Use Cases · Product

Customer requests that become Linear tickets before standup.

Cortex product agents close the loop between customer signal and engineering output — turning Slack threads into PRDs, catching spec drift before launch, and ranking every feature request without a weekly Notion session.

agent / productslack · live
request · to · linear
you
@cortex read this thread, check if we have an open Linear issue for SSO, draft a PRD with acceptance criteria, and op…
cortex
Turn a customer request from #cs-escalations into a Linear ticket with a full PRD — before the PM wakes up.
SlackLinearInstitutional memory
[ Thesis ]
Every company runs on Minimum Viable Documentation — the least a team can write down and still function. MVD is rational. But we keep asking humans to do the one thing humans were never going to do well. An agent harness with two layers of memory turns documentation into a byproduct of work, the way code does.
Read the founding thesis
[ 01 · Jobs ]

Jobs your agent handles.

Every scenario runs on a real Cortex capability. The memory split shows what stays in this session and what graduates into the team's institutional knowledge.
[ Scenario 01 ]

Turn a customer request from #cs-escalations into a Linear ticket with a full PRD — before the PM wakes up.

Trigger

CSM drops a summary in #cs-escalations: 'Three enterprise accounts asked for SAML SSO this week — they're blocking on it for legal.'

In Slack

@cortex read this thread, check if we have an open Linear issue for SSO, draft a PRD with acceptance criteria, and open a ticket tagged P1

Agent does
Slack — read escalation threadLinear — search existing SSO issuesInstitutional memory — previous decisions on authLinear — create ticket with PRD + ACsSlack — confirm ticket link in thread
Conversational memory

The three account names, the CSM's framing, the legal compliance angle named in this session.

Institutional memory

The decision to prioritize SAML SSO, the rationale, and the customer segment driving it — stored permanently. Future auth decisions surface this context without anyone re-researching.

[ Scenario 02 ]

Monday morning cross-tool status digest — what shipped, what's blocked, what's misaligned — before standup.

Trigger

Scheduled Monday 8:45am recipe. Delivered to #product before the standup starts.

In Slack

Workflow recipe (product.cross_tool_digest): cross-functional product status across Linear, GitHub, Notion

Agent does
Linear — closed + blocked issuesGitHub — merged PRs since last MondayNotion — open design/spec docsSlack — post standup-style digest to #product
Conversational memory

This week's run window and any tools that timed out.

Institutional memory

Recurring blockers tracked over time — 'design handoff consistently arrives the same week as the engineering sprint' surfaces after 3 recurrences, not after a retro.

[ Scenario 03 ]

Catch scope creep and spec drift before it becomes a post-launch surprise.

Trigger

PM runs this after a sprint closes, or automatically when a PR is merged against a Linear epic.

In Slack

@cortex compare what shipped in the [Auth Refactor] epic against the original PRD — flag anything that diverged and anything that was never built

Agent does
Linear — original epic + acceptance criteriaGitHub — merged PRs linked to epicInstitutional memory — prior scope change decisionsSlack — diff report to PM with flagged gaps
Conversational memory

The specific epic compared and the PR range scanned in this session.

Institutional memory

Scope change decisions captured — 'the retry logic was descoped from the Auth Refactor epic on [date] because [reason].' Prevents re-opening debate in the next sprint.

[ Scenario 04 ]

Triage and rank inbound feature requests without a weekly Notion session.

Trigger

Friday recipe. Runs across Slack #product-feedback, Intercom, and Notion capture pages.

In Slack

Workflow recipe: @cortex collect and rank feature requests from this week — group by theme, flag duplicates, estimate vote-weight by customer size

Agent does
Slack — #product-feedback messagesNotion — feature request capture pageInstitutional memory — prior triage decisions and customer tiersSlack — ranked weekly list to #product
Conversational memory

The raw request list and any filter criteria applied this run.

Institutional memory

Running request log by theme, with customer segment metadata. After 4 weeks, emerging patterns surface automatically — 'CSV export has appeared in 7 requests from enterprise accounts this month.'

[ Scenario 05 ]

Write customer-facing release notes from merged PRs in the time it takes to copy the Slack link.

Trigger

PM is ready to ship the release. Types one message with the version tag.

In Slack

@cortex write release notes for v2.4.1 — pull merged PRs from the last 2 weeks, translate into customer-facing language, use our changelog voice

Agent does
GitHub — merged PRs for tag rangeLinear — feature tickets linked to PRsInstitutional memory — brand voice + changelog styleNotion — draft release notes docSlack — post draft link to #product
Conversational memory

The version tag, the date range, and any features the PM explicitly called out in this session.

Institutional memory

Changelog voice and style — defined once, applied consistently to every future release. No PM has to re-explain 'we write for operators, not engineers' ever again.

[ 02 · Two layers of memory ]

Working in every job above.

Two tiers, working in parallel. One fades unless it matters; one compounds forever.
Conversational memory

What happened in this session — the Slack thread you referenced, the customer you named, the draft you iterated on. Persists through the conversation and short-term window. Volatile by design: fades unless it matters.

Institutional memory

Facts that proved their value — accessed repeatedly, confirmed accurate, promoted by the system. “Drift raised their Pro price to $400/seat.” “Acme Corp runs on Salesforce CPQ.” Written once by accident, available forever to every session.

[ 03 · Connected tools on this page ]
SlackLinearGitHubNotionGoogle Workspace

Build products from customer signal, not tribal knowledge.

Deploy in 10 minutes. No DevOps required.