A 3-Part Series: Agents, Workflows, and Skills – Build the Right Thing In Part 1, we built a bug investigation agent. In Part 2, we built a content quality pipeline. Both worked. Both had AI doing something I hope you find genuinely useful. But if you read them back to back with a sufficiently critical eye – the kind of eye a good code reviewer develops after seeing the same mistake for the fifth time – you’d notice something I deliberately left in both systems: capability reuse. The bug agent could search code. The content pipeline had its own policy checker.…
Category: <span>AI</span>
A 3-Part Series: Agents, Workflows, and Skills – Build the Right Thing There’s a phrase I’ve used in engineering reviews for years, usually right before someone’s six-week project gets redirected: “Don’t hire a strategist when you need a soldier.” In Part 1, we discuss agents – an autonomous AI system that reasons its way through open-ended problem. If you read it, you know I’m a fan. They’re genuinely capable, and when you deploy one in the right context, it feels like a superpower. But here’s the thing nobody says out loud at AI conferences: most of what you actually need to build…
A 3-Part Series: Agents, Workflows, and Skills – Build the Right Thing Every few years, “intelligent automation” gets a fresh coat of paint – and a fresh wave of hype – while leaving behind a familiar trail of abandoned projects. Expert systems. Neural nets. Rule engines. Ontologies (remember those? Well… it’s coming back. That’ll be a separate blog). Even microservices got swept into the narrative at one point. And now: agents. Here’s the take: this wave is meaningfully different. But the failure mode hasn’t changed. Engineers still tend to grab the shiny new hammer before fully understanding the nail –…
A few days after I published Part 2, a friend forwarded me a message. He’s been reading the blog on his phone during his commute, decided to open the live demo, and found that he couldn’t flag a single cell. He’s not wrong. On a touchscreen, right-click doesn’t exist. Context menus require a long-press that the browser intercepts before the game sees it. The entire flag mechanic – one of the two core interactions in Minesweeper – is simply inaccessible to anyone not sitting at a desk with a mouse. I’d known this. I’d filed it in the back of…
The planning session ended. I saved the conversation, and sat for a moment staring at the blank src/ directory on my screen. I’ve been in software long enough to know that the gap between a good architecture diagram and a working, tested application is where most projects succeed or fail. Anyone can draw boxes and arrows. The execution is where the real engineering happens – where the theory meets the edge cases, where the clean design confronts the messy reality of frameworks, tooling, and the particular perverseness of asynchronous state management. The question I was asking myself, honestly, was: does this hold up?…