The coding agent has quietly revolutionized how I work. The real question is no longer can we build it — but what are we choosing to build, and why.
LOC is a terrible productivity metric — and even this terrible metric can't hide what happened. Both figures exclude CSS, markdown, SQL, lockfiles, and binary assets.
What actually matters when code stops being the bottleneck.
The IDE is the past. Embrace the CLI.
Ship → measure → iterate. The agent writes code. Your users tell you if it matters. Close the loop in hours.
Invest in dev infrastructure that allows agent feedback loop.
Ideas that lived in the "too expensive, skip it" column a year ago now belong on the roadmap. The cost curve flipped — our ambition has to flip with it.
Without a plan, ambition just means wrong code, faster.
Dream big, but align first — or you'll ship 2,000 lines of confidently wrong code.
When the agent writes the code, human judgment becomes the constraint. We shift from can we build this to should we — identifying gaps in the spec, catching logical slips before they ship, and setting the right direction.
This is the job now. Questioning, scrutinizing, deciding — not producing.
Three expectations I'm holding myself — and this team — to, starting now.
Be an individual contributor (again).
Ask yourself — is it the right solution, or just a solution?
Automated tests are the norm, not the nice-to-have.
Design thinking must be embedded in product development. Know what to build, why, and what good looks like.
Anyone can generate code now. Fewer people can tell you which code deserves to exist.
Bad taste compounds.