Insights on AI, machine learning, and software engineering.
Layer-based folders (controllers, services, repositories) made sense for human navigation — but coding agents search, slice context, and patch locally. Domain-driven layout keeps features in one place, cuts wrong-file edits, and makes agent-led changes reviewable.
Read more →Vibe coding and "it works" manual checks feel productive — until load, cost, security, and flaky tool chains show up in production. Here's how to test agents beyond unit and E2E: performance, token budgets, capacity, security, chaos, and what changes in distributed systems.
Read more →Founders and operators keep hearing "AI agents" and assume the next step is a complex, developer-led build. In reality, a lot of "agentic" value is already available out of the box — and building too early can be the most expensive way to learn what you actually needed.
Read more →Most agent pilots fail for one of two reasons: they're too big ("replace the workflow") or they're too vague ("let's try agents"). Here's a pilot approach that produces a real answer in 2–4 weeks.
Read more →The biggest mistake I see is starting with: "Let's build an AI agent." High-ROI teams start with: "What outcome do we need — and what's the lightest-weight agentic approach that achieves it?"
Read more →Voice agents are having a moment — and for good reason: voice can remove friction where typing is slow, impossible, or unnatural. But voice also raises the bar: real-time latency, turn-taking, interruptions, escalation, and trust.
Read more →