Adapting to a new era — building with AI agents

Every few years our field gets a new tool that splits people into two camps: the ones who dismiss it, and the ones who hand it the keys. AI agents are the latest fault line — and both camps are wrong.

The engineers who dismiss them are slower than they need to be. The ones who hand over the keys ship confident, plausible, wrong code. The interesting place to stand is in between: adapt to the tool without surrendering judgment.

The job changed shape, it didn’t shrink

A year ago “senior engineer” mostly meant writing the right code. Today a huge part of the value is deciding what’s right, framing the work so a fast tool can execute it, and reviewing the output like you’d review a sharp but overconfident junior. The typing is cheaper; the thinking is worth more than ever.

That shift rewards exactly the habits good engineers already have — clear requirements, explicit architecture, small reviewable steps — and punishes the ones who skipped them, because an agent will happily scale your lack of a plan.

How I actually work with agents

My rule is simple: agents execute, I stay accountable.

• Recon and plan first. Before any code moves, the problem gets framed: what exists, what the constraints are, what “done” means. An agent pointed at a vague goal produces vague, confident damage.
• Truth over assumptions. When I rebuilt my own portfolio recently, I made the source code the single source of truth — not my memory, not an old résumé. If a claim wasn’t backed by the repo, it didn’t ship. Agents are great at sounding right; they need a factual leash.
• Small steps, real review. I’d rather approve five small, verifiable changes than one big “trust me” diff. Every step gets read, run, and checked before the next.
• Nothing sensitive, nothing public, without a human gate. Deploys, secrets, client data — those stay behind an explicit yes.

Why this matters for the work I do

The result isn’t “AI wrote my site.” It’s that I can move faster and keep the bar high — turn an idea into a reviewed, shipped, production change in an afternoon, without the quality slipping. That’s the real adaptation: the tools got powerful, so the judgment around them has to get sharper, not lazier.

The era changed. The craft didn’t. It just moved up a level — from writing every line to owning every decision.