Repository-aware execution
Use the selected codebase, its detected commands, prior run context, and the initiative’s acceptance criteria.
AI implementation agents
Move approved software work from plan to tested implementation.
The Block’s AI execution system is built for repository-backed work that needs more than a code snippet. It evaluates readiness, uses connected repository context, runs implementation in controlled environments, tests the result, and keeps approval and deployment decisions with people.
Start an initiative · Create an account
What you gain
Testing and adversarial review
Run relevant checks and route the implementation through a separate review model before presenting the result.
Controlled access
Use short-lived credentials, isolated runs, explicit capabilities, redacted secrets, and human approval for higher-risk changes.
How it works
Prepare an AI-ready task
Define the desired outcome, acceptance criteria, repository, and any required context or constraints.
Confirm readiness
The Block checks repository access, credentials, commands, scope clarity, and risk before launch.
Implement and test
An agent works in an isolated environment, runs the relevant checks, and records the execution result.
Review before release
Inspect the changes, test evidence, preview, and review findings before deciding what happens next.
Frequently asked questions
Can an AI agent deploy directly to production?
The execution path is designed around explicit capabilities and human approval. Higher-risk work and production-impacting actions require deliberate authorization.
How does the agent understand an existing repository?
The Block builds a repository profile from approved source access, manifests, commands, workflows, and prior successful runs, then combines it with the task context.
Are secrets shown to the model?
Sensitive client values are handled through controlled, audited injection paths and are not included as plaintext in model-facing responses.