WebNoviAI training and implementation

Adopting agent-based development in a team

In agent-based development a team gives tasks to AI agents (Claude Code, Codex), and the working rules live in files such as AGENTS.md and a project manifest. I help set up the process so that code is checked by tests and nothing breaks quietly.

Author: Andrey Novikov. Updated Nov 21, 2026.

Who needs it and what hurts

Engineering teams and founders who want to ship faster but fear losing control of quality.

What I will do

  1. We read your code, your team and your current process.
  2. I write the rules and connect automatic checks.
  3. We run the first tasks together and adjust the process by the results.

What you get

  • A rules file for agents: stack, bans, the order of checks, the report format.
  • A project manifest as the source of truth, with a check that the code matches it.
  • A release order: a staging environment first, then production, with rollback.
  • A report template for an agent after each task: what was done, what was checked, what was not.

What I need from you

  • Access to the repository and the build process.
  • A person on the team who accepts the agents' work.
  • A decision on what agents must never do.
  • A list of commands and environments the agents may and may not touch.

Questions and answers

Does it work on a real project?

WebNovi itself is built this way. The details, without invented figures, are in the case study.

Does it replace developers?

No. People keep the decisions and the acceptance, and agents speed up the execution.

What about secrets?

Agents never see secrets. They sit in protected storage and do not end up in code or reports.

Related