[{"data":1,"prerenderedAt":1344},["ShallowReactive",2],{"blog-articles-for-slug-how-europe-tests-ai-in-its-own-languages":3},{"articles":4,"publishedKeys":1343},[5,25,40,58,72,89,107,122,138,151,169,186,201,218,235,252,270,287,306,320,336,355,369,382,400,416,431,444,456,471,484,500,515,527,850,863,879,894,911,929,945,960,977,992,1007,1020,1032,1046,1059,1075,1093,1107,1121,1134,1147,1162,1177,1192,1207,1221,1236,1250,1267,1284,1301,1315,1330],{"id":6,"title":7,"author":8,"category":9,"date":10,"description":11,"extension":12,"featured":13,"image":14,"meta":15,"metaDescription":16,"metaTitle":17,"navigation":18,"path":19,"seo":20,"stem":21,"tags":22,"__hash__":23,"__readTime":24},"blog\u002Fblog\u002F15-advanced-features-that-turn-your-chatbots-into-gold.md","15 advanced features that turn your chatbots into gold","Len Debets","Product Features","2020-08-20","A chatbot that answers a few FAQs is easy to build. Here are 15 advanced features that move your chatbot from a simple Q&A tool to something your business depends on.","md",false,"\u002Fblog\u002F3327590-300x200.jpg",{},"A chatbot that answers a few FAQs is easy to build. These 15 advanced features turn it into something your business actually depends on.","15 Advanced Chatbot Features Worth Building",true,"\u002Fblog\u002F15-advanced-features-that-turn-your-chatbots-into-gold",{"title":7,"description":11},"blog\u002F15-advanced-features-that-turn-your-chatbots-into-gold",[],"2Ub1RORB7pfLmgONRupxdg1IOzCXXrqje5046hBrjbA","3 min read",{"id":26,"title":27,"author":8,"category":28,"date":29,"description":30,"extension":12,"featured":13,"image":31,"meta":32,"metaDescription":33,"metaTitle":34,"navigation":18,"path":35,"seo":36,"stem":37,"tags":34,"__hash__":38,"__readTime":39},"blog\u002Fblog\u002F9-things-i-hate-about-ai.md","9 Things I Really Hate About AI","AI Technology","2025-05-12","I think it's great that technology is so embedded in our daily lives. It helps us get knowledge faster, finish tasks more efficiently, gives us inspiration, and occasionally scares the hell out of us with those crazy (fake) videos. I help a lot of companies implement AI, so in the end it pays my bills. But after spending a ridiculous amount of time with all these new technologies, it's time to write down the things I really hate about AI.","\u002Fblog\u002F1747039600744.png",{},"I help companies implement AI and it pays my bills. After a ridiculous amount of time with these tools, here are the 9 things I really hate about AI.",null,"\u002Fblog\u002F9-things-i-hate-about-ai",{"title":27,"description":30},"blog\u002F9-things-i-hate-about-ai","x3Pfa3upWCZeJg3PuHjgqCGhd8w1PdeVtVLlYXYmDZk","7 min read",{"id":41,"title":42,"author":8,"category":28,"date":43,"description":44,"extension":12,"featured":13,"image":45,"meta":46,"metaDescription":47,"metaTitle":34,"navigation":18,"path":48,"seo":49,"stem":50,"tags":51,"__hash__":57,"__readTime":39},"blog\u002Fblog\u002Fa-conversation-is-not-a-case.md","A Conversation Is Not a Case","2026-10-06","The 95 percent pilot failure stat is making the rounds again, this time with a diagnosis I actually agree with: chatbots automate sessions, but a bank runs cases. Why the unit of work decides whether your AI survives production, and the day eight test that shows which one you built.","\u002Fblog\u002Fblits-session-vs-case-hero.png",{},"Chatbots automate sessions, but a bank runs cases. Why the unit of work decides whether your AI survives production, and the day eight test.","\u002Fblog\u002Fa-conversation-is-not-a-case",{"title":42,"description":44},"blog\u002Fa-conversation-is-not-a-case",[52,53,54,55,56],"agentic ai","enterprise ai","ai strategy","case management","banking","IJO6dpCGt7jjLgV0FvRqM03LiuYvz779pb3P824edrk",{"id":59,"title":60,"author":8,"category":61,"date":62,"description":63,"extension":12,"featured":13,"image":64,"meta":65,"metaDescription":66,"metaTitle":34,"navigation":18,"path":67,"seo":68,"stem":69,"tags":34,"__hash__":70,"__readTime":71},"blog\u002Fblog\u002Fa-day-in-the-life-of-ai-scale-up.md","A day in the life of an AI scale-up","Company News","2025-02-24","AI moves at breakneck speed, and being the CTO of a fast growing scaleup in it is a great job. It is also not for the faint of heart. We work on large language models (LLMs), chatbots, digital humans and voicebots, and most of our customers are large banks, international companies and professional enterprises.","\u002Fblog\u002F1740065406903.png",{},"What it is like to be CTO of a fast growing AI scaleup that builds LLMs, chatbots, digital humans and voicebots for banks and large enterprises.","\u002Fblog\u002Fa-day-in-the-life-of-ai-scale-up",{"title":60,"description":63},"blog\u002Fa-day-in-the-life-of-ai-scale-up","TiXNFQdXhajzvfAWWOBvKGpeG5WGq1A5v2MXtyvF-w0","5 min read",{"id":73,"title":74,"author":8,"category":28,"date":75,"description":76,"extension":12,"featured":13,"image":77,"meta":78,"metaDescription":34,"metaTitle":79,"navigation":18,"path":80,"seo":81,"stem":82,"tags":83,"__hash__":87,"__readTime":88},"blog\u002Fblog\u002Fagentic-ai-dialects-and-the-voice-quality-gap.md","Agentic AI Languages and Dialects: Why Voice Quality Is Still the Hard Part","2026-04-10","Agentic systems promise autonomous workflows, but speech and dialect quality fail first. Why Arabic, Turkish, Chinese and more run into the same gap.","\u002Fblog\u002Fdialect.jpg",{},"Agentic AI Dialects and the Voice Quality Gap","\u002Fblog\u002Fagentic-ai-dialects-and-the-voice-quality-gap",{"title":74,"description":76},"blog\u002Fagentic-ai-dialects-and-the-voice-quality-gap",[52,84,85,86,53],"speech ai","dialects","multilingual ai","e0ZVAmJonwtrh9SFlhMirDWXgPrHQL-HItCKsC7cOFA","14 min read",{"id":90,"title":91,"author":8,"category":28,"date":92,"description":93,"extension":12,"featured":13,"image":94,"meta":95,"metaDescription":96,"metaTitle":97,"navigation":18,"path":98,"seo":99,"stem":100,"tags":101,"__hash__":105,"__readTime":106},"blog\u002Fblog\u002Fagentic-ai-on-premises-what-to-ask-before-you-choose.md","Agentic AI On Premises: What to Ask Before You Choose a Platform","2026-12-08","Many vendors say their agentic AI platform runs on premises. Fewer can show a private AI system that runs entirely inside your own data center, models included. What has to be inside, which models you can host yourself, what it costs, and the six questions I would ask every vendor.","\u002Fblog\u002Fblits-agentic-ai-on-premises-hero.png",{},"Looking for an agentic AI platform you can run on premises? What has to run inside, which models you can host yourself, what it costs and six vendor questions.","Agentic AI Platform On Premises: What to Ask First","\u002Fblog\u002Fagentic-ai-on-premises-what-to-ask-before-you-choose",{"title":91,"description":93},"blog\u002Fagentic-ai-on-premises-what-to-ask-before-you-choose",[102,52,103,104,56],"on premises ai","data residency","ai deployment","vU7JiBJPEX9OkswERPagpyR5w-awZvCX1nyp3HKuo3w","8 min read",{"id":108,"title":109,"author":8,"category":9,"date":110,"description":111,"extension":12,"featured":13,"image":112,"meta":113,"metaDescription":34,"metaTitle":34,"navigation":18,"path":114,"seo":115,"stem":118,"tags":119,"__hash__":120,"__readTime":121},"blog\u002Fblog\u002Fagentic-ai-studio-for-enterprises.md","Introducing the Agentic AI Studio for Enterprises","2026-02-17","Introducing Agentic AI Studio: a platform agnostic, enterprise ready way to design and run autonomous workflows with your own models and data.","\u002Fimages\u002Fblits-agentic-studio.png",{},"\u002Fblog\u002Fagentic-ai-studio-for-enterprises",{"title":116,"description":117},"Introducing the Agentic AI Studio for Enterprise","Self running AI agents are changing how enterprise processes work. We're releasing the Agentic AI Studio for Blits.ai, a platform agnostic, enterprise ready way to design and run autonomous workflows with your own models and data.","blog\u002Fagentic-ai-studio-for-enterprises",[],"Iko_ojcsirkF69bOvZZlrkHTDh-kRPk1Jr-wTgc8-no","9 min read",{"id":123,"title":124,"author":8,"category":28,"date":125,"description":126,"extension":12,"featured":13,"image":127,"meta":128,"metaDescription":129,"metaTitle":130,"navigation":18,"path":131,"seo":132,"stem":133,"tags":134,"__hash__":137,"__readTime":71},"blog\u002Fblog\u002Fagentic-ai-vs-generative-ai.md","Agentic AI vs Generative AI: What Changes When AI Starts Doing the Work","2026-10-23","Generative AI writes. Agentic AI does. That sounds like a small step, but it changes what can go wrong, what you need to govern and how you measure success. The real difference between agentic AI and generative AI, why most agentic projects stall, and what the ones in production have in common.","\u002Fblog\u002Fblits-agentic-vs-generative-hero.png",{},"Agentic AI vs generative AI explained for enterprises: the real difference, why most agentic projects stall, what BNY runs in production and what to govern.","Agentic AI vs Generative AI: What Really Changes","\u002Fblog\u002Fagentic-ai-vs-generative-ai",{"title":124,"description":126},"blog\u002Fagentic-ai-vs-generative-ai",[52,135,53,54,136],"generative ai","ai governance","0m-VgMpeulu0O6n741Xc_JzXzqKjxZfQNWYe0R9Qjow",{"id":139,"title":140,"author":8,"category":28,"date":141,"description":142,"extension":12,"featured":13,"image":143,"meta":144,"metaDescription":34,"metaTitle":145,"navigation":18,"path":146,"seo":147,"stem":148,"tags":149,"__hash__":150,"__readTime":71},"blog\u002Fblog\u002Fagentic-pay-and-the-moment-ai-was-allowed-to-spend-money-by-itself.md","Agentic Pay and the Moment AI Was Allowed to Spend Money","2026-01-11","Why giving AI agents controlled access to payments changes a lot, and why the open question is who designs the rules they spend under.","\u002Fimages\u002Fblits-agentic-pay.jpg",{},"Agentic Pay: When AI Was Allowed to Spend Money","\u002Fblog\u002Fagentic-pay-and-the-moment-ai-was-allowed-to-spend-money-by-itself",{"title":140,"description":142},"blog\u002Fagentic-pay-and-the-moment-ai-was-allowed-to-spend-money-by-itself",[],"njF5BBkPJMdXpGxOH8XGMC0IglTZeDkrA99y3UB_x3c",{"id":152,"title":153,"author":154,"category":28,"date":155,"description":156,"extension":12,"featured":13,"image":157,"meta":158,"metaDescription":159,"metaTitle":160,"navigation":18,"path":161,"seo":162,"stem":163,"tags":164,"__hash__":168,"__readTime":39},"blog\u002Fblog\u002Fagentic-workflows-explained.md","Agentic Workflows Explained, With Real Examples From Banking and Insurance","Coen Doolaard","2026-10-30","An agentic workflow lets AI work through a task in steps: it plans, uses your systems, checks its own work and hands the important decision to a person. A practical explanation for banks and insurers, with the design patterns, real production examples and a checklist before you build.","\u002Fblog\u002Fblits-agentic-workflows-hero.png",{},"What agentic workflows are, how they differ from AI agents, the four design patterns behind them, and what banks and insurers already run in production.","Agentic Workflows Explained: Examples From Banking and Insurance","\u002Fblog\u002Fagentic-workflows-explained",{"title":153,"description":156},"blog\u002Fagentic-workflows-explained",[165,52,56,166,167],"agentic workflows","insurance","ai architecture","W0-tqWrSk_0dyRYFIwv9EsNTXob0h-c4aARwuDABn78",{"id":170,"title":171,"author":8,"category":9,"date":172,"description":173,"extension":12,"featured":13,"image":174,"meta":175,"metaDescription":176,"metaTitle":177,"navigation":18,"path":178,"seo":179,"stem":180,"tags":181,"__hash__":185,"__readTime":121},"blog\u002Fblog\u002Fagents-that-ask-remember-and-show-their-work.md","Agents That Ask, Remember and Show Their Work: What's New in Blits.ai","2026-10-22","Release 5.9 is about one thing: giving AI agents real work without losing control of it. Agents now ask for approval before risky actions, record every decision, publish only after their tests pass, remember returning customers within strict limits and work with other companies' agents over the open A2A protocol. What each feature does, and how it works under the hood.","\u002Fblog\u002Fblits-v59-features-hero.png",{},"Blits.ai