Your second brain, connected. Say goodbye to your subscription monster. Replace Notion, Slack, Zoom, Obsidian, Evernote and Matter + a dozen more apps for free.
Lead scoring models fail on inputs before model choice. Rules need clean fields. Predictive needs 12+ months of consistent outcomes. Agent-built needs verifiable identity anchors. Pick by data readiness. https://t.co/VOsQSHbQrY
@dan__rosenthal For prospecting, a useful test is whether the agent can explain why this account belongs in the campaign today. A hiring signal means very different things for a recruiting offer and a finance offer.
@rseroter The app-only write tool in the tutorial is a useful detail. An account-research workflow could let the agent find candidates, then let a person select exactly which records to update.
@doubledescent0 Company classification looks like an interesting fit for local inference.
Have you tested overlapping categories such as consultancy, agency, and software vendor?
@TylerBrownAT How often did Jev flag a page for review rather than place it? The split between wrong placements and abstentions would be useful for teams reorganizing a live knowledge base.
@LangChain@sydneyrunkle A useful sales-routing eval would include resellers, consultants, and vendors alongside real buyers. Same industry vocabulary, very different intent.
@dharmesh@meetgranola@YouSpotHub Does the graph distinguish the company you met with from a competitor mentioned in passing? That relationship matters as much as extracting the name.
@ffumarola@CompleteSkeptic Stable source IDs and per-item errors would make that especially useful for enrichment jobs. One bad record should be retryable without reclassifying the whole batch or losing the mapping back to the CRM.
@bbla1681 The Slack handoff should carry the account, the evidence, and the exact decision needed. Otherwise the rep just inherits a different inbox to investigate.