Composal, a London-based startup in Y Combinator's Fall 2026 batch, sells swarms of browser agents that use software the way customers would and report bugs back to coding agents. Pricing starts at $500 for 3,000 browser minutes, according to the company's documentation.
The pitch rests on a simple imbalance: AI coding agents can ship features faster than human teams can check them. Composal wants to own the checking.
"In conversations with customers and other engineering teams, we keep coming back to the same question: when an agent says a feature is done, how do you know it works?" founder Pandelis Zembashis wrote in an Oct. 23 post on the company's Reflog blog.
How the product works
Composal's homepage says it "sends a swarm of agents to use your product like your users do." The tagline is blunter: "Your agents build it. Ours break it."
The product, called Verify, runs browser agents against a managed shadow environment or any HTTP(S) URL, per the docs. The agents operate in a sandboxed browser and, the docs say, do not edit source code. Each finding comes back with video, logs and steps to reproduce.
The company's pages divide Verify into four parts:
- Discovery reads specs and a connected code repository, then builds a persistent map of screens and user flows.
- Deep Scan sends a focused swarm to pressure-test a feature or reproduce a specific bug.
- Scenarios turn user journeys into repeatable YAML tests. Verify selects relevant scenarios for each pull request diff, or runs full packs on a schedule.
- CLI lets teams run QA from inside a coding agent.
Checking the checker
An automated tester that cries wolf is worse than none, and Composal appears aware of it. Its verification docs say a separate model re-checks Deep Scan findings, and that the most serious bugs, labeled P0 and P1, are re-run in a fresh browser before being confirmed. Runs are pinned to an exact environment revision and scenario pack, with video, screenshots, console output and network data captured.
The company also argues that the agent writing code should not be the one grading it. "Seven teams described keeping the writing agent away from verifying its own code," Zembashis wrote in the Oct. 23 post.
In an Oct. 6 post, he laid out what he calls Continual Agent Verification, or CAV. "Agents use the running product, pursue user goals, and bring back evidence of what works and what breaks," he wrote. The post frames continuous integration as a test of code correctness, continuous deployment as a test of deployment correctness, and CAV as a test of product correctness.
Integrations and pricing
According to its docs, Composal supports pull request testing on GitHub and GitLab, Slack reporting, and an MCP connector (a standard that lets AI tools talk to outside software) for Claude, Codex and Cursor. Private apps can be tested through AWS Client VPN, added Oct. 13, and through Tailscale. Scheduled post-merge runs, with cron support and Slack reports, were documented Oct. 26.
Pull request tests can be triggered by pushes, ready-for-review events, labels or comment commands such as "/qa" and "/verify." Version 1.17.92 of the CLI, released Oct. 21, added the ability to set up PR testing from the terminal.
New Team workspaces are invoiced $500 upfront for 3,000 browser minutes, and the subscription renews monthly, the docs say. Composal has no separate pricing page, and the docs show no effective date, so the terms could change.
The team
Zembashis is CEO and a former CTO of Cosine (YC W23), according to the company's About page. Jonathan Davies, the CTO, co-founded mmtm and was a product engineer at Papaya. Jan Moehlmann is listed as a founder and previously held commercial roles at 11x, Encord and Tractable. Y Combinator's directory lists a team of three.
The company's legal operator is L6 Software LTD. Composal has not announced a funding round, and any funding beyond YC's standard terms is not publicly disclosed. A Dealroom entry listing a $125,000 seed round is auto-generated and unverified, and is not treated here as a round.
A crowded field
Composal is not alone. Meticulous renamed its Agent-Driven Test Harness "Agent Swarm Testing" on Oct. 1, according to its changelog, and announced a $15 million Series A led by Chemistry on Oct. 6. Meticulous's approach analyzes code diffs and replays sessions, while Composal's agents work from user goals without access to code.
Virtuoso made its Touchstone agentic QA platform generally available Sept. 30. BrowserStack launched Test Companion, an IDE-based agent, on July 29. Sauce Labs introduced AI for Test Authoring on March 17.
What comes next
Y Combinator's directory lists Composal as an active F26 company. The Demo Day page shows a date of Wednesday, Dec. 2, though the year is implied by page context rather than stated.
Zembashis, for his part, is betting on accumulation. "I think the most valuable part of an agent swarm is the history it can accumulate," he wrote on the Reflog.