release 5.9: approval per tool, a decision log per conversation, tested publishing, memory per customer, the A2A protocol and provider control.","What's New in Blits.ai: Agents That Ask and Show Their Work","\u002Fblog\u002Fagents-that-ask-remember-and-show-their-work",{"title":171,"description":173},"blog\u002Fagents-that-ask-remember-and-show-their-work",[182,52,136,183,184],"product update","human in the loop","a2a","KG_IqJdr5mhRh9ZUxFQ9BrDqSPlcQipq0nQm0zBAGE8",{"id":187,"title":188,"author":8,"category":28,"date":189,"description":190,"extension":12,"featured":13,"image":191,"meta":192,"metaDescription":193,"metaTitle":188,"navigation":18,"path":194,"seo":195,"stem":196,"tags":197,"__hash__":200,"__readTime":71},"blog\u002Fblog\u002Fai-agency-or-ai-platform.md","AI Agency or AI Platform? How Enterprises Should Choose","2026-11-10","Hire an AI agency, buy a SaaS tool, build it yourself, or work with a platform? Every enterprise faces that choice, and the data is starting to show which path reaches production. Why custom built AI so often gets abandoned, and the questions that tell you whether you're buying a project or a capability.","\u002Fblog\u002Fblits-ai-agency-vs-platform-hero.png",{},"Should you hire an AI agency, buy software or build in house? What the data says about build vs buy, why custom AI gets abandoned, and the questions to ask.","\u002Fblog\u002Fai-agency-or-ai-platform",{"title":188,"description":190},"blog\u002Fai-agency-or-ai-platform",[198,53,199,54,52],"ai agency","build vs buy","3kzZhIokvhURrrQMls8Wry1YOEDElpvbqqdDBOP_BpM",{"id":202,"title":203,"author":154,"category":28,"date":204,"description":205,"extension":12,"featured":13,"image":206,"meta":207,"metaDescription":208,"metaTitle":209,"navigation":18,"path":210,"seo":211,"stem":212,"tags":213,"__hash__":217,"__readTime":106},"blog\u002Fblog\u002Fai-agent-governance-who-authorizes-what.md","AI Agent Governance: Who Authorizes an Agent to Do What?","2026-12-11","Most organizations now have AI agents that act in their systems, and few can say which agent is allowed to do what, or who answers for it. A practical governance framework for AI agents: four questions for every action, an owner for every agent, and what regulators already expect.","\u002Fblog\u002Fblits-ai-agent-governance-hero.png",{},"AI agent governance in plain terms: identity, permissions, approval and records for every agent action, who owns an agent, and what the EU AI Act expects.","AI Agent Governance: Who Authorizes What?","\u002Fblog\u002Fai-agent-governance-who-authorizes-what",{"title":203,"description":205},"blog\u002Fai-agent-governance-who-authorizes-what",[136,214,215,216,56],"ai agents","agent security","eu ai act","ZYHt-2M1pC4_ScgYNY_Ftdr9umZ4WxJfsaNll1XEIPc",{"id":219,"title":220,"author":154,"category":28,"date":221,"description":222,"extension":12,"featured":13,"image":223,"meta":224,"metaDescription":225,"metaTitle":220,"navigation":18,"path":226,"seo":227,"stem":228,"tags":229,"__hash__":234,"__readTime":106},"blog\u002Fblog\u002Fai-call-center-voice-agents-answer-first.md","AI Call Center: What Changes When Voice Agents Answer First","2026-11-17","An AI call center is not a call center without people. It is one where a voice agent answers first, AI supports the people who handle the harder calls, and every conversation feeds quality. What changes on the floor, what the public record shows, and where the first projects went wrong.","\u002Fblog\u002Fblits-ai-call-center-hero.png",{},"What an AI call center looks like in 2026: voice agents on the first line, AI assist for human agents, every call reviewed, and the mistakes to avoid.","\u002Fblog\u002Fai-call-center-voice-agents-answer-first",{"title":220,"description":222},"blog\u002Fai-call-center-voice-agents-answer-first",[230,231,232,233,56],"ai call center","voice agents","contact center","customer experience","2DBGhBP-vQqxS7vLMSEe6pm0nww6YYyGqSblfKRgCNI",{"id":236,"title":237,"author":154,"category":28,"date":238,"description":239,"extension":12,"featured":13,"image":240,"meta":241,"metaDescription":242,"metaTitle":243,"navigation":18,"path":244,"seo":245,"stem":246,"tags":247,"__hash__":250,"__readTime":251},"blog\u002Fblog\u002Fai-customer-service-what-actually-works.md","AI Customer Service in 2026: What Actually Works, and What the Benchmarks Hide","2026-10-27","AI customer service works, but not in the way most dashboards suggest. Public deployments show a median containment rate of two thirds, yet contacts drop by only a fifth. What separates the projects that deliver from the ones that quietly get reversed, and what to ask before you buy.","\u002Fblog\u002Fblits-ai-customer-service-hero.png",{},"What AI customer service really delivers in 2026: public benchmarks, the reversals at Klarna and CBA, and what to ask before you buy AI customer service software.","AI Customer Service in 2026: What Actually Works","\u002Fblog\u002Fai-customer-service-what-actually-works",{"title":237,"description":239},"blog\u002Fai-customer-service-what-actually-works",[248,233,232,249,56],"ai customer service","conversational ai","MTic7KjEg4zMr-2JCJeEX9masmR3GM0ZsGw8b4NhwsE","6 min read",{"id":253,"title":254,"author":154,"category":28,"date":255,"description":256,"extension":12,"featured":13,"image":257,"meta":258,"metaDescription":259,"metaTitle":260,"navigation":18,"path":261,"seo":262,"stem":263,"tags":264,"__hash__":269,"__readTime":71},"blog\u002Fblog\u002Fai-in-insurance-use-cases-in-production.md","AI in Insurance: The Use Cases Already Running in Production","2026-11-03","AI in insurance is no longer an experiment. Of 117 public insurance deployments we track, 103 run in production or at scale. Where insurers use AI today, what the leaders report, what regulators expect, and why most of the industry is still stuck at the pilot stage.","\u002Fblog\u002Fblits-ai-insurance-hero.png",{},"Where insurers already run AI in production: claims, fraud, underwriting and customer service, with results from Allianz, Zurich, Ping An, Lemonade and Aviva.","AI in Insurance: Use Cases Already in Production","\u002Fblog\u002Fai-in-insurance-use-cases-in-production",{"title":254,"description":256},"blog\u002Fai-in-insurance-use-cases-in-production",[265,266,267,52,268],"ai insurance","claims automation","underwriting","customer service","7l-cJ6cSc8Ox8EmQ6T0Q2DdDPy6FW5cLHrD5gx4zvq0",{"id":271,"title":272,"author":8,"category":28,"date":273,"description":274,"extension":12,"featured":13,"image":275,"meta":276,"metaDescription":34,"metaTitle":277,"navigation":18,"path":278,"seo":279,"stem":280,"tags":281,"__hash__":285,"__readTime":286},"blog\u002Fblog\u002Fai-spending-delegation-policies-without-losing-control.md","You Already Know How to Delegate Spending. Now Write the Mandate for an Agent.","2026-03-31","You already delegate spending with cards, POs and dual sign off. An AI agent is a new delegate. How to write its mandate, and where the analogy breaks.","\u002Fimages\u002Fblog-ai-policies.png",{},"Writing the Spending Mandate for an AI Agent","\u002Fblog\u002Fai-spending-delegation-policies-without-losing-control",{"title":272,"description":274},"blog\u002Fai-spending-delegation-policies-without-losing-control",[282,136,53,283,284],"agentic payments","delegation","payments","3UinRxWgwEV2xzGlI2-7RNVqr-WksZKCQs46SjskL6E","11 min read",{"id":288,"title":289,"author":290,"category":28,"date":291,"description":292,"extension":12,"featured":13,"image":293,"meta":294,"metaDescription":295,"metaTitle":296,"navigation":18,"path":297,"seo":298,"stem":299,"tags":300,"__hash__":305,"__readTime":251},"blog\u002Fblog\u002Fai-voice-agents-for-enterprises.md","AI Voice Agents for Enterprises: Latency, Dialects and Compliance","Yuri Ihnatov","2026-11-06","An AI voice agent can answer every call at once, in many languages, at any hour. Whether customers accept it comes down to three things that rarely show up in a demo: how fast it answers, whether it understands the way they actually speak, and whether it follows the rules for disclosure, consent and authentication.","\u002Fblog\u002Fblits-ai-voice-agents-hero.png",{},"What enterprises need from an AI voice agent: response times people accept, speech recognition that works for every dialect, and the rules on disclosure and fraud.","AI Voice Agents for Enterprises: Latency, Dialects, Compliance","\u002Fblog\u002Fai-voice-agents-for-enterprises",{"title":289,"description":292},"blog\u002Fai-voice-agents-for-enterprises",[301,302,232,303,304],"ai voice agents","voice ai","speech recognition","compliance","d_UKYWERVh6gaVcc2b1uSw1fBPQO39q6odNaVlq-jC8",{"id":307,"title":308,"author":309,"category":310,"date":311,"description":312,"extension":12,"featured":13,"image":313,"meta":314,"metaDescription":34,"metaTitle":34,"navigation":18,"path":315,"seo":316,"stem":317,"tags":34,"__hash__":318,"__readTime":319},"blog\u002Fblog\u002Fbenchmarking-conversational-ai-whitepaper.md","Benchmarking Conversational AI: Whitepaper","Paul Coerkamp","Research","2022-08-25","Our research into conversational AI performance shows clear gains in chatbot effectiveness across several industries and use cases.","\u002Fblog\u002FCover-Benchmarking-Conversational-AI-Whitepaper.jpg",{},"\u002Fblog\u002Fbenchmarking-conversational-ai-whitepaper",{"title":308,"description":312},"blog\u002Fbenchmarking-conversational-ai-whitepaper","nfDZr9Jn-Owi4BXRnkyz9owwZBZPJNEZp01hok5_J-Y","2 min read",{"id":321,"title":322,"author":309,"category":28,"date":323,"description":324,"extension":12,"featured":13,"image":325,"meta":326,"metaDescription":34,"metaTitle":327,"navigation":18,"path":328,"seo":329,"stem":330,"tags":331,"__hash__":335,"__readTime":286},"blog\u002Fblog\u002Fbenchmarking-tts-for-customers-beyond-mos-and-demos.md","How to Score a TTS Bake-Off When the Demo Is Lying to You","2026-04-08","MOS has saturated and demos are hand tuned. The lab protocol we use to grade TTS on latency, jitter, entity accuracy, barge in and cloning risk.","\u002Fimages\u002Fblog-benchmark.png",{},"Benchmarking TTS Beyond MOS and Demos","\u002Fblog\u002Fbenchmarking-tts-for-customers-beyond-mos-and-demos",{"title":322,"description":324},"blog\u002Fbenchmarking-tts-for-customers-beyond-mos-and-demos",[332,302,333,334,231],"tts","benchmarking","ai evaluation","DqEdE-KfCVokeZ_dDpJUjEgDp0i4PzcbAYQtN1PjJJA",{"id":337,"title":338,"author":8,"category":339,"date":340,"description":341,"extension":12,"featured":13,"image":342,"meta":343,"metaDescription":344,"metaTitle":345,"navigation":18,"path":346,"seo":347,"stem":348,"tags":349,"__hash__":354,"__readTime":251},"blog\u002Fblog\u002Fblits-ai-bug-hunt-game-2026-results.md","What We Learned from Turning Bug Hunting into a Game at Blits.ai","Product & Culture","2026-01-08","In the first week of January we turned quality assurance into a Bug Hunt Game. This is what happened when we put Hunters, Fixers and a Judge on a shared leaderboard powered by Azure DevOps.","\u002Fimages\u002Fblits-bounty-bug-game.png",{},"In January we turned quality assurance into a Bug Hunt Game, with Hunters, Fixers and a Judge on one Azure DevOps leaderboard. Here is what we learned.","Bug Hunting as a Game: What We Learned","\u002Fblog\u002Fblits-ai-bug-hunt-game-2026-results",{"title":338,"description":341},"blog\u002Fblits-ai-bug-hunt-game-2026-results",[350,351,352,353],"quality","devops","culture","experimentation","mWTOkdFZW8DPX4ik3aCDl_en1vapZ9YS-nCMmu1FFQU",{"id":356,"title":357,"author":309,"category":61,"date":358,"description":359,"extension":12,"featured":13,"image":360,"meta":361,"metaDescription":34,"metaTitle":362,"navigation":13,"path":363,"seo":364,"stem":365,"tags":366,"__hash__":367,"__readTime":368},"blog\u002Fblog\u002Fblits-ai-joins-llms4eu-project.md","Blits.ai joins the EU-funded LLMs4EU project","2026-01-05","Blits.ai joins LLMs4EU. Who builds large language models, on what data, and under which European values and regulations, and why that matters now.","\u002Fblog\u002Fllms4eu-blog-image.png",{},"LLMs4EU: Blits.ai Joins the European LLM