What AI Training for Professional Services Firms Should Cover at the Intermediate Level
What separates real infrastructure from basic fluency — and what a curriculum built for firms past the basics looks like.
Foundational AI training - the kind covered in programs like our own 101 - teaches people to prompt reliably, understand how AI tools work, and apply them to recurring tasks. When it works, everyone on the team can write prompts that produce usable output, and the firm has a shared baseline.
What foundational training does not teach is what to do next. Getting individuals to a reliable standard is the first problem. The second problem - which is also the harder one - is building the infrastructure that makes individual capability into something the firm can compound.
What firms that have done the basics need next
After a successful foundational training, a firm typically has: a team that can prompt, a shared prompt library in some form, a Claude Project with basic firm context loaded in, and a rough sense of which workflows AI is most useful for.
What it does not yet have: Skills built from its best prompts, a library that is organized and governed well enough to grow, data connections that make AI genuinely aware of the firm's own work, or documented workflows in a form that survives personnel changes.
Intermediate training bridges that gap. It is not a continuation of prompt craft - it is a different kind of work.
The six things intermediate training should cover
01 — What a Skill is and how to build one
The distinction between a prompt and a Skill is not obvious to most people until it is explained. A Skill is a prompt designed for someone other than the person who wrote it: clear input requirements, defined output format, documented use case. Intermediate training should cover how to take a working prompt and turn it into a Skill someone else can use without any explanation from the author.
02 — How to organize a Skills library that grows
A library is not a folder. Naming conventions, use-case tagging, ownership assignment, and governance - how Skills get added, reviewed, archived - are what make the difference between a library people use and one they ignore. This is organizational design, not prompting skill.
03 — How to run longer, multi-step tasks (Cowork)
Most of the time savings from AI in professional services work come not from single-prompt exchanges but from sessions where AI handles multiple steps of a complex task while you provide direction and judgment at key checkpoints. Intermediate training should cover how to brief these sessions, where to set checkpoints, and how to turn a successful session into a reusable Skill.
04 — How to connect AI to internal data
AI that can access the firm's document archive, knowledge base, and client context is a different tool from AI working from generic training. Intermediate training should cover how to set up these connections, what to connect first, and how to prompt AI that has access to internal data.
05 — How to evaluate and connect external data sources
For firms that depend on current market data, legal databases, or regulatory tracking, connecting those sources via MCP multiplies the value of every research task. Intermediate training should cover which sources are worth connecting, how to evaluate them, and how to build Skills that use external data with appropriate sourcing requirements.
06 — How to document workflows as firm assets
The hardest thing to get right in AI infrastructure is documentation that survives. Most workflow documentation is written once and never updated. Intermediate training should cover how to write workflow documentation that is specific enough to be useful and maintained consistently enough to stay current - and how to use AI to do that documentation work more efficiently.
What makes intermediate training different from advanced training
Intermediate training should produce working infrastructure. By the end of a 202-level program, firms should have a real Skills library with real Skills in it, at least one working data connection, and at least one workflow documented in a form anyone on the team can use.
Advanced training - the 303 level - takes that infrastructure as a starting point and builds custom tooling on top of it: proprietary agents, dashboards, integrations. You cannot build those things well without the 202 foundation in place.
Apparatus 202 covers all six of these areas across six modules, and leaves every participant with working Skills, connected data, and documented workflows. If your team is at the foundational level and needs to get there first, Apparatus 101 is the right starting point.