Be careful what you wish for. On June 10, Anthropic CEO Dario Amodei published the article “Policy on the AI Exponential”, arguing that frontier AI governance must move beyond just transparency toward binding, FAA-style regulation. He explicitly proposed that the government should have the power to block or deter model deployments if third-party testing revealed unacceptable risks. Just two days later, the US government did exactly that. The sudden removal of Anthropic’s Fable 5 and Mythos 5 models on June 12 triggered a wave of anxiety across the industry. After receiving a US export control directive restricting access for foreign…

Read more…

Last week OpenAI finally released their models on Amazon Bedrock, through both the API and the updated Codex app. This removes one of the biggest barriers for companies wanting to adopt GPT-5.5 and Codex but were held back by security, compliance, procurement, billing, and governance requirements. OpenAI worked closely with Amazon on the launch and even published a short getting started tutorial: Getting Started with OpenAI Models on Amazon Bedrock. If you have only experimented with GitHub Copilot or Claude Code in your organization I really recommend that you give Codex a go. For large, complex code bases it’s in…

Read more…

Last week Cursor published the report Developer Habits, which aggregates data from their entire user base. The reports highlight two important matters that you must consider if you are working at C-level in a company with a software development department. The first one is that the top 10% of all developers (p90) have been pulling away from median developers in absolute number of lines added per week. Median developers have seen a marginal improvement the past year, where the top 10% of all developers have been improving their output performance consistently and now produce on average 10x more lines of…

Read more…

AI is getting smart in ways few people expected. For a long time, one of the standard criticisms of large language models was that they are just next-token prediction models. The argument was that they could only remix material they had already seen during training, and therefore could not produce genuinely novel content. Last week, we had two announcements that challenge this assumption on a deep level. First, Anthropic announced an initial update from Project Glasswing. They collaborated with 50 companies that used their unreleased model Claude Mythos Preview to scan over 1,000 open-source projects, where it identified 6,202 high-…

Read more…

Two weeks ago there were over 2,000 plugins waiting in queue to be manually reviewed for the popular markdown editor Obsidian. The introduction of coding agents and interest for Obsidian as a virtual brain led to an explosion of new plugins being written for the platform, and the review process up until last week was 100% manual. Somewhere in November last year the Obsidian team just stopped reviewing plugins altogether, and started working on a new automated review system that would also process plugin updates, something that was not possible with the old system. On Tuesday last week Obsidian finally…

Read more…

The past few months I have held full-day workshops in agentic engineering for hundreds of developers. And what strikes me every time is how little most people know about how AI models really work. And there’s a good reason for it. I still haven’t encountered any good management-level introduction videos that explain how these models work “under the hood”. So if you haven’t visited any of my workshops, I figured I could spend this newsletter intro talking about context windows and parametric memory. Both key concepts for understanding how LLMs work. Every time an AI model produces output, it can…

Read more…

GPT-5.5 is different. GPT-5.5 is the first model trained at OpenAI’s flagship Stargate site in Abilene, Texas, one of the supercomputing facilities funded by the $500 billion partnership between OpenAI, SoftBank, and Oracle. The site runs on Oracle Cloud Infrastructure, is powered by NVIDIA’s GB200 Blackwell systems, and is purpose-built to train frontier AI at a scale that simply wasn’t possible before. And it shows. GPT-5.5 keeps impressing me every single day. GPT-5.5 is aware of how the content it produces looks and is structured, and it can reason, iterate, and adjust the content in meticulous detail before presenting it…

Read more…

Be prepared to forget everything you have learnt about how to maximize AI productivity by maximizing token usage. In the new token economy less is more, and the new top skill for AI engineers in 2026 will be how to optimize token usage to save cost. When OpenAI launched GPT-5.5 last Thursday they doubled the prices, from $2.50 input / $15.00 output per 1M tokens to $5.00 input / $30.00 output. This matches the pricing for Anthropic Opus 4.7, which is $5.00 input / $25.00 output per 1M tokens. Many companies I talk to are not aware of what token…

Read more…

Two years ago in the summer of 2024 OpenAI launched their first ChatGPT desktop app for macOS. Back then we didn’t have agentic engineering, so the entire app was hand coded using the Apple Swift framework. It took five additional months for OpenAI to ship the ChatGPT desktop app for Windows, but that app was written in React and Typescript and had several limitations compared to the macOS app. They shared no logic internally. Fast forward to 2026, and AI now writes all the code at OpenAI. At the speed things are moving now it’s even more difficult than before…

Read more…

Last week I visited HumanX in San Fransisco, one of the largest AI conferences in 2026 with over 9,000 attendees, 500+ speakers and 400+ journalists. The conference is centered around deep-dive interviews with industry leaders and I visited 36 sessions over four days. It took the entire weekend to summarize all my notes. From the conference the message was clear: the models are good enough, your organization isn’t. The change must be driven from CEO-level, and you must start with the people, not technology. May Habib, CEO at Writer, said this about AI transformational change: “If they don’t have the…

Read more…