Know what “done” means
Frame the goal, scope, and acceptance conditions before implementation. Judge the result against an agreed contract.
AGENTFLOW / An opinionated AI-DLC for regulated industries
Give AI room to deliver.
Keep authority clear.
Move from a business goal to a reviewable change, with explicit ownership, evidence at handoffs, and human decisions where the consequences demand them.
AI-DLC: an AI-assisted development lifecycle. Open source, built around your existing repository.
One goal. Coordinated work.
See autonomy with boundaries.
Read the video description
The human defines the goal and its limits. The Goal lead coordinates three bounded tasks: implementation, test analysis, and edge-case analysis. The agents work concurrently and return separate evidence. A distinct reviewer challenges the combined result. Outside execution, the Decider advises the Steerer, which acts through a steering interface. Evidence feeds back to supervision. The human retains authorization.
On-screen tagline: An opinionated AI-DLC for regulated industries. Soft instrumental music; no spoken narration.
The value proposition
Make the decision to proceed easier to understand.
Keep accountability visible
One lead owns delivery. Parallel helpers contribute bounded work; they do not inherit approval authority.
Carry evidence forward
Keep checks, findings, handoffs, and decisions with the work, so review and resumption do not depend on a chat history.
Intentional design
More autonomy needs better boundaries.
Scale assurance with consequence.
Use a lean path for bounded changes. Add stronger challenge and human review for sensitive work; task size alone does not determine risk.
Separate doing from judging.
Independent review has its own boundary. When only self-review is available, disclose it rather than presenting it as independent assurance.
Steer without taking over.
Keep supervision outside execution. The Decider advises on triage; it cannot sign off gates, change scope, or authorize a merge.
Improve through authorization.
Turn delivery friction into refinement proposals. Review and authorize changes before applying them to the operating model.
Where it fits
Start with a real delivery decision.
A routine product change
A bounded change needs clear scope and reproducible checks, without assembling a large agent team for every task.
A sensitive change
An authorization rule or data migration needs explicit acceptance criteria, stronger review, and a human decision before proceeding.
Work that stops making progress
Repeated rework calls for diagnosis and authorized recovery, with the evidence preserved for the next decision.
For regulated environments
Support your assurance process.
Traceability, separation of duties, and explicit acceptance are design concerns from the beginning.
AgentFlow provides a structure for these controls. Your organization defines its obligations, validates its use, and retains responsibility for release and compliance decisions.
Begin with one repository and a bounded goal. Evaluate the quality of the decision record before expanding autonomy.
Plan a first evaluation →Open by design
Inspect the choices. Adapt the implementation.
The product story lives here. The current installation steps, supported tools, contracts, and implementation live in the open repository.
Built by Samuel Mota · Move the Needle
Why AgentFlow SDLC exists
Open source · Apache-2.0
Value
Turn a goal into a reviewable change, with clear ownership, evidence at handoffs, and explicit human authority.
Strategy
As AI takes on more delivery work, the decisions around that work need to stay understandable. AgentFlow makes scope, review, and acceptance part of the workflow.
Technical frontier
Proportionate assurance, bounded parallel contributions, external steering, and evidence-backed refinement proposals keep execution and decision authority distinct.