Effort","\u002Fblog\u002Fblits-ai-joins-llms4eu-project",{"title":357,"description":359},"blog\u002Fblits-ai-joins-llms4eu-project",[],"M97zWsJIDe49eUcICylhQBw6aT5EKnTNuNWbZw178Pc","4 min read",{"id":370,"title":371,"author":8,"category":28,"date":372,"description":373,"extension":12,"featured":13,"image":374,"meta":375,"metaDescription":376,"metaTitle":377,"navigation":18,"path":378,"seo":379,"stem":380,"tags":34,"__hash__":381,"__readTime":71},"blog\u002Fblog\u002Fbusiness-predictions-chatgpt.md","Business predictions based on my years of experience with ChatGPT and its predecessors","2023-02-20","Over 5 years ago I started a chatbot (ad)venture called Blits.ai. My reason was simple: I believed the way we humans deal with data would turn into a conversation. Now that ChatGPT (generative AI) is all over the news and every LinkedIn feed, more and more people are reaching the same conclusion. Great! We're clearly not there yet, though.","\u002Fblog\u002F1675857460197.jpeg",{},"Five years after starting Blits.ai, my business predictions for ChatGPT and generative AI: where conversations with data are heading, and what is missing.","ChatGPT and Business: My Predictions","\u002Fblog\u002Fbusiness-predictions-chatgpt",{"title":371,"description":373},"blog\u002Fbusiness-predictions-chatgpt","26eZfqYo1gr5l1hXYbn1uJPqnvLUXu4T4vCAb5Gy39c",{"id":383,"title":384,"author":154,"category":28,"date":385,"description":386,"extension":12,"featured":13,"image":387,"meta":388,"metaDescription":389,"metaTitle":390,"navigation":18,"path":391,"seo":392,"stem":393,"tags":394,"__hash__":399,"__readTime":39},"blog\u002Fblog\u002Fcontact-center-automation-beyond-the-voice-bot.md","Contact Center Automation: Six Places Where AI Does the Work","2026-12-15","Most contact center automation projects start and stop at the bot that answers. The larger part of the work sits around the call: routing, helping the human agent, the summary afterwards, quality review and the follow up in the back office. Six places where AI does the work, what organizations report, and the rules that apply.","\u002Fblog\u002Fblits-contact-center-automation-hero.png",{},"Contact center automation is more than an AI phone system. Six places where AI does the work, from routing to back office follow up, with public results.","Contact Center Automation: Six Places for AI","\u002Fblog\u002Fcontact-center-automation-beyond-the-voice-bot",{"title":384,"description":386},"blog\u002Fcontact-center-automation-beyond-the-voice-bot",[395,396,397,268,398],"contact center automation","ai phone system","agent assist","telecom","2gNid7yjO7cAtcgLfb7Ajp5X291iNggc4RW4J2hW2vE",{"id":401,"title":402,"author":290,"category":28,"date":403,"description":404,"extension":12,"featured":13,"image":405,"meta":406,"metaDescription":407,"metaTitle":408,"navigation":18,"path":409,"seo":410,"stem":411,"tags":412,"__hash__":415,"__readTime":286},"blog\u002Fblog\u002Fconversational-ai-design-why-bad-bots-are-a-design-problem.md","Conversational AI Design: Why Bad Bots Are a Design Problem, Not a Model Problem","2026-10-21","Most chatbots that frustrate people run on perfectly capable models. What fails is the design: unclear goals, no boundaries, fallbacks that go nowhere and handoffs that lose the context. A practical guide in questions and answers, from the team that builds and tests these systems every day.","\u002Fblog\u002Fblits-conversation-design-hero.png",{},"A practical guide to conversational AI design: user goals, decision boundaries, fallback and human handoff, opening messages, tone, voice and measuring quality.","Conversational AI Design: A Practical Guide","\u002Fblog\u002Fconversational-ai-design-why-bad-bots-are-a-design-problem",{"title":402,"description":404},"blog\u002Fconversational-ai-design-why-bad-bots-are-a-design-problem",[249,413,414,302,233],"conversation design","chatbot ux","SYan43Yp4zcenD8JNZGc8XiID_yYDPGC44wjX9n9iGg",{"id":417,"title":418,"author":8,"category":28,"date":419,"description":420,"extension":12,"featured":13,"image":421,"meta":422,"metaDescription":423,"metaTitle":418,"navigation":18,"path":424,"seo":425,"stem":426,"tags":427,"__hash__":430,"__readTime":251},"blog\u002Fblog\u002Fdo-you-need-forward-deployed-engineers-to-run-ai-agents.md","Do You Need Forward Deployed Engineers to Run AI Agents?","2026-12-01","Forward deployed engineers are the delivery model of the moment in enterprise AI, and serious money is behind it. They solve a real problem. But what decides whether your AI agents survive isn't who builds the first one. It's what your own team can do on the day those engineers leave.","\u002Fblog\u002Fblits-forward-deployed-engineers-hero.png",{},"Forward deployed engineers are enterprise AI's favorite delivery model. When you need them, what you should own when they leave, and how to test a handover.","\u002Fblog\u002Fdo-you-need-forward-deployed-engineers-to-run-ai-agents",{"title":418,"description":420},"blog\u002Fdo-you-need-forward-deployed-engineers-to-run-ai-agents",[428,53,214,429,199],"forward deployed engineers","ai implementation","TVovOJbWvP1wW73yMN4Bk-LMgV7bHlngif-CtsEhOgA",{"id":432,"title":433,"author":309,"category":28,"date":434,"description":435,"extension":12,"featured":13,"image":436,"meta":437,"metaDescription":34,"metaTitle":438,"navigation":18,"path":439,"seo":440,"stem":441,"tags":442,"__hash__":443,"__readTime":368},"blog\u002Fblog\u002Feu-ai-act-2026-enterprise-readiness-checklist.md","EU AI Act 2026: The Enterprise AI Readiness Checklist Before August","2026-03-11","Most teams treat EU AI Act compliance as a legal side project. A checklist of what enterprises should implement now to avoid disruption and rework.","\u002Fimages\u002Fblog-eu-ai-act.png",{},"EU AI Act 2026: Enterprise Readiness Checklist","\u002Fblog\u002Feu-ai-act-2026-enterprise-readiness-checklist",{"title":433,"description":435},"blog\u002Feu-ai-act-2026-enterprise-readiness-checklist",[216,53,136,304],"rluJvoZAzD04WieClVwCsoj6r18ndtcHInhwIzsy0v4",{"id":445,"title":446,"author":8,"category":28,"date":447,"description":448,"extension":12,"featured":13,"image":449,"meta":450,"metaDescription":34,"metaTitle":451,"navigation":18,"path":452,"seo":453,"stem":454,"tags":34,"__hash__":455,"__readTime":106},"blog\u002Fblog\u002Ffeeding-llms-without-leaking-secrets.md","Feeding LLMs Without Leaking Secrets: A Guide for Companies On How To Add Your Company Data","2025-05-19","A plain guide for business teams on how to add your company data to an LLM safely, without leaking secrets, and the concepts behind it.","\u002Fblog\u002F1747588091563.png",{},"Feeding LLMs Without Leaking Company Secrets","\u002Fblog\u002Ffeeding-llms-without-leaking-secrets",{"title":446,"description":448},"blog\u002Ffeeding-llms-without-leaking-secrets","8IxE-B9QPvTgygMq-My29zKji23d5TeHn16xmfgIgYY",{"id":457,"title":458,"author":290,"category":459,"date":460,"description":461,"extension":12,"featured":13,"image":462,"meta":463,"metaDescription":34,"metaTitle":464,"navigation":18,"path":465,"seo":466,"stem":467,"tags":468,"__hash__":470,"__readTime":286},"blog\u002Fblog\u002Ffrom-johannesburg-five-signals-agentic-ai-enterprise-banking.md","From Johannesburg: Five Signals on Where Agentic AI Meets Enterprise Banking","Events","2026-07-15","Five takeaways from Mastercard's summit in Johannesburg: the technology is no longer the bottleneck, governance, trust and context are.","\u002Fblog\u002Ffive-signals-mastercard.png",{},"Five Signals on Agentic AI in Enterprise Banking","\u002Fblog\u002Ffrom-johannesburg-five-signals-agentic-ai-enterprise-banking",{"title":458,"description":461},"blog\u002Ffrom-johannesburg-five-signals-agentic-ai-enterprise-banking",[52,469,249,282,334],"enterprise banking","b_mlFIOndF2DPQRqZDiXZNvaVGRyPPI3P5jsTOTRhQY",{"id":472,"title":473,"author":154,"category":459,"date":474,"description":475,"extension":12,"featured":18,"image":476,"meta":477,"metaDescription":34,"metaTitle":478,"navigation":18,"path":479,"seo":480,"stem":481,"tags":482,"__hash__":483,"__readTime":251},"blog\u002Fblog\u002Ffrom-paris-to-rio-signals-and-where-enterprise-ai-is-heading.md","From Paris to Rio: Production Over Pilots, Substance Over Spectacle","2026-07-10","Signals from VivaTech and Web Summit Rio: production over pilots, substance over spectacle, and why your operating framework beats any one model.","\u002Fblog\u002Ffrom-paris-to-rio.png",{},"From Paris to Rio: Where Enterprise AI Is Heading","\u002Fblog\u002Ffrom-paris-to-rio-signals-and-where-enterprise-ai-is-heading",{"title":473,"description":475},"blog\u002Ffrom-paris-to-rio-signals-and-where-enterprise-ai-is-heading",[53,52,249,54,136],"6DzNE9FCbGcnNxwm9Du9-SwxJ5yh585XO0lPY3tLxng",{"id":485,"title":486,"author":309,"category":28,"date":487,"description":488,"extension":12,"featured":13,"image":489,"meta":490,"metaDescription":34,"metaTitle":491,"navigation":18,"path":492,"seo":493,"stem":494,"tags":495,"__hash__":498,"__readTime":499},"blog\u002Fblog\u002Ffrom-single-agent-to-multi-agent-systems-when-to-split-roles.md","Most Multi Agent Systems Are a Solution Looking for a Problem","2026-03-19","Start with one agent, and reach for multi agent only for parallel, read heavy work. The case against splitting too early, and how to do it right.","\u002Fimages\u002Fblog-multi-agent.png",{},"When to Split Into Multi Agent Systems","\u002Fblog\u002Ffrom-single-agent-to-multi-agent-systems-when-to-split-roles",{"title":486,"description":488},"blog\u002Ffrom-single-agent-to-multi-agent-systems-when-to-split-roles",[52,496,53,167,497],"multi agent systems","llm","jh1LuV8_t23zMxcRv4qwpkRBlmGeW4eHuFvagOyENQM","10 min read",{"id":501,"title":502,"author":8,"category":9,"date":503,"description":504,"extension":12,"featured":13,"image":505,"meta":506,"metaDescription":34,"metaTitle":507,"navigation":18,"path":508,"seo":509,"stem":512,"tags":513,"__hash__":514,"__readTime":71},"blog\u002Fblog\u002Ffrom-talking-to-doing-six-months-of-blits-ai-updates.md","Blits.ai Platform Update: Five Big Additions from the Last Six Months","2026-07-20","A plain roundup of five things we shipped this year: emotional voice, interactive cards, agentic tasks, in chat payments and response feedback.","\u002Fblog\u002Fblits-login-screen.png",{},"Blits.ai Platform Update: Six Months of Additions","\u002Fblog\u002Ffrom-talking-to-doing-six-months-of-blits-ai-updates",{"title":510,"description":511},"Product Update: Five Big Additions from the Last Six Months","A plain language roundup of the five biggest things we shipped this year (emotional voice, rich interactive cards, agentic tasks, in chat payments and response feedback) and what each one does for the person on the other end of the conversation.","blog\u002Ffrom-talking-to-doing-six-months-of-blits-ai-updates",[182,52,249,302,282],"yojUU4dXMnn1LM4XJdH8rXr-2_-CI5k9oYKfDLjNOwo",{"id":516,"title":517,"author":309,"category":28,"date":518,"description":519,"extension":12,"featured":13,"image":520,"meta":521,"metaDescription":34,"metaTitle":522,"navigation":18,"path":523,"seo":524,"stem":525,"tags":34,"__hash__":526,"__readTime":319},"blog\u002Fblog\u002Fgpt-4-in-chatbots.md","GPT-4 in Chatbots: Blits.ai Breaks New Ground in Conversational AI","2023-07-14","Blits.ai now supports GPT-4, bringing new capabilities to chatbot development and deployment. What it means for conversational AI.","\u002Fblog\u002Fgpt-4.webp",{},"GPT-4 Chatbots: New Ground in Conversational AI","\u002Fblog\u002Fgpt-4-in-chatbots",{"title":517,"description":519},"blog\u002Fgpt-4-in-chatbots","G00HNe_XJ2lwAPzMes8evMiwWORzWzdgIFCzLrsptWU",{"id":528,"title":529,"author":290,"body":530,"category":310,"date":837,"description":838,"extension":12,"featured":13,"image":553,"meta":839,"metaDescription":840,"metaTitle":841,"navigation":18,"path":842,"seo":843,"stem":844,"tags":845,"__hash__":849},"blog\u002Fblog\u002Fhow-europe-tests-ai-in-its-own-languages.md","How Europe Tests AI in Its Own