AI made the product cheap. Not the company.
24,933GitHub stars on Dolt, DoltHub’s main database
179on Argon, my project
On November 17 I opened a post on DoltHub’s blog and found my own project in the title: “Version Control Database Spotlight: Argon.”
It had been up for a week. A company I had never spoken to had installed Argon, loaded a million documents into it, and timed it against their own new database, DumboDB.
Both tools bring Git-style branches to MongoDB data: work on a branch, look at the diff, then merge it or throw it away. I built Argon so that AI agents could work that way. Someone else had looked at the same problem and reached the same answer.
For an example of where a database branch stops protecting the data, see the PocketOS deletion and its backups.
The first four rows
I read the tables the way anyone reads a review of their own work: looking for my name in the winning column.
It was there. Argon runs real MongoDB underneath, so ordinary reads and writes are fast. A hundred bulk updates: 25 ms for Argon, 43 ms for DumboDB. Three more rows went the same way.
The row that mattered
Then the table reached the thing both tools exist for. Open a branch of a million documents and read from it.
Argon, DoltHub's test47.2 s
Argon, my test8.5 s
DumboDB, DoltHub's test4 ms
Time from creating a branch to the first read. Argon 2.4.0 on my machine; the other two as DoltHub published them, on theirs. Different machines, so read the size of the gap and not the exact ratio.
Argon took 47.2 seconds. DumboDB took 4 milliseconds. Diff and merge went the same way.
I did what an engineer does. I ran it again on my own machine, ten times, on the current release. The median was 8.5 seconds. Better than their number, and still about two thousand times slower than theirs.
There was nothing to argue with. Argon copies the data for each branch. DumboDB’s branches share storage, which DoltHub got by rewriting MongoDB’s interface on its own engine. I checked the rest of their description row by row and it was accurate. Argon’s site says so, and the gap is now a tracked issue.
That part took a day. It is the part I know how to do.
Who was on the other side
Then I looked up who had written it.
DoltHub was founded in 2018. It has raised about $21 million, by CB Insights’ count. Public profiles of its chief executive say he was a senior engineering executive at Snap and worked at Amazon before that. Dolt, its main database, has about 25,000 stars on GitHub. Argon has 179.
DumboDB’s first release was on May 7. Since then their blog has shown a new piece of it every week or two: indexes, garbage collection, validators, clone, push and pull.
On my side there is one person with a day job. Working with AI coding agents, I shipped a command-line tool, an MCP server, a web console, a Python SDK, and packages on npm, PyPI and Homebrew. A few years ago that was more than one person’s output.
The part I had and did not use
Here is what stayed with me after the benchmark.
I had the same idea they had. I also had a view of how it would make money: host it as a managed service and charge a subscription, or license it to companies. I did not act on it. Argon stayed an open-source project.
They acted. They raised the money, hired the people, built the harder engine, and shipped it in public every week or two. Then they wrote a post about a smaller project that was generous, exact, and put their product next to mine in front of their readers. I have not learned to do that either.
What did not get cheaper
AI lowered the bar for building a technical product. That is how one engineer got as far as I did.
It did not lower the bar for the rest. Finding users, marketing, raising money: I am an engineer, and I have not learned these. None of them is certain the way a passing test is. I could answer their benchmark in a day because I knew what a right answer looked like. I do not know what one looks like for “who needs this, and how would they hear about it.”
That, more than the engine, is the difference between a project and a company.
So I am starting that part from the beginning. I do not know yet whether Argon should be a company at all. I do know that the next thing I have to learn is not in the code.
Sources
- DoltHub, “Version Control Database Spotlight: Argon”, November 10, 2026: the comparison and its benchmark tables.
- Argon’s response, November 17, 2026: their table checked row by row, and Argon’s own measurements. The 8.5 seconds is the median of ten runs of Argon 2.4.0.
- DoltHub’s funding: CB Insights. Other trackers give between $16 million and $23 million. Its founding year and its chief executive’s time at Snap: Dealroom. His time at Amazon: The Org. I did not find these in a statement from DoltHub itself.
- Star counts were read from the GitHub API on November 17, 2026. Stars show interest. They are not users or revenue.

