Human-Centered Automation Playbook
Every episode of The Build eventually hits the same tension: automation is powerful, but trust is fragile. Tom Spencer’s conversations with founders and operators reveal a practical framework for deploying AI agents that feel like an extension of your team—not a faceless script.
Principle 1: Start with Narrative, Not Tasks
- Document the promise you make to customers. Agents should reinforce, not dilute it.
- Identify the moments where human judgment matters most and protect them.
- Use AI to draft initial responses or research, but let humans author the final story.
Starter Prompts
You are an AI partner supporting {team}. Your job is to surface context, options, and risks.
Tone: {brand_tone}. Never guess—flag uncertainty and request human review.
Principle 2: Create Guardrails and Escalation Paths
- Maintain a “too weird to automate” list of scenarios that must go to a human.
- Tag every agent action with confidence scores and supporting evidence.
- Schedule weekly transcript reviews—Tom calls this the “agent retro”—to catch drift early.
Escalation Example
- Agent identifies a complex billing dispute.
- Responds with acknowledgement + summary of known data.
- Assigns to human operator with suggested next steps and context links.
Principle 3: Instrument Trust
- Track customer sentiment before and after automated touchpoints.
- Interview frontline teammates about where automation helps vs. hinders.
- Publish a living FAQ explaining what is and isn’t automated; transparency builds confidence.
Metrics to Watch
- Resolution time saved vs. CSAT change.
- Percentage of conversations escalated to human.
- Number of “off-brand” responses caught in weekly audits.
Launch Checklist
- [ ] Narrative document shared with every agent (purpose, tone, red lines).
- [ ] Guardrail scenarios codified with owners.
- [ ] Telemetry dashboard tracking sentiment + escalations.
- [ ] Agent retro scheduled at least bi-weekly.
- [ ] Human follow-up workflow defined for complex cases.
“Automation should feel like a thoughtful teammate teeing you up—not a black box making promises you can’t keep.” — Tom Spencer
From The Build Podcast
- Ground yourself in the broader themes with The Build Podcast: Field Notes for AI-Native Teams.
- Operationalise the cadence alongside this playbook: Operating Cadence for AI Builders.
- Catch new conversations on YouTube.