Every engagement starts with a process somebody already runs by hand. This track turns one into a specification you can hand to a builder: what the steps are, which of them must never be left to a model, whether the data can support the outcome you are promising, what an error costs, and what the solution may cost per resolved ticket to run.
What you'll accomplish
This is the first track of the Forward Deployed Engineer (FDE) for MSPs ladder, and it covers the half of the job that happens before anything is built: mapping. An FDE is the person who takes an AI solution into somebody else's business and is answerable for what it costs and whether it works - and this ladder certifies that role, rung by rung. By the end you will have taken one real process from your own organisation and produced a one-page specification for it - a process map at system, field and trigger altitude, every step marked rule or judgement with the determinism boundary drawn, a note on whether the data is fit for the outcome, written acceptance criteria with a named failure taxonomy, a break-even accuracy, a cost-per-resolved-ticket estimate, the human approval gates placed and owned, and an access list naming every credential, scope and approver the build will need.
One boundary is worth stating before you start, because it defines what this credential is. This track certifies judgement about platform-delivered AI solutions: solutions assembled on an automation platform, where the engineering problem is decomposition, evaluation and cost rather than writing production-grade software. It does not certify, and does not claim to certify, the frontend and backend engineering depth that a frontier lab's forward deployed software engineer carries. That distinction is not a disclaimer - it is the content. Most of what decides whether a platform-delivered solution succeeds is decided in the mapping, and the mapping is what these ten stages teach.
The single most valuable judgement in the track is the first one: whether to build with a model at all. The credential scores it directly. The exam will ask you to decide whether to build, and a correctly argued decline is a passing answer - graded here as a first-class outcome, not as a failure to deliver.
What you'll need
- The AI Builder credential, this track's prerequisite - or a pass on its exam directly. If you already build and deploy automations for a living, the approved route is to challenge the AI Builder exam rather than work the content; same exam, same bar.
- Lived delivery experience. You should have already built and deployed at least one real automation inside your own organisation before starting. The exercises assume you have a queue, a build surface and a process owner you can talk to.
- One real process worth mapping - ideally one somebody has asked about automating, messy enough to have exceptions.
- Access to the person who does the process, not just the person who owns it. Stage 2 cannot be done from documentation.
- About 14 hours, best spread over several sittings. The stages build one artifact cumulatively, so leaving weeks between them means re-reading your own earlier work.
- A certification account so progress syncs and the exam unlocks.