A clear goal and context
Define the objective, relevant context, and what completion means.
Agentic Development Workflow
A way of building software in which AI carries out defined tasks while people guide the work and review the result.
In plain language
In an agentic workflow, AI plans steps, uses approved tools, and checks its work. People set boundaries and approve the result.
The right level of autonomy depends on the task, risk, and how easily changes can be reversed.
The foundation
What matters in Agentic workflow
Tools with guardrails
Permissions and checks limit changes and reveal failures.
Human authority
People control sensitive actions, product choices, communication, and releases.
Applied by The Forge
How we put it into practice
AI handles defined tasks. People guide, review, and remain responsible for the work.
Frame a bounded task
Define a concrete outcome, constraints, and acceptance criteria.
Provide the working context
Share project conventions, design rules, and approved tools.
Work in reviewable increments
Keep changes small, progress visible, and feedback timely.
Verify before release
Combine automated checks with human review of quality, accessibility, and purpose.
A useful boundary
Limits to keep in mind
- It is not a chatbot producing unchecked code from a single prompt.
- It does not remove accountability from the people directing and approving the work.
- It is not equally appropriate for every high-risk, irreversible, or poorly defined action.
A concrete example
Building a new website section
An agent reads existing components, edits a website section, and runs checks. A person reviews the change before release.
Further reading
Primary sources
Check the sources behind these explanations.
Related concept
Maintainability: Maintainable Software
Software that people can understand, repair, and adapt as their needs change.