104 hours

I've been quiet for a while.

Heads down. Eyes on the screen. No posts, no thought-leadership, no "clever takes" on LinkedIn posts. There's a reason. I'm trying to be about it instead of just talking about it.

This morning I went looking at my own work timeline because I was curious what the real shape of AI-augmented solo dev actually looks like once you strip out the hype. — The data was interesting enough for me to break my silence.

What I did

I have a git repo — every commit timestamped. I also have every Claude Code session I've spent on this project sitting on disk — every prompt I've typed, every response Claude generated, all timestamped down to the millisecond.

So, developed a script that merged the two timelines, grouped events that happened within 60 minutes of each other into "work sessions" and measured it all.

The script is small enough to drop into any git repo and run. I'll share it in a follow-up if anyone wants to do the same audit on their own project.

The numbers

The project is 38 calendar days old. I've worked on it on 26 of those 38 days — about two-thirds of the calendar.

In that span: 104 active hours.

Average 4 hours per active day. Median session is 49 minutes. Longest single session was 6.2 hours. The top three days were heavy ones — 12.8h, 10h, 8.7h.

14,134 conversation turns with Claude. 137 git commits.

Ninety-nine percent of the events in my timeline are dialogue. One percent are commits.

What that actually means

Let me tear down a few mental models that don't survive the data.

It's not "AI ships your product in 2 weeks while you sleep." 104 hours is real work. Spread over five-plus weeks of life happening in the background — meals, meetings, kids, the dog, sleep, the whole calendar. That's normal solo-dev cadence. AI didn't change the rhythm of my days. It changed what fits inside them.

It's not "I type a prompt and walk away." Ninety-nine percent of the timeline is me in dialogue. Reviewing output. Pushing back on choices I don't agree with. "No, that button is too big." "Use tabs, not pills." "Why are we doing it this way?" The AI is the medium the work happens through — it isn't the thing doing the work.

It's not 50× faster than a human. What 104 hours bought me on this particular project — and I'll get specific on the PROJECT in a future post — would conservatively take a solo developer 200–400 hours unaided. So the multiplier is real. It's also roughly 2–4×. Not nothing, but also, Not magic.

The honest framing is this: AI-augmented dev compresses the ramp-up cost on a thousand small decisions. Less time spent on mental syntax recall, framework-specific patterns, "where does this go again," boilerplate scaffolding. More time spent on judgment — what to build, what to ignore, what's good enough, what's not. The taste is still mine. The architecture calls are still mine. The pacing is still human.

Why I'm bothering to share this

Because the AI conversation right now is loud and lopsided. Hype on one side ("AI replaces developers in 18 months"). Pushback on the other ("it's a parlor trick"). Both feel wrong from where I'm sitting.

What I'm actually doing day-to-day is closer to the most productive pair-programming relationship of my career. A partner who has read every README. Never gets tired. Doesn't complain. Accepts "no, do it again" without ego. But also one I have to constantly steer, correct, and review.

The leverage is meaningful but the expertise absolutely still matters.

That's a useful frame to carry into 2026. Not "AI will do it." Not "AI is fake." Just a powerful tool that respects the operator who knows what they're doing.

And…

There's something at the end of these 104+ hours. Real software. Real users in mind. A real market (hopefully).

I'm not telling you what it is yet.

Be about it. Then talk about it.