Innovation matters
For the longest time, I believed there was no such thing as a bad prompt. When you ask AI to do something and get an unexpected result, that result gives you feedback. You may have left an important constraint implicit or used a word such as simple that you and the model interpreted differently. Change the prompt and try again. Learn from what happened. Working with AI forces us to make explicit…
Your agents learn while doing the work. They discover where the organization's guidance no longer matches reality. But when the conversation ends, most of that learning disappears with it. Agent memory may preserve some of it for you. What about the rest of the organization? For agents to work effectively inside an organization, they need access to its knowledge. They need to know which APIs to…
Organizations can now build software faster than they can agree on what to build. AI makes a feature that once took weeks possible in days. A product manager can create a prototype without waiting for an engineer, while a support team can automate a workflow on its own. Working software can emerge from almost anywhere in the organization. That sounds like an unambiguous win, until everyone starts…
For years, organizations invested in intranets, enterprise search, wikis, metadata, taxonomies, knowledge graphs, none of this is new. So why does it suddenly feel urgent again? Because the audience changed. Three eras of organizational knowledge We started organizing knowledge using folders. If you wanted to know how something worked, you needed to know where it lived: Engineering/ APIs/…
Every few weeks, someone declares that coding is solved . If they mean turning a well-specified problem into runnable code, they're increasingly right. But was writing runnable code ever the whole problem? Language models have solved software engineering to the extent that word processors solved journalism: they made producing the output easier. Ask one to build a REST API or implement a React…
Inference efficiency is about protecting attention. Every unnecessary step an AI agent takes, spends reasoning capacity on the tooling instead of the user's problem. For decades, we've worked to make software easier for people to use. We designed intuitive APIs and polished CLIs. We wrote better documentation and meaningful error messages. Good developer experience reduced friction, so developers…
I want to try something new with this blog. Most websites treat reading as consumption: you open a page, read it, and close the tab. But reading often starts a line of thinking that continues long after the page is gone. What if the page could support that thinking without getting in the way? I've added marginalia, a place for your thoughts next to mine. Stop when something catches your attention…
When we measure AI productivity, we tend to ask one question: how much time did this save me? That's the wrong way to look at it. The one that actually matters is how much time did this save everyone . It sounds like a nuance, but it changes the whole picture. The hidden cost shift For centuries, writing had a built-in quality filter. It was expensive. Writing an email, a report, or a proposal…
Open your favorite AI coding agent and you're greeted with a surprising number of decisions. Which model should you use? Should reasoning be low, medium, or high? How much context should the agent load? Which MCP servers should be enabled, which skills should it have access to, and should it ask before running commands? Most of us don't know the right answers. We pretend we do, but if we're…
When ChatGPT took off, everyone wanted to learn prompt engineering. People shared templates and debated personas, XML, markdown, and chain-of-thought. Entire businesses emerged around writing the perfect prompt. We've been optimizing the wrong thing. The biggest productivity gains from LLMs come from learning to externalize judgment. Better prompts barely move the needle by comparison. I'll know…