How to AI
We hire people to do Jobs — these are bundles of work.
We bundle work because a) it makes the person effective — everything they need lives in one head, and b) it’s the only way to justify hiring someone for a role.
- Research accounts
- Find contacts
- Write outreach
- Handle replies
- Book meetings
- Update CRM
AI isn’t effective at completing all of these task outcomes at once.
Research these 200 accounts, find the right contacts at each one, write personalised outreach, send three follow-ups, book meetings for next week, and keep the CRM updated as you go.
Great question! Here's a 2,000-word overview of CRM hygiene best practices…
This means understanding what to achieve — and how — is hard.
- Context — ICP, tone, real examples
- Instructions — what good looks like
- Tools — email, CRM access
- Checks — review + feedback loop
- Context — objections, pricing, FAQs
- Instructions — when to escalate
- Tools — inbox, calendar
- Checks — human sign-off
- Context — territories, routing rules
- Instructions — qualification bar
- Tools — calendar, CRM write
- Checks — no-show follow-up
AI systems are successful when you compound the creation instead of trying to get a tool to complete everything at once.
So don’t start with the job. Start with one task — the smallest one with a clear outcome.
- Research accounts
- Find contacts
- Write outreach start here
- Handle replies
- Book meetings
- Update CRM
Give that one task everything a good hire would get. Now it’s not a prompt — it’s a system.
- Context — ICP, tone, ten real emails that worked
- Instructions — one outcome: a sendable first draft
- Tools — CRM read, email templates
- Checks — a human approves every send
Then keep going. Every system you ship makes the next one cheaper to build — and they start feeding each other.
Until the branches reconnect — and the job re-emerges. Built, not hired.
- Research accounts
- Find contacts
- Write outreach
- Handle replies
- Book meetings
- Update CRM
You don’t hire AI to do a job. You grow it into one — one task at a time.
That’s how to AI.
Part two: How to roll out AI →