Languages",{"type":531,"value":532,"toc":820},"minimark",[533,544,547,554,559,562,565,568,571,575,578,586,592,595,598,607,611,614,617,621,624,627,630,633,636,639,642,646,649,652,655,658,661,664,668,671,677,680,683,686,689,693,696,701,704,707,710,713,716,720,723,726,729,732,735,738,741,744,748,751,754,757,760,763,766,769,773,776,779,782,788],[534,535,536,543],"p",{},[537,538,542],"a",{"href":539,"rel":540},"https:\u002F\u002Fwww.llms4eu.eu\u002F",[541],"noopener","LLMs4EU"," is a European Union project that adapts open language models to the official languages of the EU and to real work in five sectors. Blits.ai is one of the partners. We work on the part of the project that tests whether the finished systems are good enough to use. This article explains what the project does, how the testing works, and what we learned in the first eighteen months.",[534,545,546],{},"A visitor at a village church wants to know its history, so she asks her phone in her own language. About two million people speak that language. The answer she gets depends on a large language model, and on whether anyone has tested that model in her language. For most of Europe's languages, nobody has done that testing yet. LLMs4EU sets out to do it.",[534,548,549],{},[550,551],"img",{"alt":552,"src":553},"Holy Trinity church in Hrastovlje, a village church in Slovenia","\u002Fblog\u002Fblits-llms4eu-hrastovlje-church.jpg",[555,556,558],"h2",{"id":557},"why-smaller-languages-need-their-own-tests","Why smaller languages need their own tests",[534,560,561],{},"A large language model learns from text. Its makers give it a very large amount of text, and the model learns to predict the next word in a sentence. At that scale, the model can answer questions, summarise documents and hold a conversation.",[534,563,564],{},"Most of that text comes from the internet, and most of the internet is in English. So the models work best in English, well in French, German and Spanish, and less well in smaller languages. Irish, Latvian, Maltese and Slovene each have far less text online, so a model has far less to learn from.",[534,566,567],{},"Model makers usually publish test scores for English and a few large languages. For smaller languages, there is often no published score. So a user can get a fluent answer and still not know whether it is correct. In many cases, the company that built the model does not know either.",[534,569,570],{},"Europe builds its models under its own rules. The makers of a European model train it only on text they have the right to use. They follow EU copyright rules, which let authors and publishers opt out of AI training, and the General Data Protection Regulation. The makers also publish a summary of where the training data came from. LLMs4EU works with organisations that hold text, recordings and records in languages with little text online, and that can share them legally. For many smaller languages, that is where the useful data is.",[555,572,574],{"id":573},"what-llms4eu-does","What LLMs4EU does",[534,576,577],{},"LLMs4EU started in March 2025 and runs until early 2028. The EU's Digital Europe Programme funds part of it. The partners come from across the EU. They include universities, national research institutes, supercomputing centres, telecoms companies, an energy company, public bodies and small companies.",[534,579,580,585],{},[537,581,584],{"href":582,"rel":583},"https:\u002F\u002Fwww.alt-edic.eu\u002F",[541],"ALT-EDIC",", the Alliance for Language Technologies, coordinates the project. EU member states set up ALT-EDIC in 2024, so that Europe's languages get the data, tools and testing that the market does not provide on its own. Its offices are in the chateau at Villers-Cotterêts, north of Paris. In 1539, a French king signed an ordinance there that made French the language of law and administration.",[534,587,588],{},[550,589],{"alt":590,"src":591},"The chateau at Villers-Cotterêts, home of ALT-EDIC and the Cité internationale de la langue française","\u002Fblog\u002Fblits-llms4eu-cite-langue-francaise.jpg",[534,593,594],{},"The project takes open language models and adapts them for five areas of work: science, public services, tourism, telecommunications and energy. In each area, organisations that need a result lead the work, and partners that can build it support them. The project aims to cover all 24 official EU languages and a number of regional languages. The project plans to publish its models, data and results openly, so a small company or a public body can use them without asking permission.",[534,596,597],{},"Many public services now reach people through text and speech. When a municipality buys an assistant that works well in one language and badly in another, speakers of the second language get a worse service from the same public body. So the project builds open models on European data, adapts them to European tasks, and tests them in public.",[534,599,600,601,606],{},"LLMs4EU works alongside ",[537,602,605],{"href":603,"rel":604},"https:\u002F\u002Fopeneurollm.eu\u002F",[541],"OpenEuroLLM",", another project funded under the Digital Europe Programme. OpenEuroLLM trains new open models on European supercomputers, in all official EU languages, and publishes the weights, the code and the data. LLMs4EU adapts those models and other open models for specific tasks. It also collects domain data, checks the legal status of each dataset, and tests the results. The two projects share partners and data. So European teams train, adapt and test the same open models, and anyone can check the results.",[555,608,610],{"id":609},"why-testing-comes-first","Why testing comes first",[534,612,613],{},"A language model writes fluent sentences whether or not it knows the answer. When a model handles a language poorly, it can still write a clear answer that is wrong. So good writing does not show whether the content is correct.",[534,615,616],{},"So the project sets its test method in advance, tests against agreed standards, and publishes the results.",[555,618,620],{"id":619},"how-the-evaluation-centre-works","How the evaluation centre works",[534,622,623],{},"In LLMs4EU, this work belongs to the Distributed Evaluation Centre for LLMs, and Blits.ai is part of it. The centre is distributed because several organisations do the testing. Each one tests the use cases it knows best, and all of them follow one shared set of rules.",[534,625,626],{},"The rules come from decades of evaluation campaigns for speech recognition and machine translation. The team that builds a system does not test it. The evaluators write new test questions for each campaign. They keep the questions closed until the test ends, and then they publish them. The people who prepare the training data work apart from the people who write the test, because a system must not see the test in advance. Every system built for a use case runs on the same test material in the same period, so the evaluators can compare the results.",[534,628,629],{},"Over the past months, the evaluation group has worked out how the centre will run in practice. The group is working towards a network of specialist centres. Each centre tests the areas it knows, and ALT-EDIC both runs tests and coordinates the network. So an organisation that wants a model tested can ask the coordinator which centre fits, or contact a specialist centre directly. The group will also set criteria for adding new centres over time.",[534,631,632],{},"The group also looked at public leaderboards. When test data stays public for a long time, it can end up in the training data of the models it is meant to test. So the group prefers time limited campaigns. The evaluators share the test data only during the campaign, and they publish it when the campaign ends. Later runs on the same data still help developers, but they do not count as official results.",[534,634,635],{},"Running every submitted model on central servers takes a lot of maintenance, because software and libraries keep changing. One option is that participants run their own systems on the test inputs and send back the outputs for scoring. The group is working out how to do this while keeping the test data closed until the right moment.",[534,637,638],{},"Evaluators will use the shared platform to produce scores. They can check each result before it goes back to the developer. The platform will also hold shared resources: evaluation software, metrics, evaluation plans, guidelines and tools for data formats. A working group for the platform starts meeting in November. Several partners, Blits.ai among them, will present the evaluation tools they already use, so the working group can build on them.",[534,640,641],{},"Some new work in the project starts with the test. The team writes the data plan and the evaluation plan first, and then invites developers in the consortium to take part. So several systems can enter the same campaign from the start.",[555,643,645],{"id":644},"testing-the-whole-system","Testing the whole system",[534,647,648],{},"At the end of the project, the centre tests the systems that each use case builds. A scientist, a civil servant, a tourist, a call centre supervisor or an engineer will use these systems. So the test covers the whole system, and the language is one part of it.",[534,650,651],{},"The assistant the visitor uses at the church does several things. It searches a library of documents and picks the relevant pages. Sometimes it calls other software, for example to look up opening hours. Then it decides whether it found enough to answer. When it did not, the correct answer is that it does not know. Getting a model to give that answer takes deliberate work.",[534,653,654],{},"So the test checks several things. It checks that the system found the right page, and that the page it cites exists. It checks that any form the system filled in is correct. It checks that the system said it did not know when the library had no answer. It also checks that the system answered in the language and variety the visitor used. A system with correct grammar can still fail these checks.",[534,656,657],{},"Each use case writes an evaluation plan before testing starts. The plan states the metrics, the type and size of the test data, the test conditions and the schedule. The evaluation group reviews each plan, so it can test every use case to the same standard.",[534,659,660],{},"Blits.ai builds and runs systems of this kind for banks and large companies. Over several years, we learned what fails when a system moves from a test environment to real customers. In LLMs4EU, we jointly lead the evaluation of the tourism use case. We also put forward our own evaluation platform as one of the tools the centre can use. Our platform checks whether an answer is correct. It also checks whether the answer came back in the language and variety the user used, and whether it stayed in that variety for the whole conversation.",[534,662,663],{},"In July, the tourism team asked us to review the design of their assistant. We described how we build our commercial systems, and two of those ideas also apply to their assistant. Every document enters the library through one checkpoint, so nothing gets in without a check. When the system finds nothing relevant, it returns a fixed message that a person wrote in advance, and it does not call the model at all. We do this because a model that is asked to refuse can still choose to answer. So we write the message into the code, and the model never makes that choice.",[555,665,667],{"id":666},"what-we-learned-from-arabic","What we learned from Arabic",[534,669,670],{},"Blits.ai learned the language check in the Gulf. A few years ago, a bank there ordered a digital human from us, an on screen avatar that talks to customers. The contract said Arabic, so we delivered Modern Standard Arabic, the formal Arabic of schools and news. It was fluent and correct. Within days, the bank told us it did not sound like its customers. Its customers speak the way people in Riyadh speak, and that is different from the news.",[534,672,673],{},[550,674],{"alt":675,"src":676},"The Riyadh skyline with the Kingdom Centre","\u002Fblog\u002Fblits-llms4eu-riyadh-skyline.jpg",[534,678,679],{},"So we hired native speakers as testers. We reviewed every step, from the microphone to the spoken answer. We