The Fluency Gap.
Five levels of AI capability your team should master — and why most companies have invested in the tool but not in the people.
The most expensive software your team is barely using.
Walk into the average company that has rolled out Claude, ChatGPT, or any of the major AI products to its workforce. Look over the shoulders of the people using it. You will see something close to the following: a text box, a question typed into it, an answer copied out of it, pasted into an email or a document, and the chat window closed.
This is what nearly everyone is doing with their AI tools. Asking questions. Reading answers. Copying and pasting. Treating a system capable of executing multi-step work across real business systems as if it were a smarter version of Google.
It is the equivalent of buying every employee a laptop in 1998 and watching them use it as a typewriter. The hardware is capable of dramatically more. The barrier is not the technology. The barrier is that nobody taught the people how to use it.
This is the fluency gap. And it is the single largest source of wasted AI investment in business today.
The barrier is not the technology. The barrier is that nobody taught the people how to use it.
What the fluency gap actually costs
The economics of this gap are straightforward and ugly. A typical mid-sized company is now paying somewhere between $15 and $60 per seat per month for an AI subscription, multiplied across hundreds or thousands of employees. That is real money out the door, justified to leadership on the basis of a productivity hypothesis nobody is measuring.
Meanwhile, the actual productivity lift varies enormously across employees. The top users in any organization are pulling ten, twenty, sometimes fifty times the value from the same subscription as their colleagues. The reason is not that they are smarter or younger. The reason is that they have, usually on their own, learned what the tool can do.
The leadership question worth asking is not whether to roll out AI. That decision has been made. The question is whether the rollout will produce the kind of return that justifies the spend, or whether five years from now the company will be paying for licenses that 85% of employees still use as fancy Google.
What follows is a five-level map of AI capability. Each level represents a meaningful jump in what the tool can do for the user, and each level requires a specific set of skills to operate. The goal is not to push every employee to Level 5. The goal is to be deliberate about which employees need which level, and to invest in getting them there.
Level One: Foundations.
Foundations is where most users plateau. The skills sound basic. The fact that 85% of paid AI users never master them is the entire reason this essay exists.
How to talk to an AI system productively.
The interaction itself. Prompt construction, model selection, voice input, memory. Get this right and every subsequent level becomes meaningfully easier. Get it wrong and the user’s ceiling is permanently lower.
Treating the AI as a capable junior colleague who needs context, not a search engine that needs keywords. Every Level 1 skill flows from this single shift.
Level Two: Context.
The first real capability jump. At Level 2, the user stops describing their world to the AI in every conversation and starts giving the AI durable access to it.
Projects, connectors, and skills.
The user who masters Level 2 stops pasting documents into chat windows and starts giving the AI durable access to their actual work. The output quality jump is dramatic.
Investing in context infrastructure rather than re-explaining context every time. The shift from “what can this conversation do for me?” to “what should this AI know about my work permanently?”
Level Three: Creation.
The shift from using the AI to write text to using the AI to produce finished deliverables. This is where the productivity numbers start getting genuinely surprising.
Artifacts, visuals, and design.
At Level 3, the user stops thinking of AI as a writer and starts thinking of it as a builder. Reports, dashboards, mock-ups, decks, working prototypes — all generated inside the conversation, refined through dialogue, exported as needed.
Iteration. The first output is rarely the final output. Fluent Level 3 users have learned to refine through conversation rather than start over — ten refinements producing better work than ten fresh attempts.
Level Four: Autonomy.
The shift from working with AI to delegating work to AI. The user stops being the operator and starts being the supervisor.
Computer use, scheduled tasks, and mobile control.
At Level 4, the user stops being the one doing the work. They describe what needs to happen, hand it off to an AI agent that has access to their browser, their computer, or their messaging apps, and check the result later.
Knowing how to describe a multi-step task so an autonomous agent can execute it without needing to ask clarifying questions every few minutes. This is harder than it looks, and it’s the skill that separates Level 4 users from people who give up on agents after the first attempt fails.
The Level 4 capabilities are powerful enough that the cost of getting the supervision wrong is real. An autonomous agent that sends emails on your behalf, files transactions, or makes changes to client systems can do material damage in minutes if it’s pointed at the wrong work or given access it shouldn’t have. The skill at this level is not just operating the tool, but knowing which tasks belong in autonomous mode and which still require a human in the loop.
Level Five: Engineering.
The frontier. At Level 5, the user is not just operating AI tools but building with them. New systems, new workflows, new internal tools created on demand by people who would not have been called “technical” two years ago.
Claude Code, APIs, and custom workflows.
At Level 5, the user moves from consumer to builder. They are not just running AI tools that exist; they are building the AI tools their company needs.
A mindset shift from “What can I do with the tools that exist?” to “What tools should I build for the work I actually do?” This is the level where AI stops being a productivity tool and starts being a multiplier on what the whole organization can build.
Not every employee needs to reach Level 5. Most will never need to. But the companies pulling away from their competitors are the ones who have identified the handful of people on their team who should reach Level 5 — and invested in getting them there.
Why this doesn’t happen on its own.
