Skip to content

Translation pending — English copy shown.

Agentic software factory

Agents execute. Engineers decide.

Most teams today hand agents a single step: writing code.

We hand them all the others, except the merge decision. Your engineers set the direction and decide what goes to production.

A developer working on her laptop

The software factory at a glance

Steps in the chain
8
Risk-based review levels
3
Decisions kept by a human
2

A complete chain, steered and controlled.

Agents take on every step of the development cycle, except the merge decision. A single platform steers them, follows sessions live and measures the flow.

Why now

Producing has never been so cheap

Drafting a memo, analysing a file, writing code: AI agents can now take most of these tasks on. A well-equipped person covers end to end what used to involve several people.

Value therefore shifts towards judgement: understanding the problem closely, choosing what is worth doing, keeping the whole coherent, making trade-offs.

Risk-based review

Human control follows risk

Every change is classified by what it touches. The more sensitive it is, the more senior the person who approves it, and for critical areas, no agent merges.

  • Experimentation

    Internal tools, prototypes, tests, documentation: anything easy to fix.

    Low riskAI review, checks green, the author approves.

  • Standard

    User-facing features that can be switched on and are reversible.

    Controlled riskA first AI pass, then a person approves.

  • Critical

    Payment, authentication, security, infrastructure, core business.

    High riskA senior lead approves. No agent merges.

The chain

Eight steps, one loop

Most teams only hand agents the writing of code. We break the whole cycle into steps and hand each of them to agents, except two decisions that stay human.

Supervision in turn feeds new signals: the loop is closed.

The software factory loop: eight steps in a circle, two of them human decisions, steps 3 and 7
  1. Signals Messages, tickets, alerts and usage data feed the chain.
  2. Specification A plan is drafted, then shared between agents and humans.
  3. Specification sign-off A person approves the intent and the approach. Human decision.
  4. Code generation Several agents work in parallel, in isolated environments.
  5. Automated review House rules are applied, backed by evidence and a risk level.
  6. Tests and QA Unit tests and journeys tested like a user.
  7. Merge and deployment A person decides on the merge; deployment is progressive, with automatic rollback. Human decision.
  8. Supervision Errors, performance and field feedback are tracked, and feed new signals.

Not “vibe coding”

Speed, with more engineering

This is not a prompt and a demo, but changes that go through the same requirements as in any team: security, tests, continuous integration, release process.

  • Your environments, your accounts
  • Nothing merged without a human
  • One automated review per change
  • Flow measured step by step

Maturity

Six levels, from assistant to software factory

Each team starts from its real level. The challenge is not the tool, but the move from one agent to several, steered and controlled. We agree with you a realistic target and the path to reach it.

  1. N0

    No AI

    Code is written without AI assistance.

  2. N1

    Autocomplete

    The tool completes the current line; the developer writes the rest.

  3. N2

    Copilot

    AI suggests, the human writes.

  4. N3

    One agent

    One agent completes a whole task.

  5. N4

    Parallel agents

    Several agents, in isolated environments.

  6. N5

    Software factory

    A complete orchestrated chain, reviewed according to risk.

Our support

Four steps, until your teams are autonomous

We start from your real level, without imposing a ready-made model. The goal: shorten the time between an idea and its release to production, without giving up anything on security or reliability.

  • Diagnostic

    Where your teams stand, where time is lost, what the bottlenecks, risks and strongest opportunities are.

  • Acculturation and training

    For executives, managers, business teams and tech teams, with NextGen Academy, a Qualiopi-certified training organisation.

  • Pilot lab

    One painful problem, tackled fast and assessed over a few weeks of real use, then a decision: keep or stop.

  • Industrialisation and handover

    A tooled chain, guardrails, indicators, then a handover to your teams so that they become autonomous.

Where do your tech teams stand?

Do you want to assess what an agentic software factory would change for your teams? Contact us for a first conversation: we start from your challenges and propose a tailored roadmap.