Skip to the content
Majid Khan Mohmand

StoresWiki, AI agents & automation

Tennet.io

AI-powered collaborative development teams

Designed and built in-house

Tennet.io: its home pageTennet.io on a phone

Measured, before and after

Automated tests across Tennet.io’s services
Before
703 tests
After
5,500 tests

Linear scale

What it is

AI-powered collaborative development teams: AI agents plan and build apps together, designed for people to steer them mid-run. Node and TypeScript, React, Python (CrewAI, LiteLLM) and an HMAC-signed event bus. Its 703 passing tests mocked the very components they tested; the agent loop was restored and acceptance moved to real-browser journeys — about 5,500 automated tests across five services now.

On the record

  • Its 703 passing tests had replaced the LLM client, the event bus and the agent framework with mocks — which hid four swallowed exceptions that disconnected the whole agent loop.
  • With the loop restored, one prompt yields a plan and an agent team in about 42 seconds, and the agents’ code streams live.
  • Acceptance moved to real-browser user journeys; the codebase now runs about 5,500 automated tests across five services.