Adoption journey¶
Bringing a local LLM workflow into production is not a single decision. It is a four-phase journey. Each phase has a clear entry criterion, a clear exit criterion, and a deliverable that the next phase can build on.
flowchart LR
A[1. Discover] --> B[2. Pilot]
B --> C[3. Build]
C --> D[4. Scale]
D -.->|new team| A
style A fill:#e3f2fd
style B fill:#fff3e0
style C fill:#c8e6c9
style D fill:#f3e5f5| Phase | Question you answer | Doc |
|---|---|---|
| 1. Discover | "Is local LLM worth considering for my team?" | discover.md |
| 2. Pilot | "Can we ship one workflow in 2 weeks?" | pilot.md |
| 3. Build | "How do we put it in production?" | build.md |
| 4. Scale | "How do we roll out to 5+ teams?" | scale.md |
| 5. FDE | "What is the role that runs all four phases?" | fde.md |
The FDE lens¶
The four phases are the operations playbook. The role that runs them is the Forward Deployed Engineer (FDE) — the senior engineer embedded with the customer who ships the first working system on the customer's own hardware and hands it over before they leave. FDE is the word the rest of the AI industry uses for this job, and it is the role a local-LLM programme cannot do without: the data-residency constraint forces someone on-site, and "someone on-site" is the FDE.
FDE chapter (5) is the lens that ties the four phases to the rest of the industry. It maps Audit → Discover, Evals → Pilot, Deployment → Build + Scale, and explains why the on-prem constraint makes the FDE mandatory rather than optional. Read it after Discover, before you hire.
When to use this¶
Use the four phases when you are evaluating AI Work Flow for Business for a new team or workflow. If you already have a working local-LLM prototype, jump to Build. If you have one team in production and want to expand, jump to Scale.
What you should NOT do¶
Do not skip Discover and go straight to building. Most failed local-LLM projects fail because the team picked the wrong workflow. The Discover phase exists to prevent that.
Do not stay in Pilot indefinitely. A pilot that has not graduated in 6 weeks is a sign the workflow was the wrong choice. Kill it and try another.
See also¶
- Why AI Work Flow for Business? — the case for local-first
- Demo — what the modules actually do
- Case studies — real or representative walkthroughs
- Architecture — the five-layer stack that runs these phases
Next step¶
Finished Discover? Read the case studies to see what other teams shipped. Then go to Pilot.