The first CRM can be a disciplined ledger
A founder does not need a heavy CRM to get the first 100 customers. They need a clean ledger: person, company, source, pain, promise made, last touch, next step, owner, and learning. An agent can maintain that ledger from notes, email drafts, call summaries, and prospect research.
The most important field is not stage. It is next honest action. Send a teardown. Ask for the missing stakeholder. Follow up after launch. Mark not a fit. Ask for feedback. A useful CRM agent keeps the founder from losing context while avoiding fake pipeline confidence.
Make follow-up specific
Generic follow-up is where early relationships go cold. The agent should draft notes that cite the original pain, the promised artifact, and the next useful step. If the prospect came from X search, reference the public problem they raised. If they came from the waitlist, reference the page or question that brought them in.
Paul Graham's startup advice emphasizes manual recruiting and attentiveness. The CRM agent makes that attentiveness easier to sustain. It should remind the founder to do the human thing at the right time, not pretend the relationship is automated.
Use approvals for anything customer-facing
OpenAI's agent docs describe human review and guardrails as part of agent workflows. In founder sales, every outbound note deserves review until the voice, offer, and qualification rules are proven. The agent can prepare. The founder sends.
Have the agent flag risk: unsupported claims, too much personalization, unclear opt-out, weak relevance, or a promise the product cannot keep. This prevents the fastest path to 100 customers from becoming the fastest path to brand damage.
What to review every Friday
Review stuck deals, repeated objections, highest-intent sources, fastest conversions, and product gaps. Then update the signal map, landing page, and measurement loop. Founder sales should teach the whole system.