The Difference Between Sending Your Team to a ChatGPT Course and Building a Firm AI System
Why individual training and firm-level systems are different problems — and why one doesn't lead to the other.
A managing partner at a consulting firm recently told me their AI strategy: "We're sending everyone to a ChatGPT course." The course would cover the basics - what AI can do, how to write prompts, when to use it. Two hours. Everyone would leave with a certificate.
The impulse makes sense. The firm wanted to do something, and individual training felt like the right lever. The problem is that individual training and firm-level AI capability are separate problems. One doesn't lead to the other.
What individual training actually produces
Better individual users who still can't share
Individual AI training - even good individual training - produces individual improvement. The person who took the course gets better at using AI for their own work. That's real value. But it stays individual.
When that person figures out a great prompt for contract review, it lives in their browser history. When they write a better version next week, nobody else knows. When they leave the firm, it's gone. The firm's collective AI capability doesn't improve because nobody is doing anything with what individuals learn.
Individual training also produces fragmented standards. When everyone learns AI from a different source using different frameworks, you end up with ten different mental models of how to prompt. You can't share prompts effectively across those mental models - people can't read each other's prompts, can't improve them, can't build on them. The AI use stays ad hoc.
What a firm AI system actually is
Three things that make it institutional
A firm AI system isn't a software product. It's a set of shared practices that make individual AI use accumulate into firm capability.
A shared prompt standard
Everyone uses the same framework for structuring prompts. Not the same prompts - the same structure. When the firm has a common vocabulary for what goes into a good prompt, people can read each other's prompts, evaluate them, improve them, and actually use what others have built. That shared structure is what makes sharing possible.
A maintained prompt library
Somewhere that good prompts actually live, organized by workflow, maintained by someone who owns it. Not a folder with 50 experiments nobody opens. A library with 10 well-built prompts that people find, trust, and use regularly. The library is the mechanism for individual learning to become institutional knowledge.
A norm around contribution
A simple rule that keeps the library alive: if you use a prompt three times and it works, it goes in. That threshold is low enough to encourage contribution without requiring formal evaluation. The library grows as the team uses it. Without a norm, the library stagnates after the first week.
Why one doesn't lead to the other
Individual skill and collective infrastructure are different problems
Individual training addresses individual skill. That's valuable. But it doesn't produce a prompt standard, a maintained library, or a contribution norm. Those require deliberate decisions at the firm level - decisions that nobody makes during a two-hour ChatGPT course.
The reason firms end up with the training-but-no-system problem is that individual training feels like a concrete action while building infrastructure feels vague. Sending people to a course produces a measurable output (certificates, completion rates). Building a prompt library requires someone to make decisions, own something, and maintain it over time.
Both things matter. Individual skill is the prerequisite for being able to contribute to the library. Firm infrastructure is what makes individual skill compound. Doing only individual training gets you better individual users who still can't build on each other's work. Training that combines both - individual skill and firm infrastructure, in the same session - is what actually moves the needle.