Cosine
Autonomous AI software engineer — picks up tickets and ships pull requests like a teammate.
About
Cosine is a UK-based startup building Genie, an AI software engineer designed to work end-to-end on real codebases. It plugs into a team's existing repos and issue tracker, picks up tickets, reasons over the full codebase and its history, and submits pull requests for review — operating either autonomously or collaboratively, with Slack and notification integrations so it behaves like a colleague.
Genie made its name with a 30.08% score on SWE-Bench back in 2024 — a state-of-the-art result at the time, far ahead of GPT-4's ~12% — achieved by training on data that codifies human engineering reasoning rather than just code. The agentic loop covers planning, retrieval, code writing, and code running, with self-correction learned from synthetic failure data. The company raised a $2.5M seed round led by SOMA and Uphonest Capital.
Worth knowing: access has historically been limited/invitation-only with custom enterprise pricing — this is not a self-serve developer tool. The headline benchmark is from 2024 and the field has moved fast since; evaluate current Genie 2 claims (like SWE-Lancer results) against your own tasks before committing.
What it can do
- Ticket-to-PR autonomous engineering workflow
- Repo-scale context over full codebase and history
- Self-correcting agentic loop: plan, retrieve, write, run
- Slack and notification integrations
- Human-in-the-loop review of generated PRs
Setup
Contact Cosine for access — invitation/enterprise onboarding via cosine.sh. Connects to GitHub repos and issue trackers.
Prerequisites
A GitHub organization; an issue tracker (GitHub Issues, Jira, or Linear).
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 45; Trending score 70/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.