How to investigate a customer issue with Maito

An AI support engineer needs the same starting point as your team: a clear customer report, the relevant project, and access to evidence. This guide walks through setting up a Maito investigation and reviewing its result.

The export issue below is an illustrative scenario. It is not a customer case study or a claim about a measured outcome.

1. Connect the project and choose your provider

Create a workspace and connect your AI provider. Add the repository and documentation for the product you support. Connect the tools needed for this investigation, such as a ticket system or application logs, through the MCP connections your team enables.

Start with the access the task needs. Store credentials in Vault and reference them by name rather than pasting secrets into a conversation. Connecting a repository does not automatically grant access to your logs, production database, or deployment system.

You can work with Maito in Slack or through your AI assistant. Both should point to the workspace with the relevant context. See what Maito works with for the connection model.

2. Describe the report and the expected behavior

Suppose a customer says that a large export stops halfway through. Share the time range, the expected result, the affected workflow, and a sanitized example if one is available.

Give the investigation a concrete instruction:

Investigate why this export stops before completion. Check the connected code and available logs, identify missing evidence, and propose a reproduction. Explain the likely cause with references. Prepare a fix for review if the cause can be verified; do not deploy it.

Avoid sending customer secrets or unrelated personal information. If a source is unavailable, tell the agent rather than expecting it to infer what happened.

3. Review the evidence before deciding on a fix

Ask which code paths and log entries support the explanation. Separate observations from hypotheses. A plausible explanation is a starting point; a reproduction or a relevant failing check is stronger evidence.

Your team may need to provide additional context or access. If the agent cannot reproduce the issue, keep that uncertainty in the investigation and in any proposed customer reply.

4. Review the proposed change and its checks

Independent coding tasks use separate sandboxes, branches, and checkouts. Ask Maito to run your project's relevant checks and explain what changed, why it addresses the issue, and what remains unverified.

Review the proposed pull request using your normal engineering process. Check the behavior reported by the customer, error handling, and nearby workflows. Your team decides whether to merge and release the change.

Passing configured checks does not guarantee that a change is safe in every production environment. Missing test coverage and unavailable tools should be stated explicitly.

5. Keep the customer response and follow-up together

Ask for a draft reply grounded in the findings. Review it before sending it to the customer, especially when the cause or impact remains uncertain.

If a recurring check would help, create a Loop with a specific goal, the required tools, and a schedule or trigger. Decide who will review its findings and when the work should be paused.

What to prepare before the first task

• A connected AI provider and the relevant repository.
• Project documentation and access to the evidence needed for the issue.
• A clear report, expected behavior, and scope of allowed actions.
• The commands or checks your team uses to verify a change.
• An engineer who can review findings and proposed fixes.

Maito is priced per workspace; AI provider usage is billed separately. Read the FAQ for provider authentication, cloud credits, and review workflows, or contact us about a demo.