The Day Claude Fails

I work with AI every day. I lead development teams and still keep my hands on the code daily; my main tool is Claude Code: Opus, Sonnet, or Haiku, depending on the moment and the task. Lately, as happens with any AI tool living at the edge of global demand, Claude has its rough days: it degrades, goes down, or doesn't respond the way it should.

And that's when the ground gives way.

The first feeling is pure frustration: I can't work. You sit staring at the screen as if your hand had been cut off. It happens to me. I see it in my teams: devs with years of craft under their belts, enough to know how to work without AI, yet paralyzed when the tool doesn't respond.

For someone with +20 years in tech, it's a strange feeling. I learned to code without AI. I debugged the hard way for years. I wrote thousands of lines without intellisense, without an agent reviewing my PRs, without anything remotely close to what we have today.

So why do we freeze up when AI isn't available?

After turning it over these past few days, I reached a conclusion: it's not that we can't work without AI. It's that we can't deliver at the scale we deliver today.

Those are two very different things.

The Paradigm Shift We Didn't See Coming

Teams used to deliver at human scale. Some more, some less, depending on performance and individual strengths, but always within a reasonably human range. One feature per sprint, one PR a day, a careful 200-line review, one deployment a week.

Today that output has multiplied. And not "a little": it has multiplied enormously.

With subagents wired into ad-hoc frameworks to spec, plan, develop, run code reviews, generate automated tests on the fly, package releases, assist deployments, and orchestrate the flows of our development methodologies, what used to be a sprint is now an afternoon. What used to be a feature is now, without a doubt, many more. Controls that used to be optional are now automatic and exhaustive.

The scale has become, literally, inhuman.

The Real Dependency

That's why, when Claude fails, the feeling of "I can't work" is misleading. I can work. I know how to work. I did it for two decades. And the teams I lead can too: they have the craft to do it without AI. But that's not what's at stake.

What no team can do today, what none of us riding this wave can do, is sustain the pace and volume our clients, the market, and we ourselves have grown accustomed to. AI didn't replace our ability to code: it replaced the ceiling on how much we can deliver in a day.

And that new ceiling, once you know it, isn't easy to step back from without feeling that something has broken.

Maybe that's the honest conversation we owe ourselves. We're not debating whether AI will leave us jobless. We're living, in real time, how it redefined what "a day's work" means. And how, quietly, it made us dependent not on AI to code, but on AI to keep up with the new normal we all built together.

That new normal, and how to design teams, processes, and architectures that take advantage of it without hanging on the next timeout, is the question that interests me most right now. It's also the one we're exploring at Analytix Code Groove with clients going through exactly this transition. If you're thinking about the same thing on your end, I'd like to hear from you.