AI can now build an entire application in hours.
But that created a new problem:
Who verifies the software that AI writes?
We can generate code.
Generate tests.
Review PRs.
Ship faster than ever.
But when a real user clicks the wrong button, an API slows down, authentication breaks, or something gets exploited — passing tests don't always mean the product works.
That's what we worked on at Bharat Builds by WeMakeDevs, powered by AWS.
We built VERIFAI — an Autonomous Software Verification Lab.
You connect a GitHub repository.
VERIFAI creates an independent AI verification team that can:
→ use the product like real users
→ test APIs
→ look for security issues
→ inject failures
→ stress-test performance
→ reproduce findings
→ collect evidence
→ verify a fix before proposing a PR
The idea is simple:
AI can propose what might be wrong.
Executed evidence decides what is actually wrong.
We built it using Strands Agents, Amazon Bedrock AgentCore, CodeBuild, ECR, ECS/Fargate and open-source tools including Strix, Cua, MiroFish, Browser Use, ZAP, Schemathesis, Toxiproxy, Locust and k6.
A lot broke while building this.
That was probably the best part.
Huge thanks to Kunal Kushwaha, Hitesh Choudhary, WeMakeDevs and Amazon Web Services (AWS) for creating a space where builders can actually experiment and ship.
AI gave developers a coding team.
We want VERIFAI to give that team an independent QA, security and user-research department.
#BharatBuilds #BuildersBharatBuilds #WeMakeDevs #AWS #AI #AIAgents #AgenticAI #OpenSource #Hackathon
Loading article