rebuilt our tests around one question: would a person from that city accept this voice as their own. After that work, the answer was yes.",[534,681,682],{},"We kept the testing. Our platform now checks every answer for language and dialect. It checks that the system replied in the variety the user spoke, in the register the user expected, and that it kept that variety for the whole conversation. A system can drift back to the formal language after a few turns. That drift shows up in longer conversations, so we test full conversations.",[534,684,685],{},"Today we run assistants in Egyptian Arabic, in several Gulf dialects including Saudi and Qatari Arabic, in Libyan Arabic and in Turkish. We test every candidate model on the same recordings, with the same background noise and the same time limits. We also ask a panel of native listeners to judge the voices, because an automatic score counts correct words and cannot tell whether a voice sounds natural to the customer.",[534,687,688],{},"One result from this work matters for Europe. On our own Arabic tests, a newer generation of speech models did not always do better on dialect than an older one. So we test every new model on our own data before we use it. A speaker of a smaller European language needs the same check as a customer in Riyadh.",[555,690,692],{"id":691},"lessons-from-the-first-eighteen-months","Lessons from the first eighteen months",[534,694,695],{},"These lessons come from the consortium meeting in June, from the working meetings since then, and from the evaluation plans the teams have written. They apply to any team that builds or tests language systems.",[697,698,700],"h3",{"id":699},"writing-the-test","Writing the test",[534,702,703],{},"**Native speakers write the test questions. **Machine translation can help produce training data. In a test, machine translation would measure the translation software, and that is the wrong thing to measure. So native speakers write, translate or correct every test question by hand. They also need to work independently of the team that builds the system. For smaller languages, finding such speakers takes extra effort.",[534,705,706],{},"**Partners translate test queries into their own languages. **The team that builds the agentic search for the project's data catalogue wrote a set of test queries in English. Then it asked partners to translate the queries into their own languages. The team now compares the results and the response time in each language, and uses them to choose the model.",[534,708,709],{},"**The test data includes speech that mixes languages. **In one speech use case, the broadcasts also contain English words. So the evaluation team builds test data and scoring that handle speech that switches between languages.",[534,711,712],{},"**The evaluation team keeps part of the test unseen. **The evaluation team holds back a set of questions that no developer sees before the run. It also adds hard cases on purpose. These include questions from outside the domain, questions with missing information, and questions that test whether the system can be misled.",[534,714,715],{},"**Partners make the test harder when scores get too high. **One partner found that his test set of web pages returned very high scores. So he asked the other partners for data to make the test harder, because a test that every system passes no longer shows the differences between systems.",[697,717,719],{"id":718},"running-the-test","Running the test",[534,721,722],{},"**The evaluation team checks the starting point first. **Before it tested adapted models, one evaluation team reran a public benchmark on standard open models. It wanted to see whether it got the published scores. For some model sizes the scores matched, and for others they differed. So the team is now checking settings such as the generation parameters, because it needs a correct starting point before it measures any improvement.",[534,724,725],{},"**Teams run a dry run on the data they have. **Partners often need time to share real data, because of contracts and legal checks. So the teams run the full evaluation on data they already have. Then, when the real data arrives, the teams have already tested their pipeline.",[534,727,728],{},"**The team freezes the test before the run. **Before a run starts, the team fixes the test package. Then it takes a snapshot of the documents the system searches. Every run gets the same inputs. Once the run starts, nobody edits the outputs by hand. The team runs the full system in a closed test environment, so it tests retrieval, citations and answers together.",[534,730,731],{},"**A language model scores first, and people review the hard cases. **A language model scores the answers in a first pass. Then people review the cases where the model is unsure. They also review the results that someone disputes, and the answers that someone flagged. When an answer confidently invents a fact, the answer counts as a failure.",[534,733,734],{},"**Teams use different measures at different stages. **Early in development, the team looks at overall precision and recall, to see whether a model is on track. Later, it looks at a breakdown by error type, to see what each small change does. Before a system goes into production for a sensitive task, the team uses a much stricter measure. For removing personal data from customer conversations, a partner developed a measure that heavily penalises a single missed item, because one missed name can reveal who a person is. The same measure also penalises a system that removes too much, because then the text is no longer useful. All of these measures come from the same test set, so developers can choose which one to look at.",[534,736,737],{},"**Teams report results per language. **Each team reports its scores per task and per language. So a reader can see how the system performs in each language separately.",[534,739,740],{},"**Teams measure speed as well as quality. **Several teams measure how fast a system answers, in tokens per second and in time to the first token. The tourism team also runs a stress test with thousands of synthetic queries. Then it checks the logs, to confirm that the system records resource use correctly for each user.",[534,742,743],{},"**Evaluation teams share their test tools. **One evaluation team built a simple interface for its tools. You pick a task and load a dataset, and then you see the results per metric and per sample. The team plans to share the tool with partners, so they can rerun the tests and check the results themselves.",[697,745,747],{"id":746},"building-the-system-and-its-data","Building the system and its data",[534,749,750],{},"**Small refinements can remove content without a warning. **A partner compared tools that turn web pages into clean text. The plain tool with its default settings gave the best result. Each refinement he added removed the tables from the pages, and no tool warned him. He found the problem when he read the outputs side by side.",[534,752,753],{},"**Scanning software can drop letters. **Much heritage material exists only as scanned books and old PDF files. Text recognition software often drops accents and special letters, such as č, š and ž. A model cannot restore a letter that the scanner lost, so the project fixes the scanning step itself.",[534,755,756],{},"**The training data teaches the system to say it does not know. **The tourism team adds some questions with no answer in the library to the training conversations. So the model learns that it may say it does not know.",[534,758,759],{},"**The tourism team uses synthetic data for training and human data for checking. **The team generated a large set of synthetic questions and answers for training. Then it wrote a smaller set by hand for checking. Now linguists review the answers by hand, with a language model as an assistant judge. The team is also building an automated test that follows conversation scenarios step by step.",[534,761,762],{},"**Each item in a collection has its own rights. **A large European heritage collection publishes its catalogue records under an open licence. The texts and images that those records describe carry their own rights, and the project cannot use many of them for training. So the project checks the rights item by item. The project's lawyers also noted that anonymised data is not always anonymous in law, because someone can sometimes identify a person by combining it with other data.",[534,764,765],{},"**The law can limit what evaluators may see. **In one public service, the public body publishes documents for a short period. After that period, strict legal limits apply. The system owner may keep using the documents, and independent evaluators may not. So the evaluation team agrees with the owner which parts of the system it can test on its own, and the owner tests the rest.",[534,767,768],{},"**The project labels its own datasets. **The project built a dashboard that reads the main public catalogues of language data. In many entries, the size field was empty or used different units, such as kilobytes, hours, tokens or words. The domain field was usually free text, or empty. So the project labels its own datasets with size, units and domain.",[555,770,772],{"id":771},"what-comes-next","What comes next",[534,774,775],{},"The tourism team has collected its training and test data. It is now fine tuning models together with a retrieval system, and linguists are reviewing the first answers. Other use cases are finishing their evaluation plans and preparing their data. The teams run the first evaluation rounds over the coming months. Larger campaigns follow in 2027. The project ends in early 2028.",[534,777,778],{},"For Blits.ai, this is the work we already do, applied across Europe. We build and test conversational systems in languages and dialects where the difference between supported and usable decides whether a product works. We bring the tools we built for that to the evaluation centre. In return, we help shape the rules that European public bodies will use to decide which systems to trust.",[534,780,781],{},"When the visitor at the church asks her question, the system she uses will have been tested in her language, and the project will publish the results.",[534,783,784],{},[785,786,787],"em",{},"LLMs4EU is funded in part by the European Union under the Digital Europe Programme, Grant Agreement 101198470. Blits.ai is a consortium partner. Views and opinions expressed are those of the author only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them.",[534,789,790],{},[785,791,792,793,798,799,804,805,810,811,798,816,819],{},"Photos via Wikimedia Commons: Holy Trinity church in Hrastovlje, Slovenia, by ",[537,794,797],{"href":795,"rel":796},"https:\u002F\u002Fcommons.wikimedia.org\u002Fwiki\u002FFile:Cerkev_sv._Trojice_-_Hrastovlje.jpg",[541],"Miha Peče"," (",[537,800,803],{"href":801,"rel":802},"https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby-sa\u002F4.0\u002F",[541],"CC BY-SA 4.0",", cropped); Cité internationale de la langue française by ",[537,806,809],{"href":807,"rel":808},"https:\u002F\u002Fcommons.wikimedia.org\u002Fwiki\u002FFile:Villers-Cotter%C3%AAts_-_Cit%C3%A9_internationale_de_la_langue_fran%C3%A7aise_17.jpg",[541],"Renhour48"," (CC0); Riyadh skyline by ",[537,812,815],{"href":813,"rel":814},"https:\u002F\u002Fcommons.wikimedia.org\u002Fwiki\u002FFile:Riyadh_Skyline.jpg",[541],"B.alotaby",[537,817,803],{"href":801,"rel":818},[541],").",{"title":821,"searchDepth":822,"depth":822,"links":823},"",2,[824,825,826,827,828,829,830,836],{"id":557,"depth":822,"text":558},{"id":573,"depth":822,"text":574},{"id":609,"depth":822,"text":610},{"id":619,"depth":822,"text":620},{"id":644,"depth":822,"text":645},{"id":666,"depth":822,"text":667},{"id":691,"depth":822,"text":692,"children":831},[832,834,835],{"id":699,"depth":833,"text":700},3,{"id":718,"depth":833,"text":719},{"id":746,"depth":833,"text":747},{"id":771,"depth":822,"text":772},"2026-11-16","What Blits.ai does in the LLMs4EU project, and what we learned in the first eighteen months. How