1Learn
Multi-source truth
Combining Crunchbase + LinkedIn + news + internal data. Entity resolution and dedup logic. Verification passes (AI checking AI). Confidence scoring.
4
AGENTIC
Orchestrator
Multiple data sources feed systems that verify, dedupe, and act. Agents work for you while you're in meetings.
01 — Modules
2Apply
Agentic workflows
Subagents and parallel work. Long-running agents. Giving agents tools (MCP from the other side). Guardrails and human-in-the-loop points.
3Extend
Monitoring & reliability
The deal-alert agent pattern. Evaluating output quality over time. Cost control. Knowing when an agent silently broke.
02 — Tool belt
- ▸Claude Code + subagents / Agent SDK
- ▸Scheduled agents
- ▸Slack / email as output channels
- ▸Logging
03 — Say it out loud
Terms you should be able to explain to a colleague.
- agent vs. workflow
- entity resolution
- human-in-the-loop
- evals
- why verification is the hard part
04 — VC playbook
- 01Deal-alert agent that monitors signals and pushes verified Slack alerts
- 02Inbound triage agent that scores decks against your thesis
- 03Portfolio early-warning from cross-source signals
- 04Automated first-pass diligence brief with cited sources
Checkpoint challenge
The Sleepless Analyst
An agent that has run reliably for 2+ weeks, monitoring something you care about, pushing verified alerts to Slack/email — including at least one alert that was genuinely useful and one false positive you diagnosed and fixed.
Sign up to work through this→