How an AI agent can carry customer evidence into scoped product and engineering work
This workflow gives an AI agent a defined job, a bounded set of records, and a result a person can review. The agent reads the relevant Unthread context and matches it with Linear, applies the rules in the prompt, and keeps the source behind every recommendation. It returns a proposed handoff rather than taking consequential actions on its own.
Can an AI agent carry customer evidence into scoped product and engineering work?
Yes. Start with the scope, date range, decision rules, and fields that identify the right records. The agent can collect the evidence, compare states or sources, mark conflicts and missing data, and organize the result around the outcome above. A reviewer then checks the matches and judgment calls before approving messages, record updates, bookings, purchases, publishing, or other write actions. The guide below shows the records, boundaries, prompt, and handoff needed for this specific workflow.
Prove the problem before drafting an issue
Unthread supplies the customer report, environment clues, troubleshooting steps, attachments, impact, and support status. Linear supplies issue history, product area, current owner, and acceptance criteria. Used together, they can distinguish a known defect from a configuration question or a new reproducible problem.
Define the Unthread cases and Linear team in scope. Ask the agent to redact customer identity and secrets before moving evidence. It should search Linear using error text, behavior, component, and reproduction terms, then explain why a candidate issue matches or differs.
Example starter prompt
Use Unthread cases [case links] and Linear team [team] to investigate [reported problem]. Remove customer names, contact details, credentials, and unrelated account data from the handoff.
Build the smallest supported reproduction from the conversations, including environment, expected result, observed result, frequency, impact, and troubleshooting already tried. Search Linear for duplicates and show match evidence.
Do not create, update, or prioritize an issue and do not contact customers. Return duplicate candidates or a complete proposed Linear issue with redacted support links.Keep impact and priority separate
Verify the reproduction against the source Unthread cases and note where evidence differs. Then review Linear candidates with the owning team. Several customers reporting a problem establishes support impact, but engineering priority also depends on severity, reach, workarounds, and current plans.
Questions this workflow answers
Do these support reports describe one reproducible bug, and is engineering already tracking it?
The agent compares expected and observed behavior, environment, account state, error text, frequency, workaround, and troubleshooting across selected cases. Customer identity, credentials, and unrelated account detail are removed before evidence leaves the support system. Differences stay visible; reports with the same symptom may have separate causes.
A minimal reproduction uses only steps supported by the conversations and marks missing preconditions. The agent searches issue history by behavior, component, error, and environment, then classifies candidates as direct, partial, related, or unrelated. If an issue matches, it drafts a redacted impact update rather than another ticket.
New issue drafts include reproduction, impact boundaries, workaround, evidence links under correct access, acceptance checks, and unknowns. Report count describes support impact, not automatic priority. Product and engineering still weigh severity, reach, current plans, and effort.
The handoff allows owners to create, merge, revise, or reject work after checking the source cases. No issue status, priority, or customer communication changes during the run.
The handoff should include case links under proper access, reproduction, evidence, duplicate analysis, impact, workaround, acceptance checks, and questions. Product or engineering owners control issue creation and priority.
The agent should extract environment, account state, user action, expected result, observed result, timing, errors, workaround, and safe evidence from each case. It can group reports only when the same sequence and failure mode fit, keeping look-alike symptoms separate. Existing Linear issues are compared on reproduction and scope, not title. The draft records the number and dates of confirmed cases without inflating impact from duplicates or unrelated complaints. Acceptance checks replay the representative state and protect a neighboring successful path. Customer identities and private attachments stay in support, linked under existing access rather than copied into the engineering issue.