Europe adapts open language models to its own languages, why testing comes first, how the distributed evaluation centre works, and the lessons from writing, running and building the tests.",{},"What Blits.ai does in the EU project LLMs4EU, how the distributed evaluation centre tests AI in every EU language, and what we learned in eighteen months.","LLMs4EU: How Europe Tests AI in Its Own Languages","\u002Fblog\u002Fhow-europe-tests-ai-in-its-own-languages",{"title":529,"description":838},"blog\u002Fhow-europe-tests-ai-in-its-own-languages",[846,847,86,334,848],"llms4eu","european ai","language technology","Le5SM4Pjlf-dqqPnqWJJHVwjDgM_tTYJNuV8Cok0-Mg",{"id":851,"title":852,"author":309,"category":28,"date":853,"description":854,"extension":12,"featured":13,"image":855,"meta":856,"metaDescription":857,"metaTitle":858,"navigation":18,"path":859,"seo":860,"stem":861,"tags":34,"__hash__":862,"__readTime":319},"blog\u002Fblog\u002Fhow-practical-is-gpt-3-for-conversational-ai-chatbots.md","How practical is GPT-3 for Conversational AI chatbots?","2022-01-25","GPT-3 has changed what people expect from conversational AI. Ask it almost anything and you get a fluent, often surprisingly good answer. Whether that makes it practical for a chatbot or voicebot that talks to your customers is a different question, and that is what this article is about.","\u002Fblog\u002FGPT-3.jpg",{},"GPT-3 gives fluent, often surprisingly good answers. Whether that makes it practical for a chatbot that talks to your customers is another question.","How Practical Is GPT-3 for Chatbots?","\u002Fblog\u002Fhow-practical-is-gpt-3-for-conversational-ai-chatbots",{"title":852,"description":854},"blog\u002Fhow-practical-is-gpt-3-for-conversational-ai-chatbots","5bcufuhpdudmHxSJDIeisrRbO9tOk1zdavIsJZcNhBY",{"id":864,"title":865,"author":8,"category":28,"date":866,"description":867,"extension":12,"featured":13,"image":868,"meta":869,"metaDescription":870,"metaTitle":871,"navigation":18,"path":872,"seo":873,"stem":874,"tags":875,"__hash__":878,"__readTime":286},"blog\u002Fblog\u002Fhow-to-build-an-ai-control-tower-for-agentic-operations.md","The AI Control Tower: Air Traffic Control for Agents in Production","2026-03-23","A dashboard tells you the plane landed. A control tower keeps two aircraft off the same runway in real time, with the authority to divert one. What that distinction means for running AI agents in production, and why an HTTP 200 will lie to you.","\u002Fimages\u002Fblog-control-tower.png",{},"A dashboard tells you the plane landed. A control tower keeps two aircraft off one runway. What that means for running AI agents in production.","How to Build an AI Control Tower for Agents","\u002Fblog\u002Fhow-to-build-an-ai-control-tower-for-agentic-operations",{"title":865,"description":867},"blog\u002Fhow-to-build-an-ai-control-tower-for-agentic-operations",[876,52,136,877,53],"ai observability","guardrails","tekVT7bbrPjtuBBrGEuTtVN495S40tyKTdMJ6WTqSmQ",{"id":880,"title":881,"author":290,"category":28,"date":882,"description":883,"extension":12,"featured":13,"image":884,"meta":885,"metaDescription":886,"metaTitle":887,"navigation":18,"path":888,"seo":889,"stem":890,"tags":891,"__hash__":893,"__readTime":106},"blog\u002Fblog\u002Finsurance-ai-claims-underwriting-fraud-on-top-of-the-core-system.md","AI for Insurance Claims, Underwriting and Fraud: What Sits on Top of the Core System","2026-12-18","Insurers don't need new claims management software or underwriting software to use AI. The deployments that work put an agent layer on top of the core system they already have. How that layer is built for claims, underwriting and fraud detection, what insurers report, and where the limits are.","\u002Fblog\u002Fblits-insurance-claims-underwriting-fraud-hero.png",{},"How AI agents work with insurance claims management software, underwriting software and fraud detection, without replacing the core system. With public results.","AI for Insurance Claims, Underwriting and Fraud","\u002Fblog\u002Finsurance-ai-claims-underwriting-fraud-on-top-of-the-core-system",{"title":881,"description":883},"blog\u002Finsurance-ai-claims-underwriting-fraud-on-top-of-the-core-system",[166,266,267,892,214],"fraud detection","08_0XCOfCZBlzrSuqPhuMlZpHYAtN7j4KvH3vzEJ0ME",{"id":895,"title":896,"author":309,"category":9,"date":897,"description":898,"extension":12,"featured":13,"image":899,"meta":900,"metaDescription":901,"metaTitle":902,"navigation":18,"path":903,"seo":904,"stem":905,"tags":906,"__hash__":910,"__readTime":39},"blog\u002Fblog\u002Fintroducing-blits-ai-financial-wellbeing.md","Introducing Blits.ai Financial Wellbeing: Guidance, Never Financial Advice","2026-08-29","Our newest product turns a bank's own transaction data into a wellbeing score, plain language insight and gentle nudges toward healthier money habits. It runs inside the bank, it measures its own impact, and it will never give financial advice. Here is how it works.","\u002Fblog\u002Fblits-fw-dashboard.png",{},"Our newest product turns a bank's transaction data into a wellbeing score, plain language insight and gentle nudges. It runs inside the bank, never as advice.","Blits.ai Financial Wellbeing: Guidance, Not Advice","\u002Fblog\u002Fintroducing-blits-ai-financial-wellbeing",{"title":896,"description":898},"blog\u002Fintroducing-blits-ai-financial-wellbeing",[907,469,877,908,909],"financial wellbeing","product","nudges","fr19gXuC3QnydGpIKcJpc9c9ZjA6F-HESxQnaqK8dfY",{"id":912,"title":913,"author":290,"category":28,"date":914,"description":915,"extension":12,"featured":13,"image":916,"meta":917,"metaDescription":918,"metaTitle":919,"navigation":18,"path":920,"seo":921,"stem":922,"tags":923,"__hash__":928,"__readTime":39},"blog\u002Fblog\u002Fknowledge-graphs-vs-vector-search.md","Knowledge Graphs vs Vector Search: What Enterprise AI Actually Needs","2026-11-24","Vector search finds passages that mean the same thing. A knowledge graph knows how things are connected. Most enterprise AI needs the first, some needs the second, and a surprising number of questions need neither. A practical guide to choosing, from someone who builds both.","\u002Fblog\u002Fblits-knowledge-graph-hero.png",{},"Knowledge graph or vector search? What each one is good at, where each breaks, and how to decide which questions your enterprise AI needs to answer.","Knowledge Graphs vs Vector Search for Enterprise AI","\u002Fblog\u002Fknowledge-graphs-vs-vector-search",{"title":913,"description":915},"blog\u002Fknowledge-graphs-vs-vector-search",[924,925,926,927,53],"knowledge graph","vector search","rag","retrieval","IPa50HdvBjLxCAI5Q_SAlhu7UaoQewgBoWLBooU0CkQ",{"id":930,"title":931,"author":309,"category":28,"date":932,"description":933,"extension":12,"featured":13,"image":934,"meta":935,"metaDescription":34,"metaTitle":936,"navigation":18,"path":937,"seo":938,"stem":939,"tags":940,"__hash__":944,"__readTime":286},"blog\u002Fblog\u002Fmcp-security-in-production-auth-and-least-privilege.md","MCP Security: The Threat Model Changes the Day After You Ship","2026-03-15","Most MCP breaches needed no exploit code; the servers behaved as specified. What goes wrong in production, and why the new OAuth spec is half a fix.","\u002Fimages\u002Fblog-mcp-security.png",{},"MCP Security in Production: Auth and Privilege","\u002Fblog\u002Fmcp-security-in-production-auth-and-least-privilege",{"title":931,"description":933},"blog\u002Fmcp-security-in-production-auth-and-least-privilege",[941,942,943,53,52],"mcp","ai security","prompt injection","Hz41mPrcsT9FC-R_MkrwnwIFerSFFo6bu9tJw9xeKTA",{"id":946,"title":947,"author":8,"category":28,"date":948,"description":949,"extension":12,"featured":13,"image":950,"meta":951,"metaDescription":34,"metaTitle":952,"navigation":18,"path":953,"seo":954,"stem":955,"tags":956,"__hash__":959,"__readTime":368},"blog\u002Fblog\u002Fmcp-servers-and-tools.md","MCP Servers and Tools: How LLMs Connect to the Real World","2026-01-27","MCP servers standardize how LLMs find and use tools. What they are, why they matter, how we use them at Blits.ai, and where this is going next.","\u002Fimages\u002Fmcp-server-header.webp",{},"MCP Servers and Tools: How LLMs Reach the World","\u002Fblog\u002Fmcp-servers-and-tools",{"title":947,"description":949},"blog\u002Fmcp-servers-and-tools",[497,957,941,958],"tools","agents","H-rIH_hhI4AEIuhMm5XBRn8GqeNa7Ie_6MmiMlzHllw",{"id":961,"title":962,"author":309,"category":459,"date":963,"description":964,"extension":12,"featured":13,"image":965,"meta":966,"metaDescription":967,"metaTitle":968,"navigation":18,"path":969,"seo":970,"stem":971,"tags":972,"__hash__":976,"__readTime":71},"blog\u002Fblog\u002Fmeasurable-beats-impressive-blits-ai-at-ai-everything-abu-dhabi.md","Measurable Beats Impressive: 75 Conversations at Ai Everything Abu Dhabi","2026-10-15","Len, Coen and I spent two days on our stand at Ai Everything Abu Dhabi and had more than 75 conversations with banks, government entities, telcos, investors and fellow founders. Five themes kept coming back, and together they point one way: from impressive demos to measurable results in production.","\u002Fblog\u002Fblits-ai-everything-abu-dhabi-recap-hero.png",{},"What more than 75 conversations at Ai Everything Abu Dhabi taught us about enterprise AI in the Gulf: production, sovereignty, agents, language and trust.","75 Conversations at Ai Everything Abu Dhabi 2026","\u002Fblog\u002Fmeasurable-beats-impressive-blits-ai-at-ai-everything-abu-dhabi",{"title":962,"description":964},"blog\u002Fmeasurable-beats-impressive-blits-ai-at-ai-everything-abu-dhabi",[973,974,52,975,53],"events","abu dhabi","financial services","ElFxoh8zy2QDQJsKiNXkHDIYRHEUUfMjIaRxR4vFaD0",{"id":978,"title":979,"author":8,"category":28,"date":980,"description":981,"extension":12,"featured":13,"image":982,"meta":983,"metaDescription":34,"metaTitle":34,"navigation":18,"path":984,"seo":985,"stem":986,"tags":987,"__hash__":991,"__readTime":39},"blog\u002Fblog\u002Fmeasure-ai-performance-and-set-the-right-kpis.md","Measure AI Performance and Set the Right KPIs","2026-03-07","AI performance means more than accuracy. These are the production KPIs that matter (latency, P99, TTFT and error rates) and how to measure them.","\u002Fimages\u002Fblits-performance.jpg",{},"\u002Fblog\u002Fmeasure-ai-performance-and-set-the-right-kpis",{"title":979,"description":981},"blog\u002Fmeasure-ai-performance-and-set-the-right-kpis",[988,989,990,53],"ai kpi","ai performance","observability","7izChjS6RtzVMstKY8mFca2uii-3N31S4GP1zVQOxX0",{"id":993,"title":994,"author":8,"category":28,"date":995,"description":996,"extension":12,"featured":13,"image":997,"meta":998,"metaDescription":999,"metaTitle":1000,"navigation":18,"path":1001,"seo":1002,"stem":1003,"tags":1004,"__hash__":1006,"__readTime":106},"blog\u002Fblog\u002Fmost-enterprise-ai-never-leaves-the-pilot.md","Most Enterprise AI Never Leaves the Pilot. Ours Goes Live.","2026-11-13","The line on our home page is a big claim, so here is what sits behind it. What the research says about enterprise AI pilots, the ten ways we see pilots die, and what we do differently so that a use case reaches production and stays there.","\u002Fblog\u002Fblits-pilot-to-production-hero.png",{},"Why most enterprise AI stalls in the pilot, what the studies from MIT, S&P, IDC, Deloitte and Gartner show, and how we get enterprise AI into production.","Why Most Enterprise AI Never Leaves the Pilot","\u002Fblog\u002Fmost-enterprise-ai-never-leaves-the-pilot",{"title":994,"description":996},"blog\u002Fmost-enterprise-ai-never-leaves-the-pilot",[53,54,1005,136,56],"ai in production","tsDVW3yzbg6ITwAmCYpbM1m5Fmicz2U6xP685igxzPQ",{"id":1008,"title":1009,"author":8,"category":28,"date":1010,"description":1011,"extension":12,"featured":13,"image":1012,"meta":1013,"metaDescription":34,"metaTitle":34,"navigation":18,"path":1014,"seo":1015,"stem":1016,"tags":1017,"__hash__":1019,"__readTime":121},"blog\u002Fblog\u002Fnobody-budgets-in-tokens.md","Nobody Budgets in Tokens","2026-08-12","Per seat licensing carried software for a decade, but agents don't sit in seats and token pricing