
thought · ai
Living with Agents
Agents are already in the room. Not as a movie villain — as a new kind of coworker that never sleeps, never gets tired of boilerplate, and never replaces the part that has to care.
What I think
I don't think agents are magic. I think they're leverage.
A good agent amplifies whatever you already are. If your intent is sharp, it multiplies your speed. If your intent is muddy, it multiplies the mess. That's why this moment feels both exciting and uncomfortable: the bottleneck is shifting from "can I type this fast enough?" to "do I actually know what I want?"
For me, as a developer, that's the real shift. Agents aren't here to write my identity. They're here to remove friction between an idea and a running thing. The craft doesn't disappear — it moves upstream, into taste, architecture, and the courage to say "no, that's not good enough."
Will they replace us?
Replace us? No — not the whole human.
Replace tasks? Yes. Quietly. Relentlessly. A lot of the work we used to call "being a developer" was really orchestration: searching docs, wiring the same patterns, translating vague tickets into concrete diffs. Agents eat that for breakfast.
They also replace roles we refuse to evolve. If your value is only producing volume — lines, tickets closed, meetings attended — you're competing with something that doesn't get tired. If your value is judgment, accountability, and the ability to own an outcome when it breaks at 2 a.m., you're still necessary. Maybe more than before.
So the better question isn't "will AI take my job?" It's: which parts of my job am I still proud to keep?

How to live with them
Treat agents like junior teammates with infinite speed and zero shame. You direct. You review. You own the merge.
- Direct clearly. Vague prompts get vague software. Constraints are kindness.
- Review like it matters. Speed without scrutiny is just faster technical debt.
- Keep your hands dirty. If you never touch the code, you lose the intuition that tells you when something smells wrong.
- Own the outcome. When an agent ships a bug, the signature on the commit is still yours.
Living with agents isn't about worship or fear. It's about building a practice: ask better questions, verify harder, ship cleaner.
What I'm expecting
I expect the floor to rise. Mediocre work gets cheaper. Good work gets more visible because the excuses get thinner.
I expect smaller teams to punch above their weight — and larger teams to feel awkward if they can't explain what humans are still for.
I expect myself to keep learning in public. This blog exists partly because of that: writing forces clarity, and clarity is the skill agents can't fake for you.
And I expect a quiet split — not overnight, not cinematic — between people who use this wave to become more themselves, and people who wait for permission that never arrives.

If we don't do anything
Here's the part I keep coming back to.
If we do nothing — if we scroll past the tools, shrug at the demos, wait for someone else to "figure out AI" — we don't get a dramatic ending. We just lose.
We lose relevance, slowly. We lose agency over how the work gets done. We lose the chance to shape the tools instead of being shaped by them. Not because machines hate us. Because the people who practice will move, and the gap compounds.
Inaction isn't neutral. It's a decision to fall behind while pretending we still have time.
So I'm choosing the other path: learn, build, review, write, repeat. Agents won't wait. Neither should we.