I think Claude and I are moving too fast
I built a Rails API in 45 minutes last week. Registration, login, session management, content endpoints and 112 rspec tests. This used to take me 2-3 days, possibly with tears. I shipped it before lunch and felt like Bradley Cooper in Limitless, the first act, where everything clicks and the world bends to your will.
Then I spent the next two hours untangling a session management bug that Claude had confidently introduced in minute 12, which I’d missed because I was already on the next feature. I’d moved so fast through the codebase that I couldn’t retrace how I’d got there. Limitless, the second act—waking up somewhere unfamiliar with no memory of the journey.
The deeper I go with AI, the more I recognise that it is not any single tool or workflow, but an attitude that shifts how you move through your day. You stop tolerating friction. You develop a kind of ambient impatience that bleeds out of the terminal and into the rest of your life.
I spoke to someone last week whose partner had complained that he’d started talking to her in imperatives. He blamed his constant, terse Wispr dictations to Claude. He was joking—mostly. But I recognised it, because I’ve noticed something similar in myself: the twitchiness when anything takes more than thirty seconds, the reflex to delegate a thought to Claude before I’ve actually had the thought. I’ve started treating my own thinking as a bottleneck to be optimised away, which is—if you stop and look at it—a genuinely unhinged thing for a human to do.
I know my limits with knowledge work and I can tell you that playing with Claude most of the day, I reach that place 4-5x faster than anything else I’ve ever done. I’ve never experienced anything like it. There are no guardrails, no speed limit. That sounds exciting and it is. But is also means is you can now smash your head against a wall at 20x the speed.
And because everything looks like work, you don’t catch it. You’re not doomscrolling—you’re shipping features, wiring up automations, generating reports. Sure, you need a second job to read all the reports Claude has generated for you. But things are happening.
I wrote last time about engineers building elaborate MCP pipelines to manage problems that really needed a boundary, not a dashboard. That piece was about other people’s avoidance. This one’s about mine. I’ve spent entire afternoons building Claude commands and automating reports—felt enormously productive doing it—and some of those commands are genuinely useful. Others were a way of staying busy while avoiding a harder question. The tool didn’t force that. But it made the avoidance feel like craftsmanship.
The real cost isn’t dramatic. It’s the slow erosion of your ability to sit with a problem without reaching for a solution. Especially in leadership, where the most valuable thing you do often produces no visible output: sitting with ambiguity long enough for the real problem to surface, resisting the urge to ship a decision just because you can.
I’m not arguing against speed. I use Claude every day. I built five new skills last week and I’ll build more. But the most sophisticated thing I’ve done with these tools isn’t a technical workflow—it’s learning to trust signals that don’t come from a terminal. A tight chest at 2pm. The realisation I haven’t looked out of the window since breakfast. The moment I catch myself composing a prompt about something I haven’t actually thought about yet.
Those signals are asking for the same thing every time: a pause. Not a long one—just enough to ask whether the thing I’m about to build is the thing that actually needs building. Enough to let a half-formed idea stay half-formed for another hour, instead of immediately asking Claude to finish my sentence.
AI compresses the pause between stimulus and response to near zero, if you let it. That pause is where judgment lives. Protect it.