Checksum has launched its Steady High quality Agent, an autonomous system that runs nightly in opposition to deployed functions and robotically heals damaged assessments with out ready for an engineer to open a dashboard or write a immediate.
AI coding has modified the constraint in software program growth. Groups can now ship much more code than earlier than, however each PR nonetheless must be examined, validated, and trusted earlier than it reaches manufacturing. Even assessments written by AI require human upkeep to replace selectors, triage failures, and separate actual bugs from damaged assessments. The hole widens each dash, and shutting it requires an agent that generates assessments robotically and heals them because the product adjustments.
“The business solved code era. It hasn’t solved high quality,” stated Gal Vered, CEO of Checksum. “Groups utilizing AI to put in writing code are delivery extra and catching much less. With the Steady High quality Agent, 70% of take a look at failures resolve with out an engineer touching them. That’s the distinction between AI as a copilot and AI as infrastructure.”
The agent works the total high quality loop, fine-tuned on greater than 1.5 million take a look at runs. It detects gaps in protection, generates assessments for particular flows, runs them in opposition to dwell functions, and heals damaged assessments autonomously. Each take a look at is commonplace Playwright code, dedicated on to the group’s personal repo as a pull request. No proprietary format, no lock-in.
The agent meets builders the place they already work:
- From the net app, groups get full visibility into each session, each failure classification, and a dwell Characteristic Well being Dashboard that separates actual product bugs from damaged assessments.
- From the IDE, /checksum slash instructions in Claude Code, Cursor, and extra let builders set off, steer, and overview the agent with out leaving their editor.
“For lower than half the wage price of an offshore developer, I’ve the affect of a full QA group,” stated Ron Alexssen, Engineering Supervisor at Counterpart. “If I had been making an attempt to interchange what Checksum is doing, it could take me at the least a full group of six to 10 folks. Once I present as much as company-wide conferences and report zero manufacturing outages, I really feel like a hero.”