breaks the moment AI succeeds. Where AI pricing lands.","\u002Fblog\u002Fblits-pricing-eras-hero.png",{},"\u002Fblog\u002Fnobody-budgets-in-tokens",{"title":1009,"description":1011},"blog\u002Fnobody-budgets-in-tokens",[1018,53,52,54],"ai pricing","hkBVqOB8OWhD605LVQ0T2z7EUAGFurKvJc45xadSVxs",{"id":1021,"title":1022,"author":309,"category":28,"date":1023,"description":1024,"extension":12,"featured":13,"image":1025,"meta":1026,"metaDescription":34,"metaTitle":1027,"navigation":18,"path":1028,"seo":1029,"stem":1030,"tags":34,"__hash__":1031,"__readTime":24},"blog\u002Fblog\u002Fopening-the-natural-language-understanding-nlu-blackbox.md","Opening the Natural Language Understanding (NLU) Blackbox","2020-06-23","NLU is often treated as a black box, but knowing how it works is key to building effective conversational AI. A look inside.","\u002Fblog\u002FMonitoring-300x234.png",{},"Opening the NLU Black Box","\u002Fblog\u002Fopening-the-natural-language-understanding-nlu-blackbox",{"title":1022,"description":1024},"blog\u002Fopening-the-natural-language-understanding-nlu-blackbox","qPChjo2cCcMnFSlIS91oCjCT84N-DppSPKyROYaYC2Q",{"id":1033,"title":1034,"author":290,"category":28,"date":1035,"description":1036,"extension":12,"featured":13,"image":1037,"meta":1038,"metaDescription":1039,"metaTitle":1040,"navigation":18,"path":1041,"seo":1042,"stem":1043,"tags":1044,"__hash__":1045,"__readTime":251},"blog\u002Fblog\u002Fprompt-injection-enterprise-ai-agents.md","Prompt Injection: The Attack Every Enterprise AI Agent Will Face","2026-11-20","Prompt injection is the number one risk on the OWASP list for LLM applications, and security agencies say it may never be fully solved. For AI agents that read emails and documents and act in your systems, that changes how you design them. What prompt injection is, real incidents, and five layers that keep a successful attack from doing real harm.","\u002Fblog\u002Fblits-prompt-injection-hero.png",{},"What prompt injection is, why it is the top risk for LLM applications, why it cannot be fully solved, and five layers that limit the damage for AI agents.","Prompt Injection: The Attack Every AI Agent Will Face","\u002Fblog\u002Fprompt-injection-enterprise-ai-agents",{"title":1034,"description":1036},"blog\u002Fprompt-injection-enterprise-ai-agents",[943,942,52,136,56],"mBu03eY3ykp-MVizY4m6FATerfgXKC3E7Rka_o9M8vc",{"id":1047,"title":1048,"author":8,"category":28,"date":1049,"description":1050,"extension":12,"featured":13,"image":1051,"meta":1052,"metaDescription":34,"metaTitle":1053,"navigation":18,"path":1054,"seo":1055,"stem":1056,"tags":1057,"__hash__":1058,"__readTime":499},"blog\u002Fblog\u002Frag-in-2026-why-enterprise-pipelines-still-fail.md","The RAG That Passed Its Demo and Quietly Died in Production","2026-04-16","A knowledge assistant that aces its pilot can fail silently, recall sliding while dashboards stay green. Why good RAG rots after the demo.","\u002Fimages\u002Fblits-hallucination.jpg",{},"RAG in 2026: Why Enterprise Pipelines Still Fail","\u002Fblog\u002Frag-in-2026-why-enterprise-pipelines-still-fail",{"title":1048,"description":1050},"blog\u002Frag-in-2026-why-enterprise-pipelines-still-fail",[926,53,497,927,334],"30WucP_fwRnTHUJ8UHdi6VXrjrevV1S2Uhl6sQMEEMY",{"id":1060,"title":1061,"author":8,"category":28,"date":1062,"description":1063,"extension":12,"featured":13,"image":1064,"meta":1065,"metaDescription":1066,"metaTitle":1061,"navigation":18,"path":1067,"seo":1068,"stem":1069,"tags":1070,"__hash__":1074,"__readTime":39},"blog\u002Fblog\u002Fretail-ai-use-cases-that-survive-the-demo.md","Retail AI: The Use Cases That Survive Beyond the Demo","2026-11-27","Retail AI demos are all shopping assistants that talk like a friend with great taste. The retail AI that survives in production is mostly pricing, forecasting, fraud screening and product data. What 123 public deployments tell us, what Walmart, Amazon and IKEA actually report, and where the shopping assistant finally earns its place.","\u002Fblog\u002Fblits-retail-ai-hero.png",{},"Which retail AI use cases make it to production? What 123 public deployments, Walmart, Amazon, IKEA and Lowe's show, and why the boring ones pay first.","\u002Fblog\u002Fretail-ai-use-cases-that-survive-the-demo",{"title":1061,"description":1063},"blog\u002Fretail-ai-use-cases-that-survive-the-demo",[1071,1072,52,233,1073],"retail ai","ecommerce","ai use cases","-ud86QKo5OoTfyhYMSq4I40xq-UYZWhtOdJTUPcddHU",{"id":1076,"title":1077,"author":8,"category":28,"date":1078,"description":1079,"extension":12,"featured":13,"image":1080,"meta":1081,"metaDescription":1082,"metaTitle":1083,"navigation":18,"path":1084,"seo":1085,"stem":1086,"tags":1087,"__hash__":1092,"__readTime":121},"blog\u002Fblog\u002Fswiss-data-protection-law-vs-gdpr-for-ai-services.md","Swiss Data Protection Law vs GDPR: What It Means for AI Services","2026-12-22","Switzerland has no AI Act and its data protection law looks a lot like GDPR, so AI companies often assume one compliance program covers both. It nearly does. Where the Swiss FADP differs for AI services, why the EU AI Act reaches Swiss companies anyway, and what to arrange if you serve customers on both sides of the border.","\u002Fblog\u002Fblits-swiss-law-vs-gdpr-ai-hero.png",{},"AI service compliance in Switzerland vs the EU: how the Swiss FADP differs from GDPR for AI, why the EU AI Act still reaches Swiss companies, and a checklist.","Swiss Data Protection Law vs GDPR for AI Services","\u002Fblog\u002Fswiss-data-protection-law-vs-gdpr-for-ai-services",{"title":1077,"description":1079},"blog\u002Fswiss-data-protection-law-vs-gdpr-for-ai-services",[1088,1089,1090,216,1091],"data protection","switzerland","gdpr","ai compliance","HTi0ta-8IDYFkEmZSIVcvzJD8nuPg8rih8YxluBvEKE",{"id":1094,"title":1095,"author":8,"category":28,"date":1096,"description":1097,"extension":12,"featured":13,"image":1098,"meta":1099,"metaDescription":34,"metaTitle":1100,"navigation":18,"path":1101,"seo":1102,"stem":1103,"tags":1104,"__hash__":1106,"__readTime":39},"blog\u002Fblog\u002Ftext-to-speech-engines-and-why-they-matter.md","Text-to-Speech Engines: The Voice Layer Every AI Product Needs","2026-03-03","Text to speech is core AI infrastructure. What TTS engines are, the model approaches, the providers that matter, and why speed and dialect decide.","\u002Fimages\u002Fblits-voice.jpg",{},"Text to Speech Engines: The Voice Layer for AI","\u002Fblog\u002Ftext-to-speech-engines-and-why-they-matter",{"title":1095,"description":1097},"blog\u002Ftext-to-speech-engines-and-why-they-matter",[332,302,1105],"ai infrastructure","Eu3je_dRENeJZno3LYJ_fO5Q8qAgtwBhXUoOKz3wGcA",{"id":1108,"title":1109,"author":8,"category":28,"date":1110,"description":1111,"extension":12,"featured":13,"image":1112,"meta":1113,"metaDescription":1114,"metaTitle":34,"navigation":18,"path":1115,"seo":1116,"stem":1117,"tags":1118,"__hash__":1120,"__readTime":106},"blog\u002Fblog\u002Fthe-ai-bank-runs-on-governance.md","The AI Bank Runs on Governance","2026-09-30","Our banking use case map grew from 21 to 81 jobs, so we turned it into a playbook. Along the way the story changed: the destination is the AI bank, where the work itself runs on governed AI, and what decides who gets there is governance far more than the models.","\u002Fblog\u002Fblits-ai-bank-eras-hero.png",{},"Our banking use case map grew from 21 to 81 jobs. The destination is the AI bank, and what decides who gets there is governance more than the models.","\u002Fblog\u002Fthe-ai-bank-runs-on-governance",{"title":1109,"description":1111},"blog\u002Fthe-ai-bank-runs-on-governance",[1119,136,469,52,304],"ai bank","ziFVXhwwYmus7n690c2_Rwb-4y8O4_M3dywY49MRwnc",{"id":1122,"title":1123,"author":8,"category":28,"date":1124,"description":1125,"extension":12,"featured":13,"image":1126,"meta":1127,"metaDescription":34,"metaTitle":34,"navigation":18,"path":1128,"seo":1129,"stem":1130,"tags":1131,"__hash__":1133,"__readTime":106},"blog\u002Fblog\u002Fthe-best-ai-model-doesnt-exist.md","The Best AI Model Doesn't Exist","2026-07-27","62 AI model versions got a retirement date this year. Why locking your business to one model is a hidden risk, and why staying model agnostic wins.","\u002Fblog\u002Fblits-model-agnostic-hub.png",{},"\u002Fblog\u002Fthe-best-ai-model-doesnt-exist",{"title":1123,"description":1125},"blog\u002Fthe-best-ai-model-doesnt-exist",[1132,53,497,333,54],"model agnostic","JqdV1Cu_gxRCBVp7H6x-wfyip2yVMkHRo7Tu0G_4gnM",{"id":1135,"title":1136,"author":8,"category":61,"date":1137,"description":1138,"extension":12,"featured":13,"image":1139,"meta":1140,"metaDescription":1141,"metaTitle":1142,"navigation":18,"path":1143,"seo":1144,"stem":1145,"tags":34,"__hash__":1146,"__readTime":24},"blog\u002Fblog\u002Fthe-best-way-for-chatbots-to-actually-understand-your-customers.md","Blits.ai: The Best Way for Chatbots to Actually Understand Your Customers","2021-02-11","We're introducing a new method at Blits.ai that changes how chatbots work under the hood. From the start our idea was simple: companies should be able to use the best performing engines on the market without having to pick one vendor and hope for the best. Right now we are two things at once, a middleware layer and a low code platform, and through it our customers get access to 40+ cognitive AI services.","\u002Fblog\u002F1611759270728.png",{},"A new Blits.ai method lets chatbots use the best AI engines on the market without betting on one vendor, through 40+ cognitive AI services.","How Chatbots Truly Understand Your Customers","\u002Fblog\u002Fthe-best-way-for-chatbots-to-actually-understand-your-customers",{"title":1136,"description":1138},"blog\u002Fthe-best-way-for-chatbots-to-actually-understand-your-customers","jlCmU9T6D-chVvdsZIMpRg5Gk4hbzvt7Mut-plLU6ZY",{"id":1148,"title":1149,"author":309,"category":28,"date":1150,"description":1151,"extension":12,"featured":13,"image":1152,"meta":1153,"metaDescription":1154,"metaTitle":34,"navigation":18,"path":1155,"seo":1156,"stem":1157,"tags":1158,"__hash__":1161,"__readTime":39},"blog\u002Fblog\u002Fthe-chargeback-arms-race.md","The Chargeback Arms Race","2026-10-03","Chargebacks are heading for 337 million a year, most of them aren't real fraud, and each one costs a multiple of its face value. A practical look at the six AI jobs hiding in one dispute, what each of them needs to be governed, and why agentic commerce will change the volume.","\u002Fblog\u002Fblits-chargeback-loop-hero.png",{},"Chargebacks head for 337 million a year and most are not fraud. The six AI jobs in one dispute, how to govern each, and what agentic commerce changes.","\u002Fblog\u002Fthe-chargeback-arms-race",{"title":1149,"description":1151},"blog\u002Fthe-chargeback-arms-race",[1159,284,52,53,1160],"chargebacks","fraud","HPFO_kwJvdXKmDRJf83bSyREvy4-rH815TsB1ButS64",{"id":1163,"title":1164,"author":154,"category":28,"date":1165,"description":1166,"extension":12,"featured":13,"image":1167,"meta":1168,"metaDescription":1169,"metaTitle":34,"navigation":18,"path":1170,"seo":1171,"stem":1172,"tags":1173,"__hash__":1176,"__readTime":106},"blog\u002Fblog\u002Fthe-company-that-calls-first.md","The Company That Calls First","2026-09-23","A brief landed on my desk asking for 600,000 outbound calls a month, by AI. The pattern is not a banking story: it applies to every enterprise with customers worth calling. Why human dialling died, what insurers, banks, lenders and software companies already run, and the anatomy of a governed outbound call.","\u002Fblog\u002Fblits-outbound-hero.png",{},"A brief asked for 600,000 outbound AI calls a month. Why human dialling died, what insurers, banks and lenders already run, and how a governed outbound call works.","\u002Fblog\u002Fthe-company-that-calls-first",{"title":1164,"description":1166},"blog\u002Fthe-company-that-calls-first",[1174,302,53,52,1175],"outbound ai","sales","lex9hp4VJze5EHsfvgwSNDTKzRgcx7xuHKtS3Co1hxc",{"id":1178,"title":1179,"author":309,"category":28,"date":1180,"description":1181,"extension":12,"featured":13,"image":1182,"meta":1183,"metaDescription":1184,"metaTitle":1185,"navigation":18,"path":1186,"seo":1187,"stem":1188,"tags":1189,"__hash__":1191,"__readTime":251},"blog\u002Fblog\u002Fthe-coordination-layer-what-to-build-when-models-converge.md","The