Vaiz
Sign up
Help center
Tutorials

How to delegate a task to an AI coding agent in Vaiz

Coding sessions let you run AI coding workflows directly from Vaiz. Delegate a task to Cursor or Claude and review the proposed code changes. Decisions are saved in the task's Memory field, with links to the corresponding blocks.

Coding sessions and Memory are included in the Premium plan. You will need access to your Vaiz space, the GitHub repository, and your chosen agent's API.

Step 1. Connect GitHub in App center

In your Vaiz space, open App center → GitHub and click Connect. Follow the setup flow to select a GitHub account or organization, grant access to the project repository, and complete the connection. Return to Vaiz and check that the status is Connected.

See the GitHub setup guide for detailed instructions and screenshots.

Grant the integration access to the repositories needed for the work. When connecting your agent, check its access to the same project separately.

Vaiz App Center showing GitHub connected for linking pull requests, branches, and commits to tasks.

Step 2. Connect Cursor or Claude through the API

Create an API key for your chosen agent:

Copy the key and return to Vaiz. In the agent connection settings, select Cursor or Claude, paste the corresponding API key, and save the connection. Check that the agent can access the repository it will work on.

If your agent is already connected, continue to the next step.

Vaiz App Center highlighting Cursor and Claude integrations for connecting AI coding agents via API keys.

Step 3. Prepare the task

Open the task in Vaiz. Describe the expected outcome, add acceptance criteria, and link relevant documents or designs.

Define the scope, technical constraints, and how the result should be verified. Reference relevant components, architectural decisions, and repository instructions. Identify existing patterns to reuse and behavior to preserve.

If the implementation approach is still open, ask the agent to investigate and prepare a plan first. Review the plan before it changes the code.

Vaiz task outlining agent instructions, tool behavior, and layout requirements.

Step 4. Delegate the task

In the task's Coding agents field, click Delegate to agent. This opens the session configuration window.

Vaiz task showing the Coding agents field and the “Delegate to agent” option.

Step 5. Configure and start the session

In the window that opens, select a connected agent (Cursor or Claude), then choose the repository, starting branch, and model.

Optionally, add any additional requirements or constraints for the session.

Select Move and update task to allow the agent to move and update the task in Vaiz.

Review the settings, then click Delegate task to start the session.

Respond to clarification requests as work progresses. If requirements change, update the task description to match the current instructions.

Coding session setup window in Vaiz.

Step 6. Review the changes

When the session finishes, open the task and review the agent's report. If a pull request was created, follow its link to review the changes. If no pull request was created, check the report for blockers and next steps.

Check the changes against the acceptance criteria. Review the test results and, for interface changes, verify the behavior in the running application. Request corrections by describing the problem and the expected behavior.

Inspect the changed files and automated checks in the pull request. Confirm that the changes stay within the agreed scope. Use any attached screenshots or recordings to support your review.

Check which tests the agent could not run and why. Complete the required verification, then merge after review under your repository's rules. Completing the task depends on meeting its acceptance criteria.

Cursor session report and pull request link in Vaiz.

Step 7. Review the task record

All decisions made on the task are added to the Memory field, with links to the corresponding blocks. Follow the links to review each decision.

These records help the next developer understand why changes were made and continue from the decisions already taken.

Check that the task includes the pull request link, verification results, and open questions. After review, update its status according to your team's workflow.

If the pull request is not yet linked to the task, add it through the GitHub field.

Vaiz project memory with saved decisions and coding session logs.

Usage and limits

When budgeting, account for your Vaiz plan and the API usage terms of your chosen provider. Cursor bills Cloud Agents at the selected model's API rates and prompts you to set a spending limit. Claude API costs depend on the model and usage. You can check your Claude API spending and limits in Claude Console. See Cursor billing, Claude API pricing, and Claude limits.

Before starting, check the balance and limits of the account associated with the API key. Set a budget and assign someone to monitor usage. Include repeat runs and revisions after review when estimating costs.

Evaluate the workflow

Start with a small set of tasks with clear acceptance criteria. Before expanding usage, track:

  • Time from task creation to an accepted change.
  • Developer time spent on preparation, review, and revisions.
  • Session costs, including repeat runs.
  • Quality: acceptance criteria met, corrections required, and defects found after merging.

Compare these results with similar tasks your team has completed. Keep outcomes and decisions linked to their tasks so the evaluation can be checked against the work history.