Something feels different now

I’ve used ChatGPT and Claude since they came out, including a fair bit for coding. Helpful, sometimes surprisingly so. Useful, but not transformative.

This week I tried the new agentic tools: Claude Code (and to a lesser extent OpenAI’s Codex). It feels like a completely new gear.

Not a new UI. Not a marginal gain. More like instructing clones of yourself that can draft, refine, explain its reasoning, and work with you iteratively. It’s still just a tool, but suddenly a much more capable one.

I haven’t been one to jump on the LinkedIn CEO/Founder/AI Salesman hype, but some thoughts are forming:

• This increases output significantly. 2–10× for many tasks feels within reach. In 2 hours yesterday, I did what would have taken me 1–2 days last week.
• Engineers are still responsible for every line. Maybe even more so, because it’s so tempting to just accept a good-looking draft.
• It doesn’t replace skill. The quality of the output matched the quality of my instructions, corrections, and review.
• It shifts the effort: less time typing, more time designing, testing, documenting.

One interesting side effect is that it probably makes small internal tools worth building like: – A tool to verify that critical data landed in the data platform for specific time slices – A quick dashboard to debug a rare heisenbug – A one-off HTTP proxy to rewrite legacy API responses – A CLI to inspect Redis cache usage across workloads – A GitHub Action tailored to one quirky project

You haven’t missed the train if you're not using these yet. Whether you adopt it now or in a month won’t decide who wins. But it does feel inevitable.

I’ll be arguing for using Claude Code at work. I think our backlog throughput will prove the business case on its own.

I don't care for hype, but it might be a real shift in how we work as engineers.