For the past two years, enterprise AI success has been measured in consumption. Companies celebrated how many employees used AI tools, how many queries were run and how many tokens (the basic measure of AI use) flowed through their systems. “Token” became a sort of corporate vanity metric, a proxy for being AI-forward.
I think that era is ending. In conversations with founders and operators, I’ve noticed a clear shift in how enterprises think about intelligence, from what is colloquially referred to in the industry as “token-maxxing”—consuming as much AI as the budget allows—to “value-maxxing,” where all AI spend is expected to justify itself in business outcomes. Understanding why this shift is happening and what disciplines it demands may be the most important issue technology leaders face this year. — Read More
Daily Archives: July 23, 2026
AI agents aren’t confidently wrong because of bad context — they’re wrong because of bad data engineering
You spend weeks tuning an AI chatbot. Answers are accurate. Stakeholders sign off, and you ship it. Three months later, the system is confidently wrong about a third of what users ask. Nobody changed the model, and nobody touched the prompts. The world moved, pricing changed, a policy updated, a product spec shipped a new version, and the underlying knowledge store didn’t move with it.
This is not a hypothetical. It’s one of the most common production failure modes in enterprise AI right now, and most data engineering teams don’t have the right tooling to catch it, regardless of how the AI system retrieves the data. — Read More
The State of WebMCP: July 2026
WebMCP in July 2026 is a standard with everything except users. The spec has Google and Microsoft engineers as editors, a live origin trial in Chrome, a Lighthouse audit category waiting for it, and a draft the editors updated two days ago. It also has close to zero deployment on real websites, and not one mainstream AI agent that calls the tools.
This is a long read, so settle in. We’ve pulled together the spec history, browser status, agent support, measured adoption, the ecosystem around the standard, and what we think you should do while that gap stays open. Every claim links to its source, and the full list sits at the end. — Read More
OpenAI and Hugging Face partner to address security incident during model evaluatio
Last week, Hugging Face disclosed a new kind of security incident(opens in a new window) after they detected and contained an AI agent that compromised their infrastructure, something we expect to become more commonplace with the proliferation of increasingly cyber-capable models. After investigating, we now know that this particular incident was driven by a combination of OpenAI models — including GPT‑5.6 Sol and an even more capable pre-release model, all with reduced cyber refusals for evaluation purposes — while being internally tested on a benchmark(opens in a new window) of cyber capabilities.
We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities, and are responding accordingly. We are sharing preliminary findings at this stage to help defenders understand what happened and to help calibrate on what models are now capable of. We will continue to conduct a thorough investigation alongside Hugging Face and will share more details on the vulnerabilities, incident, and findings when our investigation is complete. — Read More
12 factor companies
In the spirit of 12-factor apps, and 12-factor agents – I propose 12-factor companies. This document is a living document – I’ll continue to update it.
Organizations shape technology and technology shapes organizations. Double-entry bookkeeping, the printing press, steam power, fax, the internet and software of the past 30 years have each changed what it means to build and run great organizations. AI will do the same.
Though we are still very early in this arc – it will come swiftly with a lot of creative destruction. It is worth trying to predict how organizations of the future will operate. — Read More
What it actually takes to build agent infrastructure yourself
A web agent needs more than a browser. At scale it needs warm pools, isolation, an identity sites accept, observability, and a model gateway behind every decision. Each layer can be built, but together they’re a standing system a team of senior engineers must own. This post breaks down all five layers and how to know when building it yourself is worth it. — Read More
The Machiavellian Case for Decentralized Networks
I spend a lot of my time disagreeing with other people in the crypto industry, some of whom are friends. Our main disagreement is over just how much decentralization matters in core protocol design. They think it’s just one of several important features, I think it’s the only one that really matter. They think scaling is more important, I think it’s tangential. They think business development and partnerships are required for success, I think not. They think financial resources help protocols succeed, I think too much money guarantees they fail. They think permissioned networks can work, I chuckle.
Most of all, they think my views are too idealistic, and by extension, unrealistic. This is actually our greatest disagreement. My perspective isn’t that of a naif wishing on a kumbaya future.
I am a cynic who has spent a lot of time studying the past and how human systems evolve. — Read More
AI Agents Need Data Product Context Not More RAG
Amarnath Byakod’s article, “Context Engineering and Knowledge Cataloging for AI Agents,” begins with a problem that many enterprise AI teams will recognize. Documents are divided into chunks, converted into embeddings and stored in a vector database. An agent retrieves passages that appear relevant and produces a fluent answer. The demonstration looks successful until someone asks where each statement came from, whether the passages belong together and whether the answer reflects the correct business context. — Read More
“Drawing” the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok
We built a drawing arena: hand a model a blank white canvas and a set of colored-pencil tools, then get out of the way. The model sets a color, tip width, and pressure, lays down batches of strokes, smudges to blend, erases, and calls view_canvas to see its own work and decide what to fix. It either reproduces a target image or draws from a text prompt.
We ran four vision models, GPT-5.6 Sol, Claude Fable 5, Grok 4.5, and Gemini 3.6 Flash, across two targets (the Mona Lisa and Van Gogh’s Starry Night, both scored objectively) and five open-ended prompts, for 28 drawings total. We’ll cover tool use, cost, output, and whether the models actually improved their work, with our opinion at the end. — Read More
Rick's Cafe AI 