Coordination Layer: What to Build When Models Converge","2026-10-20","Models are converging and getting cheaper to swap, so the durable engineering in enterprise AI is moving somewhere else: the layer that decides who may act, what is remembered, when a human takes over and what gets evidenced. A build guide for the part of your AI stack you'll still own in five years.","\u002Fblog\u002Fblits-coordination-layer-hero.png",{},"Models converge and get cheaper to swap, so lasting work moves to the layer that decides who acts, what is remembered and when a human takes over.","What to Build When AI Models Converge","\u002Fblog\u002Fthe-coordination-layer-what-to-build-when-models-converge",{"title":1179,"description":1181},"blog\u002Fthe-coordination-layer-what-to-build-when-models-converge",[167,52,53,136,1190],"integration","6bV3tAmavHYlr6iTqMhgdg-TOmMYZipHTMmtAENHZCE",{"id":1193,"title":1194,"author":8,"category":28,"date":1195,"description":1196,"extension":12,"featured":13,"image":1051,"meta":1197,"metaDescription":1198,"metaTitle":1199,"navigation":18,"path":1200,"seo":1201,"stem":1202,"tags":1203,"__hash__":1206,"__readTime":106},"blog\u002Fblog\u002Fthe-danger-of-ai-hallucinations-and-how-businesses-should-handle-it.md","The Danger of AI Hallucinations and How Businesses Should Handle It","2026-02-25","AI hallucinations are a business risk as much as an error. What they are, why models make things up, how often it happens, real cases from courts and companies, practical mitigations, and what you gain by getting it right.",{},"What AI hallucinations are, why models make things up, how often they do, real cases from courts and companies, and how businesses reduce the risk.","AI Hallucinations and How Businesses Handle Them","\u002Fblog\u002Fthe-danger-of-ai-hallucinations-and-how-businesses-should-handle-it",{"title":1194,"description":1196},"blog\u002Fthe-danger-of-ai-hallucinations-and-how-businesses-should-handle-it",[1204,497,1205,53],"ai","risk management","hXB0xURMAestXPn4SvNI1RX6xuzd8zazx12PKSMkFJU",{"id":1208,"title":1209,"author":8,"category":28,"date":1210,"description":1211,"extension":12,"featured":13,"image":1212,"meta":1213,"metaDescription":34,"metaTitle":1214,"navigation":18,"path":1215,"seo":1216,"stem":1217,"tags":1218,"__hash__":1220,"__readTime":499},"blog\u002Fblog\u002Fthe-many-forms-of-rag-explained-without-the-jargon.md","The Many Forms of RAG, Explained Without the Jargon","2026-07-02","Hybrid search, reranking, ColBERT, GraphRAG, self-RAG: a plain language guide to what each RAG variant does and which ones you probably don't need.","\u002Fblog\u002FBlits-ai-RAG-used-to-be-one-thing.png",{},"The Many Forms of RAG, Without the Jargon","\u002Fblog\u002Fthe-many-forms-of-rag-explained-without-the-jargon",{"title":1209,"description":1211},"blog\u002Fthe-many-forms-of-rag-explained-without-the-jargon",[926,53,497,927,1219],"ai explained","I_4mb1aJRQPXhlA7DS18iNF9x69fHflEzyBjmTit6RE",{"id":1222,"title":1223,"author":290,"category":28,"date":1224,"description":1225,"extension":12,"featured":13,"image":1226,"meta":1227,"metaDescription":34,"metaTitle":1228,"navigation":18,"path":1229,"seo":1230,"stem":1231,"tags":1232,"__hash__":1234,"__readTime":1235},"blog\u002Fblog\u002Fthe-missing-text-layer-why-speech-to-speech-fails-enterprise.md","The Missing Text Layer: Why Speech-to-Speech Agents Fail Enterprise Requirements","2026-07-23","Native speech to speech demos beautifully, then fails enterprise needs. Why the text layer, where guardrails and audit live, still wins regulated voice.","\u002Fblog\u002Fblitsai-voice-isnt-safe.png",{},"Why Speech to Speech Voice Agents Fail Enterprise","\u002Fblog\u002Fthe-missing-text-layer-why-speech-to-speech-fails-enterprise",{"title":1223,"description":1225},"blog\u002Fthe-missing-text-layer-why-speech-to-speech-fails-enterprise",[302,1233,53,167,249],"speech to speech","_wRru7BtLUWa3kxmqXFqOH19LuZw0aj0P7uY7jSmSYc","15 min read",{"id":1237,"title":1238,"author":8,"category":28,"date":1239,"description":1240,"extension":12,"featured":13,"image":1241,"meta":1242,"metaDescription":1243,"metaTitle":1244,"navigation":18,"path":1245,"seo":1246,"stem":1247,"tags":1248,"__hash__":1249,"__readTime":106},"blog\u002Fblog\u002Fthe-universal-commerce-protocol-and-why-llms-need-a-new-economic-language.md","The Universal Commerce Protocol and Why LLMs Need a New Economic Language","2026-01-19","Why commerce as we know it does not work for AI and what the Universal Commerce Protocol is trying to fix","\u002Fimages\u002FbannerUCP-blits.png",{},"Why commerce as we know it does not work for AI, what the Universal Commerce Protocol is trying to fix, and why LLMs need a new economic language.","The Universal Commerce Protocol for LLMs","\u002Fblog\u002Fthe-universal-commerce-protocol-and-why-llms-need-a-new-economic-language",{"title":1238,"description":1240},"blog\u002Fthe-universal-commerce-protocol-and-why-llms-need-a-new-economic-language",[],"m0tr5KARy4ydjJJuxAi4PVAYnRO7OhZnTm_iqltnddk",{"id":1251,"title":1252,"author":8,"category":28,"date":1253,"description":1254,"extension":12,"featured":18,"image":1255,"meta":1256,"metaDescription":34,"metaTitle":1257,"navigation":18,"path":1258,"seo":1259,"stem":1260,"tags":1261,"__hash__":1266,"__readTime":88},"blog\u002Fblog\u002Fuse-cases-in-what-the-future-of-banking-ai-actually-looks-like.md","21 Use Cases In: What the Future of Banking AI Actually Looks Like","2026-08-05","21 banking AI use cases, live or in delivery, across the Middle East, Africa and Europe: dialect voice, digital humans, KYC, fraud screening and outbound agents.","\u002Fblog\u002Fblits-banking-ai-hero.png",{},"21 Use Cases: The Future of Banking AI","\u002Fblog\u002Fuse-cases-in-what-the-future-of-banking-ai-actually-looks-like",{"title":1252,"description":1254},"blog\u002Fuse-cases-in-what-the-future-of-banking-ai-actually-looks-like",[469,52,1262,302,1263,1264,1265],"digital humans","super apps","outbound voice","use cases","8xwelyyim84wM89UJ3_AwxnBerKWf2IZG1930y4KjcI",{"id":1268,"title":1269,"author":8,"category":9,"date":1270,"description":1271,"extension":12,"featured":13,"image":1272,"meta":1273,"metaDescription":1274,"metaTitle":1275,"navigation":18,"path":1276,"seo":1277,"stem":1278,"tags":1279,"__hash__":1283,"__readTime":121},"blog\u002Fblog\u002Fusing-cloud-ai-without-sending-personal-data-abroad.md","Using Cloud AI Without Sending Personal Data Abroad","2026-10-16","Strict data rules, no local AI cloud region, no GPUs, and a board that wants agents live this quarter. For a lot of banks and other regulated organizations that used to mean picking two out of three. Our new privacy shield removes the trade off: the best cloud models do the thinking, and the personal data never leaves your building.","\u002Fblog\u002Fblits-privacy-shield-hero.png",{},"Banks and other regulated organizations without a local AI cloud or GPUs can use the best cloud models while personal data stays in their own environment.","Cloud AI Without Sending Personal Data Abroad","\u002Fblog\u002Fusing-cloud-ai-without-sending-personal-data-abroad",{"title":1269,"description":1271},"blog\u002Fusing-cloud-ai-without-sending-personal-data-abroad",[1280,103,1281,975,1282,52],"privacy","on premises","regulated industries","6RznfjQ5ouJ8-p4PHDvoYFz7sJc2d3OrPwHh9ZsvWek",{"id":1285,"title":1286,"author":8,"category":28,"date":1287,"description":1288,"extension":12,"featured":13,"image":1289,"meta":1290,"metaDescription":1291,"metaTitle":1292,"navigation":18,"path":1293,"seo":1294,"stem":1295,"tags":1296,"__hash__":1300,"__readTime":39},"blog\u002Fblog\u002Fwe-read-650-ai-deployments-so-you-dont-have-to.md","We Read 1,300+ AI Deployments So You Don't Have To","2026-10-13","We built a public library of 257 AI use cases and 1,377 real deployments, with every number quoted from its source and labelled with who made the claim. The verdict: AI is doing real work, the industry can't prove it, and the sellers are writing the benchmarks.","\u002Fblog\u002Fblits-use-case-library-hero.png",{},"A public library of 257 AI use cases and 1,377 real deployments, every number sourced. AI does real work, but the industry cannot prove it yet.","Lessons From 1,300+ AI Deployments","\u002Fblog\u002Fwe-read-650-ai-deployments-so-you-dont-have-to",{"title":1286,"description":1288},"blog\u002Fwe-read-650-ai-deployments-so-you-dont-have-to",[1073,1297,53,1298,1299],"ai adoption","ai benchmarks","research","1t8kKQ1aRAV876lRNXk4uMDvFc_IGXhaAaVoH9WtrnI",{"id":1302,"title":1303,"author":8,"category":28,"date":1010,"description":1304,"extension":12,"featured":13,"image":1305,"meta":1306,"metaDescription":34,"metaTitle":1307,"navigation":18,"path":1308,"seo":1309,"stem":1310,"tags":1311,"__hash__":1314,"__readTime":39},"blog\u002Fblog\u002Fwere-not-saas-not-an-integrator-on-purpose.md","We're Not SaaS. We're Not an Integrator. On Purpose.","Procurement wants us to be a software vendor or an implementation partner. We are neither, and I think the model in between is what gets AI into production.","\u002Fblog\u002Fblits-positioning-hero.png",{},"Not SaaS, Not an Integrator, On Purpose","\u002Fblog\u002Fwere-not-saas-not-an-integrator-on-purpose",{"title":1303,"description":1304},"blog\u002Fwere-not-saas-not-an-integrator-on-purpose",[53,54,1312,1313],"platform","company","NNBRDZDagTR5umxOvb2Os1ZCgfhPE4FIwUzmjgyZSYc",{"id":1316,"title":1317,"author":154,"category":28,"date":1318,"description":1319,"extension":12,"featured":13,"image":1320,"meta":1321,"metaDescription":1322,"metaTitle":1323,"navigation":18,"path":1324,"seo":1325,"stem":1326,"tags":1327,"__hash__":1329,"__readTime":71},"blog\u002Fblog\u002Fwhat-ai-native-means-for-a-bank.md","What \"AI Native\" Actually Means for a Bank","2026-12-04","Every bank now has AI. Very few are AI native. The difference is not the number of pilots but whether the bank's processes run on AI, with people making the decisions. What AI native means in practice, what the leaders already show, and five shifts that separate AI added from AI native.","\u002Fblog\u002Fblits-ai-native-bank-hero.png",{},"AI native is more than a chatbot. What an AI native bank looks like, what DBS, JPMorgan and BNY already do, and five shifts from AI added to AI native.","What AI Native Actually Means for a Bank","\u002Fblog\u002Fwhat-ai-native-means-for-a-bank",{"title":1317,"description":1319},"blog\u002Fwhat-ai-native-means-for-a-bank",[1328,56,54,52,136],"ai native","fXtWAaRr-S_BtaDs4aqjaPZPdQRfQ7L3j276gRIvzeI",{"id":1331,"title":1332,"author":8,"category":28,"date":1333,"description":1334,"extension":12,"featured":13,"image":1335,"meta":1336,"metaDescription":1337,"metaTitle":1338,"navigation":18,"path":1339,"seo":1340,"stem":1341,"tags":34,"__hash__":1342,"__readTime":251},"blog\u002Fblog\u002Fwhat-is-artificial-intelligence.md","At this point, most people are afraid to ask a simple question; what is artificial intelligence (AI)?","2023-03-28","Someone once told me stupid questions don't exist. Maybe so, but people are still afraid to look stupid by asking about things that are assumed to be general knowledge, and Artificial Intelligence (AI) is one of them. With ChatGPT and Midjourney everyone loves to talk about AI, yet I only know a handful of people who actually know how it works. So here are the basics.","\u002Fblog\u002F1678649727454.jpeg",{},"Everyone talks about ChatGPT and Midjourney, yet few people know how AI actually works, and many are afraid to ask. So here are the basics.","What Is Artificial Intelligence (AI)?","\u002Fblog\u002Fwhat-is-artificial-intelligence",{"title":1332,"description":1334},"blog\u002Fwhat-is-artificial-intelligence","puXtdqK7g8r2GrDbxm4d77lDflWUdtOTBEqLRtvmeIs",[21,37,50,69,82,118,133,148,163,180,196,246,263,280,299,317,330,348,365,380,411,441,454,467,481,494,512,525,844,861,874,905,939,955,971,986,1003,1016,1030,1056,1103,1117,1130,1145,1157,1172,1188,1202,1217,1231,1247,1260,1278,1295,1310,1341],1794848291254]