The five levels described above are not secrets. The features are documented. Anthropic publishes guides. There are tutorial videos on YouTube, blog posts, courses, and free resources covering every capability in this essay. So why are 85% of users still stuck at Level 1?
Three reasons. All of them structural. None of them solved by an internal Slack message saying “the new AI tools are great, everyone should learn them.”
One. The curse of optionality
The same flexibility that makes AI tools powerful makes them paralyzing for new users. Faced with infinite possibilities, most people retreat to the one use they already understand — typing a question. There is no forcing function pushing them toward Level 2 or 3 capabilities. They never discover what they could be doing.
Two. The taste problem
The difference between a Level 1 output and a Level 3 output is often invisible to the person producing it. Both look like answers. The Level 3 output is dramatically better — but if you don’t know what good looks like, you can’t tell the difference. Most employees lack a frame of reference for what fluent AI use produces, so they have no signal that they should be aiming higher.
Three. The time problem
Learning a new capability requires deliberate practice on tasks that matter, with feedback from someone who knows what good looks like. Most employees don’t have time to experiment freely with their AI tools during a workday with deadlines. They use what they already know how to use, even when better options exist five clicks away.
The companies pulling away are not the ones with the best AI subscriptions. They’re the ones who treated AI fluency as a skill worth teaching, the same way they teach Excel or sales methodology.
The fix for all three problems is the same: deliberate teaching. Showing people what fluent use looks like. Walking them through real work using their actual tools. Building practice into the workday rather than expecting it to happen on the side. None of this is exotic. It’s how every other professional skill in the history of business has been transmitted, and it’s what is conspicuously missing from most companies’ approach to AI.
What actually works.
Over the last eighteen months, we have run AI fluency programs for teams ranging from five-person founder-led companies to professional services firms with dozens of employees. The pattern of what works has been consistent enough that it’s worth describing concretely.
Identify who needs which level
- Not every employee needs Level 5. Most should get to Level 2 or 3. A handful should reach Level 4 or 5.
- Map roles to target levels based on what the role actually requires
- Identify the highest-leverage learners — people whose hour is most valuable, who will use the skills hardest
- Be honest about which employees will adopt and which will not. Most rollouts fail because they try to drag along the people who don’t want to come.
Use the team’s actual deliverables as training material
- Generic AI training using fictional case studies almost never sticks
- The training material should be the team’s own current projects, refactored as fluency exercises
- Each session ends with a piece of real work produced using new capabilities
- The before-and-after is a piece of work the team needed anyway
Capture what works as durable infrastructure
- Every effective prompt, skill, and workflow gets documented
- The team’s collective knowledge becomes a shared library that new hires inherit
- This library is the actual deliverable of a fluency program — the workshops are the means; the library is the asset
- Twelve months in, the library is worth more than the training that produced it
Office hours and feedback after the workshop
- Workshops alone produce knowledge. Coaching produces fluency.
- Weekly office hours where employees bring real work and get feedback on how they used the tools
- One-on-one coaching for the people targeted at Level 4 or 5
- The compounding curve starts to bend around month 3
What changed in the work itself
- Before-and-after on specific deliverables: how long, how good, how repeatable
- Adoption metrics by level: who is using what, how often, on what
- Business outcomes the program was supposed to produce: throughput, quality, hiring deferral, project speed
- Honest assessment of who graduated to higher levels and who didn’t
Eighteen weeks is roughly the right shape for a serious fluency program in a company of fifty employees. It is significantly more work than a one-day workshop. It is dramatically less work than the alternative of having paid for two years of AI subscriptions while watching most of the value evaporate.
Where Purple AI fits in.
We work with companies in three configurations on AI fluency. None of them are off-the-shelf training programs. All of them are built around your team, your tools, and your actual work.
Workshops
Half-day and full-day sessions for teams ready to make a step-change in capability. Tailored to your stack, your industry, and the level your team is actually starting from. We come in, we teach against your real work, and your team leaves with skills they can use Monday morning. Best for teams between five and forty people.
Fluency programs
Multi-week engagements that include workshops, office hours, library-building, and coaching. The full curriculum described in Section Eight. Best for companies serious about treating AI as a core competency rather than a perk in the benefits package. Typical engagement runs eight to eighteen weeks.
Speaking
Keynotes, fireside chats, and executive briefings for conferences, leadership retreats, and industry events. The content is shaped to the audience, and the goal is always practical: people should leave with a meaningfully different understanding of what AI can do for their business and a clearer view of what to do next.
If you have read this far and recognized your own organization in any of it, the first conversation is free. We will spend thirty minutes talking through your team’s current AI maturity, where the highest-leverage gaps are, and what a workshop, program, or talk could look like for you.
Three ways to close the gap for your team.
Workshops
Half-day and full-day sessions taught against your team’s actual work.
Fluency programs
Multi-week engagements that build durable AI capability into the team.
Speaking
Keynotes, fireside chats, and executive briefings for conferences and leadership events.
Send us a few sentences on your team’s current AI maturity and what you’re trying to accomplish. We’ll come back with a candid read on which engagement fits and what the first month would look like.
