Startup Agent Adoption

This skill is derived from: https://claude.com/blog/building-ai-agents-for-startups (published 2025-11-03).

Instructions

• Identify candidate workflows where time is spent repeatedly and outcomes can be reviewed by a human.
• Choose an initial project with high impact and low-to-moderate risk; define a clear acceptance test for outputs.
• Design the agent as modular capabilities (shared components, shared prompts, shared tools) so it can expand to other teams.
• Make decisions observable: log inputs/outputs and provide a rationale or evidence trail when possible.
• Add human-in-the-loop checkpoints for high-stakes actions or ambiguous decisions.
• Scale from the first win by expanding coverage (more cases) and raising automation (more steps) only after quality is stable.

Examples

Example: picking a first agent project

• Workflow: inbound support triage

• Human oversight: agent drafts classification and response; human approves before sending

• Acceptance: 95% correct routing + fewer than N unsafe responses per week

Example: scaling after an initial win

• Reuse the same classification module across support, sales lead qualification, and ops requests

• Add observability (reason + citations) before increasing automation

Templates

See:

• templates/first-project-scorecard.md
• templates/scaling-plan.md

References

See:

• references/adoption-principles.md