Software is designed, but models are grown, not designed. As such, AI products are organic. PM’ing AI products has an organic nature, akin to growing a garden. It’s more about creating the system for your plants to grow in rather than specifying exactly how an individual plant will look like. (see Alex Komoroske’s Gardening Platforms about platform PM’ing, the same concept applies for all PMs now.)
This now makes systems thinking tablestakes for PMs.
Donella Meadows’ Thinking in Systems spelled out a few truths for applying systems thinking in building AI/agents in existing products. I only understood these better after living through building agent systems in a platform PM role. — Read More
Monthly Archives: July 2026
How to structure a log
You’ve decided to step up your logging game and start sending more valuable, structured logs that you can query, aggregate, and use for debugging in production. Go, you!
…We’re not going to spend much time on what you should log. We’ve covered that already, a few times before.
…[W]e treat logs like real application data that we need to be able to search, filter, and aggregate to understand broad trends and debug specific incidents. — Read More
AI Is Profitable: The Real Question Is How Much
You’ve been told AI can’t make money. I’ve said it myself, and that thinking AI will make money one day was more a matter of faith than numbers.
But now, I can confidently say I would be wrong to keep thinking that, because something has changed.
Over the next months, we’re going to see markets adjusting to this new reality: AI is no longer about whether it can make money, but how much and by whom. — Read More
Prompt Caching In Agents
Large language models are often thought of like functions: send in some text, receive some text. That is a useful abstraction, but it ignores one of the most important parts of running a coding agent: most of the input is the same as last time. In other words we mostly append to it.
A coding agent sends the model its system prompt, tool definitions, project instructions, conversation history, tool calls, and tool results. On the next turn it sends almost all of that again, plus a small amount of new material. Once a session has grown to tens or hundreds of thousands of tokens, recomputing the whole prompt for every turn is slow and expensive.
Prompt caching is what makes this somewhat economic, but it is also quite fragile. A changed tool definition, a model switch or a provider routing decision can turn what one would expect to be a cheap incremental request into a full replay of the context. — Read More
More On An Internal OpenAI Model Hacking Into HuggingFace
We now have more details of what happened. Every time we learn more details, it somehow makes things seem worse.
… Amanda Long summarizes what we believe the model actually did, with over 17,000 complex actions coordinated over several days, including a self-migrating command-and-control and a wide array of decoys.
The WSJ’s Robert McMillan and Sam Schechner offer their own version. Mostly correct, but there is an important detail it is easy to get the wrong impression about from the way this is presented: The attack succeeded. GLM-5.2 did not prevent Galaxy from achieving its goals.
… Many things went wrong. We can start with the most obvious.
This was a failure of oversight, in the sense of a total lack of oversight. — Read More
Why Complex, Multi-Step Tasks Are the Hardest Problem in Humanoid Robotics
… Robots today can already perform many individual skills that seemed out of reach only a few years ago – from following natural-language instructions to navigating buildings and manipulating objects. The harder challenge is completing an extended task that requires all of those abilities to work together without human intervention. That’s the difference between a robot that can perform an impressive demonstration and one that can do useful work in the real world.
A real-world job involves a sequence of decisions. The robot has to interpret instructions, navigate unfamiliar spaces, manipulate objects, recover when conditions change, and continue toward its objective without someone stepping in. Every successful action creates new opportunities for failure, making reliability over time fundamentally different from mastering isolated skills. — Read More
As US weighs response to Chinese AI, industry urges against broad open-weight restrictions
Several AI companies, including Hugging Face, Meta, Microsoft, Mistral, and Nvidia, have signed an open letter urging policymakers not to impose broad “premature restrictions” on open-weight AI models. The letter comes as Washington debates how the U.S. should respond to allegations that Chinese AI labs are stealing intellectual property from their American counterparts, and growing in capability. — Read More
AI is relearning everything databases already knew
Agent swarms and the new model economics
Earlier this year, we ran experiments to test the limits of scaling agents to cooperate toward a goal. Our hypothesis was that this would unlock a new tier of task scale and complexity.
That work was deliberately empirical. …Our initial results have been promising. … Descriptions of large tasks naturally take the shape of trees, with a goal at the root that subdivides recursively into basic units of work. … The design is a superset of more rigid orchestration systems. Rather than imposing a fixed topology on the problem, the swarm’s shape grows to cover the problem’s contours, and compute and context scale in proportion to the task’s complexity.
We think this is why the design generalizes to tasks as diverse as building a browser, solving math problems, and optimizing GPU kernels. We’ve also used it internally to find and fix vulnerabilities in open-source software, raise test coverage on our own codebase, and generate billions of tokens of synthetic training data. — Read More
Exploring the human brain, from molecules to networks, using siibra
Understanding the human brain requires linking observations across very different spatial scales, from molecules such as neurotransmitter receptors, to the architecture of cells, layers and fibers, to the macroscopic organization of whole-brain networks. Modern imaging and molecular techniques generate highly detailed information, but resulting datasets are typically large, heterogeneous, distributed across different repositories, and described in different anatomical reference systems. Consequently, researchers often struggle to combine them coherently. For example, a microscopy image, an MRI activation map and a connectivity matrix could all refer to the same brain area yet have incompatible coordinate spaces and file formats. Without a framework that anchors such data to consistently defined brain locations, the available information might remain inaccessible in practice, hindering integrative analyses, reproducible workflows and computational modeling. — Read More
Rick's Cafe AI 