Drew Breunig’s AI strategy ideas—surfaced by Simon Willison—cut through the hype: ship AI where users already are, measure impact in minutes saved, and prioritize distribution over model size. Here’s the nugget-sized playbook.
Source: Simon Willison on Drew Breunig
1) Distribution beats model size
Great models don’t matter if no one uses them. Distribution—inside existing products, channels, and habits—wins. Make AI an upgrade, not a detour.
2) Ship into existing workflows
Embed AI into the tools people already trust (email, docs, CRM, IDEs). Minimize context switching. Surface value where the task starts and decisions get made.
3) Measure value in minutes
Quantify time saved per user per week. Tie that to dollars using fully loaded costs. If your AI feature can’t show up in a CFO-friendly model, it won’t scale.
4) Data advantage compounds
Private usage data (prompts, outcomes, corrections) is the flywheel. Use it to improve routing, retrieval, and UI—not just to fine-tune models.
A 30-day AI rollout playbook
- Week 1: Pick one high-friction task. Baseline it (median minutes spent, error rate, rework).
- Week 2: Ship a lightweight copilot inside the existing tool (no new login). Add clear guardrails and one-click rollback.
- Week 3: Instrument everything (time-on-task, adoption, prompted vs. unprompted use, correction rate).
- Week 4: Review with finance. If ROI > 3x and users opt in unprompted, expand to adjacent tasks. If not, kill or iterate.
Corroborating trend: the Stanford AI Index shows broad experimentation and rising investment—making distribution and measurable ROI even more critical.
Takeaway
Treat AI as a product, not a demo. Win on distribution, workflow fit, and a simple metric: minutes saved per user per week.
Get more nuggets
Enjoyed this? Subscribe to our free newsletter for 2-minute, practical AI insights: theainuggets.com/newsletter
