Honcho

@honcho what have you learned about Sarah's communication preferences from our last 20 conversations?

Based on 18 sessions with Sarah spanning 3 weeks, Honcho has identified several patterns: She prefers direct, bullet-point summaries over narrative explanations (observed in 14 of 18 sessions). She typically asks follow-up questions about implementation details rather than high-level strategy. She responds most actively to messages sent between 9-11 AM EST. Recent sessions show increased interest in automation workflows and reduced engagement with manual process discussions.

This example uses Chat With Peer at a higher reasoning level, which incurs usage-based costs and takes 3-8 seconds to process. The quality of learned representation depends on how many sessions and messages have been stored for that peer — new peers return minimal insights.

@honcho store these three conclusions: Alex prefers async updates over meetings, Jordan needs context on legacy decisions before weighing in, and Sam wants cost implications called out explicitly.

Successfully stored 3 conclusions across peers in workspace 'team-alpha'. Honcho has indexed these preferences and will surface them when you query peer behavior or ask for communication recommendations. Each conclusion is timestamped and can be updated or deleted independently. These explicit conclusions will influence future Chat With Peer responses and session-level reasoning.

Create Conclusions writes directly to Honcho's peer models without triggering immediate reasoning costs. However, once stored, these conclusions affect future usage-bearing queries. You need write permissions on the workspace, and conclusions persist until explicitly deleted.

@honcho add the last 15 messages from this thread to a session called 'q1-planning' and summarize what we decided about the API redesign.

Added 15 messages to session 'q1-planning' with 4 peer attributions (you, Sarah, Jordan, Alex). Session-level reasoning identified three decisions: (1) REST endpoints will be versioned with /v2/ prefix by Feb 15, (2) GraphQL layer deferred to Q2 pending team capacity review, (3) breaking changes require 90-day deprecation notice. Two open questions remain: authentication migration path and rate-limit policy for legacy clients.

Create Messages with reasoning enabled triggers usage-bearing processing that extracts decisions, questions, and peer stances. Processing happens asynchronously — the summary above may take 5-15 seconds to generate. Rate limits apply per workspace, so batch-adding messages from long threads can hit quota faster than expected.

When Honcho makes sense for rotating support shifts

A 6-person support team rotates coverage across three time zones. Each agent picks up tickets mid-conversation with no shared context beyond the ticket history. Honcho's peer-and-session model fits this scenario cleanly: each customer becomes a peer, each ticket thread a session, and the "Chat With Peer" tool lets the next agent query what the customer cares about without re-reading 40 Slack messages. The reasoning levels cost tokens, so you'll want to benchmark whether a medium-reasoning query is cheaper than having your agent summarize the transcript itself. If your median ticket spans fewer than three handoffs, the overhead probably isn't worth it. If handoffs are daily and your team burns 20 minutes per shift catching up, Honcho pays for itself in the first week.

Honcho for deal context across a 10-rep sales floor

A 10-person sales team works 200 active deals. Reps jump between accounts, and the CRM notes are stale or missing. Honcho's "Get or Create Peer" and "Create Conclusions" tools let you treat each prospect company as a peer and store explicit takeaways after every call—budget authority, pain points, competitor mentions. When a rep picks up a warm lead three weeks later, "Chat With Peer" surfaces what matters without scrolling through call transcripts. The catch: if your team already lives in Salesforce and writes decent notes, Honcho is redundant. The win is when your CRM discipline is weak and you need the AI to extract and retrieve context automatically. For teams under 5 reps or deals that close in one call, the setup cost outweighs the benefit.

When Honcho beats manual tagging for user research

A 3-person product team runs 40 customer interviews per quarter. Each interview generates a transcript, and the team needs to answer questions like "which enterprise users mentioned SSO as a blocker?" without re-reading every file. Honcho's session-and-peer structure maps naturally: each interviewee is a peer, each interview a session, and "Create Messages" ingests the transcript with reasoning enabled. Later, "Chat With Peer" queries let you ask cross-interview questions in plain English. The trade-off is token cost—reasoning on 40 hour-long transcripts isn't free. If your interviews are short (under 15 minutes) or you only run a handful per quarter, a shared Notion doc with manual tags is faster and cheaper. Honcho wins when interview volume exceeds your team's tagging discipline and you need retrieval that understands intent, not just keywords.