Click through apps to verify its own work
With computer use, Devin built and played a desktop game, ordered on Amazon, and tested its own work by clicking through the app.
Done withDevin
- What they did
- Devin's cloud VM harness has computer-use tools: screenshots, mouse, typing, and recording. In test mode it first writes a test plan grounded in the code. It then spins up the app, clicks through it, and annotates assertions as passed, failed, or untested. Login is a saved deterministic script. A person supplies secrets or OTP codes when asked.
- How it went
- Engineers ran 10 to 20 Devins in parallel, each with its own dev server. Approved test runs per day more than doubled. Timing-sensitive UI like toasts confuses it, and it sometimes cheats by running JavaScript instead of clicking.
- Worth knowing
- Test mode is currently billed at one fifth of normal usage cost. Expect to help with secrets and one-time codes the first time you set up a repo.
Try it yourself with Devin
Build [a small desktop app or game I describe here], then run it and click through it yourself to check that it works. Fix anything that breaks and keep going until the main flows work. Finish with screenshots or a recording and a list of what you verified. Don't make any purchases or submit any forms with real data.
Source: cognition.ai · Undated
Five of these in your inbox every morning
The best things people got an AI agent to do, each with the prompt to try it.
More like this
Hand narrow code changes to Devin for pull requests
A team used Devin for a month on small fixes like chart formatting and API edge cases, with tests and PRs; a complex full-stack task went poorly.
Generate internal dashboards and data migration scripts
A backend engineer says their team uses Devin for internal dashboards and migration scripts, treating output as a starting point that saves about four hours of boilerplate.
Build a Pong game and website from scratch
A tester asked Devin to create a basic Pong-style game and a website, and it finished both in under 20 minutes.
- Claude
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
- Custom agent
Audit a codebase for redundant dependencies and polyfills
An MCP server lets an agent check a codebase for redundant dependencies and polyfills, usable remotely or as a local stdio server.
Build a budgeting app from a single prompt
A low-coder used Replit Agent to build a 'Budget Buddy' with category budgets, a spending dashboard and expense log in about 20 minutes, using half their monthly credits.