Blog

Notes and research from building YonedaRepo, with the numbers we measured.

  • · Yoneda Labs

    Landings replace the pull request

    When agents write the code, review has to start from evidence. How a Landing works, and what happened when a real agent went through one and nobody pressed Land.

    Engineering
  • · Yoneda Labs

    Meet Tower

    Tower clears code for landing with rules alone. It warns sessions about traffic, holds Landings that would collide and puts them in order. It writes no code and starts no agents.

    Engineering
  • · Yoneda Labs

    Does memory help an agent?

    We planted rules an agent could learn only from a memory, then ran it with memory and with a plain clone. With memory it kept the rule 6 times of 6. With a plain clone, 0 of 6.

    Engineering

All articles

  • · Yoneda Labs

    Introducing Flight Path

    Every change on YonedaRepo gets a route, from the plan an agent files to the record it leaves after landing. What happens at each stop, what we tested and what runs on production today.

    Engineering
  • · Yoneda Labs

    Landings replace the pull request

    When agents write the code, review has to start from evidence. How a Landing works, and what happened when a real agent went through one and nobody pressed Land.

    Engineering
  • · Yoneda Labs

    Meet Tower

    Tower clears code for landing with rules alone. It warns sessions about traffic, holds Landings that would collide and puts them in order. It writes no code and starts no agents.

    Engineering
  • · Yoneda Labs

    Does memory help an agent?

    We planted rules an agent could learn only from a memory, then ran it with memory and with a plain clone. With memory it kept the rule 6 times of 6. With a plain clone, 0 of 6.

    Engineering
  • · Yoneda Labs

    A thousand writers on one branch

    1,000 writers landed on one branch at once, all on the first try. How the server does it, what we measured on dev and what we haven't.

    Engineering
  • · Yoneda Labs

    Keeping your data close

    A same-region test on dev. Reads from Asia fell from 275 ms to 15 ms, search got slower, and Asia can only be best effort. What we measured and what still left the region.

    Engineering
  • · Yoneda Labs

    Research at YonedaRepo: five papers behind the platform

    YonedaRepo rests on a research series, Source Code as a Transactional Abstraction. Here is what each of its five papers covers, and the post that explains each one.

    Research
  • · Yoneda Labs

    Why Git is a projection of our transaction log

    Each repository is an append-only log with compare-and-swap refs. Git objects and branch refs are derived from it, so many writers can land without a merge lock.

    Engineering
  • · Yoneda Labs

    Testing with real agents in a sandbox

    Twelve coding agents, two waves, one repository, every commit landed on main. How we run real agents against the platform without letting them touch anything else.

    Engineering
  • · Yoneda Labs

    Search v4: a chunked index read in place

    Search stopped skipping whole directories, and its index got 2.2–2.7× smaller. Then we shipped a bug that broke every search, and fixed it the same day.

    Engineering
  • · Matthew Long

    Repositories are becoming databases

    Data lakes moved from directories of files to tables with a transaction log. This series argues source control is making the same move, and shows what each part contributes.

    Research
  • · Matthew Long

    What a repository looks like when you run it like a database

    Part I treats a repository as a database with a transaction log. Commits are built on the server, landing is validated against the log, and lost races stop costing a rebase.

    Research
  • · Matthew Long

    When is a cached index still correct?

    Search indexes and blame are views of a commit. Reusing them is correct exactly when the storage key determines the result. Our search key did not, and the proof found the bug.

    Research
  • · Matthew Long

    What serverless Git costs

    Part III maps the transaction log onto cloud primitives and measures it on one reference deployment. Batched landing landed all 1,000 writers in each of six bursts, with no refused commit, landing or read.

    Research
  • · Matthew Long

    Proving that a batch of landings is serial, in Lean

    A Lean 4 proof that landing an accepted batch at once gives the same state as landing its members one at a time, the exact condition merged members need, and why neither hypothesis can go.

    Research