Verification Agents
It’s increasingly clear to me that becoming the best coding assistant will require being the best at verifying code changes. I predict verification agents—AI agents capable of autonomously ensuring your code fulfills requirements—are a huge, unexplored opportunity with AI, and I don’t think enough people are working on this today.
Imagine having agents for each of the following tools in your stack today:
- Unit tests
- Integration tests
- Automated UI tests
- E2E tests
- Synthetic data creation
- Observability tools
- …and many others.
Each of these agents could directly consume requirements, deeply understand your codebase, and automatically establish guardrails for your code changes, giving you supreme liberty to rapidly iterate through code variations via prompting until you arrive at a version that passes verification.
Instead of writing code and slogging through all the verification steps, you’d have immediate feedback about correctness, drastically reducing manual QA and increasing deployment velocity with unprecedented levels of confidence.
We already have agents writing code. But the real leap forward will come when we have agents verifying it—continuously, intelligently, and collaboratively.
