Skip to content

AI development with control over scope, context, and quality.

Generate an implementation plan, create a branch, modify code under tool allow-lists, run tests, open a pull request, summarise the diff, and request review — without losing engineering discipline.

Watch one change go from a reviewed plan to a pull request with its reviewers tagged.

Capability · Development
  1. Plan: Implementation plan · Step-by-step plan reviewed before code edits begin.
  2. Branch: Branch creation · Feature branch scoped to the mission, naming-policy enforced.
  3. Code: Code modification · File edits inside the mission scope only.
  4. Test: Test execution · Suite runs before PR is opened. Failures block.
  5. PR: Pull request · PR opened with diff, rationale, risks, and rollback note. Diff summary · Reviewer-friendly summary auto-attached to PR body.
  6. Review: Review request · Required reviewers tagged per CODEOWNERS + policy.

capability runtime · streaming

Uptime
99.98%
Latency
14.2ms
Nodes
128/128

Evidence

412 / 418 evidence packs signed today

release readiness · evidence-backed

Delivery

3.4× velocity · 22% burn

  • playwright
  • bdd
  • api
  • a11y
  • visual

What development agents do.

Read it as the history of one pull request: every entry is a job a development agent does before a person reviews the change.

Pull request · one mission Sample
  1. Implementation plan

    Step-by-step plan reviewed before code edits begin.

  2. Branch creation

    Feature branch scoped to the mission, naming-policy enforced.

  3. Code modification

    File edits inside the mission scope only.

  4. Test execution

    Suite runs before PR is opened; failures block.

  5. Pull request

    PR opened with diff, rationale, risks, and rollback note.

  6. Diff summary

    Reviewer-friendly summary auto-attached to PR body.

  7. Review request

    Required reviewers tagged per CODEOWNERS + policy.

Agents pass work like a relay.

The box is one change. Each agent does its part and hands it on. Nothing moves until an operator approves the plan: press Approve plan to start the run.

  1. Planning agent

    Produces implementation plan; operator approves.

    Plan ready · waiting for the operator
  2. Coding agent

    Edits files inside scope; runs unit tests locally.

  3. Review agent

    Static analysis, style, regression risk, security flag.

  4. QA agent

    Generates and runs integration + Playwright tests.

Boundaries the coding agent cannot cross.

Pick a boundary to see where it sits between the agent's edits and a merged change.

  1. Agent can only touch files declared in the mission plan.

  2. Protected paths require named reviewer beyond the agent.

  3. Suite must pass before PR opens. No PR on red.

  4. Hardware-token gate on protected branches.

  5. Every PR ships with an automated rollback playbook.

Start a Software Factory pilot.

  1. One repository
  2. One ticketing tool
  3. One mission
  4. Evidence

Connect one repository, one ticketing tool, and one mission. See evidence in days, not quarters.