PromptQL ads

I gave a talk at AWS re:Invent on why enterprise AI keeps failing. Not because the technology isn't ready, but because we're building it wrong. Here's what I mean. Every chat-with-data product can technically work today. You can connect an LLM to your database, ask questions, get answers. The capability exists. But when a business user asks a loaded question with real impact, they know for a fact they can't trust the output. Even a human would struggle with that question, but at least the human signals uncertainty. AI just delivers a confident answer. This is the "confidently wrong" problem, and it's the real barrier. Not lack of capability. The issue is that AI never admits when it doesn't know something. And here's why that matters: you can only teach something when it says it doesn't know. Think about the humans you trust in business. They're not necessarily the smartest people in the room. What makes them valuable is their ability to pick up context, learn from failures, and signal when they're uncertain. That uncertainty is what lets you teach them. AI doesn't do this. So even when your AI is 90% accurate, you can't tell which answers fall in that problematic 10%. This forces verification on every output. Your team forensically checks results because one wrong answer costs more than ten correct ones. That verification tax destroys ROI. And you can't fix it through feedback because AI never signals that it needs help. The learning loop is broken before it starts. The solution isn't complicated. Build systems that explicitly signal confidence. Show users what the AI knows versus what it's assuming. When it encounters something unfamiliar, flag it. Let domain experts step in naturally. Capture that knowledge. That conversation between user and expert becomes the seed for improvement. The system learns through actual work instead of requiring some separate training process. This is what allows systems to scale beyond the 20-25 table limits you see elsewhere. When AI can admit uncertainty and learn from feedback, the scale problem becomes solvable. Without that loop, more scale just means more confident wrongness. The market thinks enterprise AI is failing because models aren't capable enough. That's wrong. Enterprise AI is failing because systems haven't been designed to communicate limits and learn from feedback. The fix isn't bigger models. It's humbler ones that know when to ask for help. --> Full talk linked in comments.

promptql.io

Thrilled to be named one of the Top AI Startups to Watch in 2026 by Business Insider 🚀 What they spotlighted as our killer feature: “AI accuracy at scale – without requiring messy data to be prepped or moved elsewhere.” Since day one, we’ve been obsessed with one thing: AI accuracy on data as it is today. Your AI should learn your business language – and keep learning as your business and data evolve. That’s the only way to scale accuracy – not by waiting for “perfect” data. We’ve always believed this was the real game-changer. Amazing to see the world take note. 👉 Full feature in the comments. #EnterpriseAI #BusinessInsider #AccurateAI #TopAIStartups

VentureBeat says we're coming for McKinsey and Deloitte's AI business. I can't say I disagree. Here's why I think engineering-led consulting is about to disrupt a $200B industry. After working with a variety of Fortune 500s, including customers we shouldn't have taken, ones we almost lost, and ones where everything clicked, I've learned what separates AI that compounds from AI that stalls. Two non-negotiables: 1. 𝐋𝐞𝐚𝐝𝐞𝐫𝐬 𝐚𝐧𝐝 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬 𝐢𝐧 𝐭𝐡𝐞 𝐬𝐚𝐦𝐞 𝐥𝐨𝐨𝐩. Skip the layers. Put AI leaders with clear KPIs directly with engineers who've shipped AI at scale. 2. 𝐓𝐞𝐜𝐡 𝐭𝐡𝐚𝐭 𝐦𝐚𝐤𝐞𝐬 𝐚𝐜𝐜𝐮𝐫𝐚𝐜𝐲 𝐬𝐜𝐚𝐥𝐚𝐛𝐥𝐞, 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐢𝐦𝐩𝐫𝐞𝐬𝐬𝐢𝐯𝐞. The silent killer at scale: confidently wrong responses. You need an AI solution that knows when it doesn't know and uses that to incentivize users to correct that in real-time and thus scale & maintain accuracy organically in your org. Nothing else can scale. This is why engineering-led consulting with embedded FDEs and platforms built for organizational learning will reshape enterprise AI. The future isn't more pilots. It's AI that compounds with use, when leaders and engineers build it together.

Get the latest jobs and industry news