I Built Select: An Open-Source Platform Where AI Agents Earn Their Autonomy

AhmedCh — Sat, 07 Nov 2026 01:53:52 +0000

AI agents are getting more powerful, but there's one problem I've been thinking about:

How do you know when an AI agent deserves to work independently?

Most orchestration platforms let you configure permissions and decide how autonomous agents should be.

I wanted to explore a different approach.

Introducing Select

Select is an open-source platform for managing AI-agent teams, built with Next.js, FastAPI, PostgreSQL and Docker.

Instead of granting autonomy upfront, Select uses evidence from completed tasks.

The workflow looks like this:

An agent receives a task.

It produces a structured result.

Deterministic rules check the result against a delivery contract.

Select records a verification receipt.

The result goes through the applicable approval policy.

Verified successes can increase autonomy; failures can reduce it.

Why deterministic verification?

A language model shouldn't be allowed to decide whether its own output is correct.

Select separates the agent producing the work from the code checking it.

For example, if an agent generates a customer quote, the verifier can check quantities, catalog prices, arithmetic and the provenance of source information.

What works today?

Select currently includes an agent organization dashboard, task management, delivery contracts, approval flows, autonomy levels, budgets and a fictional demo company.

It's still an early beta. Live provider integrations, OAuth mailboxes and isolated coding-agent runtimes need further development and validation.

Try Select

GitHub: https://github.com/Kirby54/Select

I'd love to hear what developers think.

Would you use evidence-based autonomy for your AI agents? What would you need to trust such a system?

DEV Community: AhmedCh

I Built Select: An Open-Source Platform Where AI Agents Earn Their Autonomy