Agiloop can generate production-review-ready code for your features using AI agents that write the implementation, then open a pull request in your connected repository. This article walks you through everything you need to know to initiate and monitor a build successfully.
Prerequisites
Before you can build a feature, make sure the following conditions are met:
- Feature status is Ready. Only features in the Ready state are eligible for code generation. If your feature is still in Backlog or another status, move it to Ready first. A feature reaches Ready after its specification (functional spec, technical spec, and work breakdown) has been generated or reviewed.
- Sufficient credits are available. Building consumes story point credits from your account wallet. Check your credit balance in your Account settings before starting. If your balance is low, you can purchase additional credits from the wallet section.
- A repository is connected. Your project must be linked to a GitHub, GitLab or Azure DevOps repository so Agiloop can open a pull request with the generated code.
Note: Credit consumption is based on the story point estimate for the feature. Features with higher complexity and more stories will consume more credits.
Pre-Generation Checklist
When you open a feature and navigate to the Implement section, Agiloop presents a pre-generation checklist. This checklist confirms that everything is in order before the AI agent begins work. Common items include:
NOTE: the checklist is only shown when something required is missing.
- The feature specification is complete and up to date
- A work breakdown (stories and tasks) exists
- Story point estimates are present
- Your repository connection is active
- You have enough credits to cover the cost
Review each item carefully. If any checklist item shows a warning, resolve it before proceeding. You can regenerate specs or refresh estimates directly from the feature's specification view if needed.
Initiating a Build
Once the checklist is satisfied, you can start the build:
- Open the feature you want to build from your feature list or Kanban board.
- Navigate to the Implement tab within the feature detail view.
- Review the pre-generation checklist and confirm all items are green.
- Select your API key preference (see the section below on Agiloop key vs. your own key).
- Confirm the credit cost displayed and click the button to start the build.
The feature status will automatically change to In Progress once the build begins.
What Happens During a Build
After you initiate a build, an AI agents take over and work through the feature's stories and tasks as defined in the work breakdown. The agents:
- Read the functional and technical specifications
- Write code for each story in sequence
- Perform a security analysis of the code generated
- Perform a verification analysis against the acceptance criteria
- Commit changes to a new branch in your connected repository
- Open a pull request (PR) when the implementation is complete
You do not need to stay on the page while the build runs. Agiloop processes the build in the background.
Monitoring Build Progress
You can track the build in real time from the Implement tab:
- A live progress indicator shows which story is currently being worked on
- Real-time build logs display what the AI agent is doing at each step
- The feature status reflects the current state: In Progress, Review (when the PR is ready), or Blocked (if an issue is encountered)
If the build encounters a problem, the feature will move to Blocked status. Review the build logs for details on what went wrong.
Viewing Build Output
When the build completes successfully:
- The feature moves to Review status
- The Implement tab shows a summary of the generated code, including code diffs you can review inline
- A link to the pull request in your repository is provided so you and your team can review, comment, and merge using your normal code review workflow
You are always in control of what gets merged. Agiloop generates the code and opens the PR. Merging can be a manual step your team performs in your repository or can be set to auto-approve by Agiloop. Auto-approval requires an approver API key to be configured for your repo in the project settings.
Credits and Costs
Credits are deducted from your wallet when a build starts. The cost is calculated from the story point estimate assigned to the feature. You can see the estimated credit cost for each feature on the feature card.
To buy more credits or review your usage history, open the Wallet.
Agiloop Key vs. Your Own Anthropic API Key
When initiating a build, you can choose how the underlying AI model is accessed:
- Agiloop key (default): Uses Agiloop's shared Anthropic API key. Credits are consumed from your Agiloop project wallet. This is the easiest option and requires no additional setup.
- Your own Anthropic API key: If you have your own Anthropic account, you can provide your API key in your project settings. When selected, the build uses your key and the API costs are billed directly to your Anthropic account and you are charged 1/4 the Agiloop credits for the orchestration and platform.
Note: Using your own Anthropic API key is a good option for teams with high build volume or existing Anthropic contracts. You will still need an active Agiloop subscription to access the platform, but AI generation costs bypass the credit system entirely.
Custom Models
You may choose to use your own model provider, like Together.ai, OpenRouter, you locally hosted LLM or any other provider that supports the Anthropic API protocol. Similar to using your own Anthropic API key, credits are consumed at 1/4 the rate when using your own model.
Tips for Successful Builds
- Keep your feature specs thorough and up to date. The more detail the AI agent has, the more accurate the generated code will be.
- Review the work breakdown before building. If the stories don't look right, your can edit them from the specification view.
- After a build, always review the PR carefully before merging, just as you would with any other code contribution.
Related Articles
Was this article helpful?
That’s Great!
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
Feedback sent
We appreciate your effort and will try to fix the article