# AI agents from feature request to release: where people still decide

A product manager describes a feature in plain words. Agents write the stories, tests and code; people answer the questions, review and sign off.

- Author: Laszlo Csiki (https://www.linkedin.com/in/laszlo-csiki-7ba454108/)
- Published: 2026-03-21
- Kind: Live demo
- Demo: https://x2labs.ai/demos/agentic-delivery-workflow?embed=1 (Live demo · click through, or use the arrow keys)
- Page: https://x2labs.ai/notes/agentic-delivery-workflow

Most of the time between a feature request and a release isn’t spent deciding anything. It goes on turning a request into stories, stories into tasks, tasks into code and tests, and then checking that what came out matches what was asked. We wanted to know how much of that agents can carry, and where a person still has to decide.

## How it works

A product manager describes the feature in plain words, the way they’d explain it to a colleague. The demo above follows one real example: an analytics dashboard for a SaaS product, with metrics, charts and exports.

- **Questions first.** Before any code, an agent asks what a good lead would ask: which metrics, which time ranges, real-time or refreshed. Gaps show up here instead of in QA.
- **Stories with acceptance criteria.** The answers become structured stories, ready for sprint planning in your tracker.
- **Parallel work.** The feature is split into tasks and handed to several agents at once. Each one writes failing tests first, then implements until they pass.
- **People review.** A developer reviews the code, the way they would a colleague’s pull request.
- **Checked against the criteria.** Nothing is marked done until it passes every acceptance criterion from the stories. Then QA signs off.

## What we learned

Agents are fast and dependable when each one has a narrow task and a failing test to make pass. Give one a wide, vague task and it drifts. So most of the work is in the split: small tasks, clear criteria, tests that fail for the right reason.

People stay where judgement matters: the questions up front, code review and the QA sign-off. Those are also the steps your team would least want to hand over.

## What it means for your team

Your team keeps its tools and its reviews. The work between the decisions gets faster: stories, tests, the first implementation and the checks. The practical first step is one feature on your own codebase, measured against how long it usually takes.