Scale · founder · 8 min read

Who Wrote This Line? Why Code Provenance Just Became a Founder Problem

The ex-GitHub CEO's new startup wants to store your agent's reasoning next to your code. Here's why provenance matters if you vibe-coded your product.

On July 8, Thomas Dohmke — who ran GitHub until last year — launched the preview of his new company, Entire. The headline feature is a distributed Git network: mirror your GitHub repo, and AI agents clone from a copy that’s geographically closer and doesn’t rate-limit them. Useful plumbing, not obviously your problem.

But buried in the announcement is a line worth reading twice. Dohmke said session logs are “now the second most important artifact in software development, and they belong in the repository alongside the code.”

That’s not plumbing. That’s a claim about what a codebase is now — and if you built your product by prompting, it’s a claim about you.

The gap nobody warned you about

When a human engineer writes a function, there’s a chain you can follow. A ticket, a pull request, a code review, maybe a Slack thread arguing about it. Six months later, when the function does something surprising, someone can reconstruct why it exists.

When an agent writes that function, the chain evaporates. You have the code. You have a commit. The conversation that produced it — the constraint you mentioned in passing, the three approaches the agent rejected, the assumption it quietly made about your data — lives in a chat window you closed. Maybe in a tool’s history tab. Maybe nowhere.

You end up with a codebase where every line is technically documented and nothing is actually explained.

For a weekend project, fine. For a company with customers, this is the thing that bites you at the worst possible moment: during an outage, during diligence, during a security review, or the first time you try to hand the codebase to someone who wasn’t there when it was built.

What Entire is actually selling

Strip the infrastructure story away and Entire’s pitch is provenance. Three features do the work:

  • Entire Blame — traces a line of code back to the agent conversation that produced it. Git blame tells you who and when. This tells you why.
  • Entire Review — automated code review that uses that context, so the reviewer knows what you were trying to do, not just what shipped.
  • Semantic search — query the history of changes and the reasoning behind them.

It integrates with Claude Code, Codex, Cursor, Factory, and GitHub Copilot. The company raised a $60M seed in February led by Felicis (Madrona, Microsoft’s M12, and Basis Set also in), which Felicis called the largest seed ever for a dev-tools startup.

The distributed-Git part is real too — Dohmke’s argument is that centralized hosting buckles when billions of agents hammer it, showing up as rate limits, latency, and outages. If you’ve ever had an agent stall mid-task on a git clone, you’ve felt the edge of that.

Why a non-technical founder should care

You are not going to set up a distributed Git mirror this afternoon. That’s not the point. The point is that a serious, well-funded team has looked at the vibe-coding boom and concluded the missing piece isn’t better code generation — it’s a record of intent.

Three places where the absence of that record costs you real money:

Diligence. We’ve written before about diligence teams cloning products to test defensibility. The other half of that conversation is provenance. An acquirer’s technical reviewer will ask what your code depends on, what data it touches, and where it came from. “Lovable wrote it” is not an answer. It’s a discount.

Incidents. When a vibe-coded app leaks data — and it keeps happening — the first question is always how long has it been like this, and what else did that change touch? Without the session that produced the change, you’re reverse-engineering your own product under time pressure.

Handover. The day you hire your first engineer, everything the agent knew and you didn’t becomes their problem. A codebase with recorded reasoning is a codebase a new person can join. One without is an archaeology dig, and they will tell you so in week two.

What to do before you buy anything

Entire is in preview and hasn’t announced pricing, so this isn’t a “go sign up” recommendation. But you can close most of the provenance gap with discipline rather than tooling, starting today.

Commit the prompt, not just the code

When you ship a meaningful change, put the actual instruction you gave the agent in the commit message or a /decisions folder in the repo. Two sentences is enough: what you asked for, and what you told it not to do. This is the single highest-leverage habit on this list and it costs you thirty seconds.

Keep a decision log

Not a changelog — a decision log. Why you chose Supabase over Firebase. Why auth works the way it does. Why that one table has a weird column. Agents make these calls constantly and never tell you they made a call at all.

Never let an agent touch auth, payments, or data access unreviewed

The provenance problem is worst exactly where the stakes are highest. If you can’t explain in plain English what a change to your login flow does, it doesn’t ship. Read our security guide if you want the longer version.

Keep your tool’s session history

Most agentic tools store conversation history. Most founders never export it. Whatever your tool offers — export it, and keep it somewhere that isn’t the tool. If you switch from Cursor to something else next year, that history is the only map of how your product got built.

The honest read

Entire might not win. Git hosting is a brutal market, GitHub is entrenched, and the company is currently mirroring repos rather than replacing anything. Betting the analysis on this specific company would be silly.

But the thesis is very hard to argue with. We now generate code faster than we generate understanding of it, and the gap compounds. Every tool in this space has spent two years optimizing the first half of that sentence. The interesting money is starting to move to the second.

If you’re a founder with a vibe-coded product in production, treat this launch as a prompt of a different kind: go look at your repo, pick a file you don’t recognize, and try to explain why it’s there. However that goes is your provenance score.

Related guides

Recommended next step

Was this helpful?