Start with one pilot. Architect for all seven.
As Casey noted, we'll bring a proposal with a recommended pilot and a longer-term plan. A pilot-first path gets working software in front of a Carnegie team in weeks, proves value on real data, and de-risks the broader rollout.
How we'd prioritize your seven
Our initial read — we'll refine it with your teams in Discovery.
| Use case | Impact | Effort | Pilot candidate |
|---|---|---|---|
| Concur receipts | High | Low | ★ Recommended |
| China AI KMS | High | Medium | ★ Recommended |
| Ask-Andy intranet | High | Medium | Phase 2 |
| Contact / CRM | Medium | Low | Phase 2 |
| QC monitoring | Medium | Low | Phase 2 |
| Event logistics | Medium | Medium | Phase 2 |
| Library research | High | High | Phase 2 |
The Concur receipt agent is the fastest high-impact win; the China AI KMS is the highest-visibility flagship. Either makes a strong first pilot.
01Discovery & prioritization
Week 1–2We meet the teams behind these ideas — Finance, the China AI Initiative, Climate/DDD, Comms, HR — to map current workflows, systems, and data access, then score each use case on impact vs. effort.
02Pilot design & security review
Week 2–3Design the pilot end-to-end: agent behavior, the MCP/connector, the human-review step, and the data-governance model. Your IT and security teams sign off before we build.
03Build the pilot
Week 3–7We build on Vercel + Supabase + Claude — the working software you've already seen in these demos — wired to your real systems in a sandbox, with weekly demos.
04Measure & prove
Week 7–10Run the pilot with a real team. Track time saved, error reduction, and adoption. Tune based on how staff actually use it.
05Scale across the seven
OngoingReuse the platform to light up the next use cases — each one mostly configuration on the backbone from the pilot. PRR supports and evolves it.
These demos are Phase 3 — already standing.
We built working versions of all seven before the call to show we move fast and think in production. Next: a Discovery session to pick the pilot and shape it around how Carnegie works.
Revisit the demos