There's a quiet truth about AI productivity that nobody in the rollout business wants to say out loud: most of the people using AI today aren't actually using it well. They've adopted the surface (they ask questions, they get answers, they paste) and they've stalled there. The economic gap isn't between people who use AI and people who don't. It's between people who use AI as a chatbot and people who use it as a working partner.
That gap is the entire reason Lucy, our AI coach, exists. AI does not transform business. People transform business with AI. Lucy is the missing link between the people doing the work and the AI tools that could revolutionize it. To understand where Lucy meets a team, you need a clearer picture of how organizations actually mature with AI, and where the value really lives.
The five-stage spine
We map AI maturity inside an organization across five stages: Trust → Personal value → Adoption → Capability → Scale. Each stage builds on the one before it, and each one is its own kind of work.
- 1Trust
- 2Personal value
- 3Adoption
- 4CapabilityThis article
- 5Scale
- Trust. Will my data be safe? Will I lose my job? Without this, nothing else lands.
- Personal value. The first time AI saves a person an hour on something they actually do. That's the moment the story changes.
- Adoption. Daily, repeated use. Not pilots. Not demos. Real workflows.
- Capability. Real skills. Universal AI fluency, then specialized depth where the work demands it.
- Scale. What used to be one person's edge becomes how the company operates.
Skip trust and your teams avoid the system. Skip personal value and adoption is just for show. Skip capability and wins don't compound. Skip scale and know-how stays with a few people: best practices never compound across teams. The stages are not a checklist: they are a sequence, and the sequence matters. (For the full framework, see How Lucy Works.)
This article is about the fourth stage: Capability. It is the stage where most economic value actually shows up, and the stage that almost every rollout under-invests in.
Universal AI skills: the foundation for roughly 70% of the work and value
This is where most of the economic value lives. And almost everyone overlooks it.
Universal AI skills are the muscle that turns AI from a curiosity into a working partner. You know how to write a prompt that gets useful output the first time. You know how to iterate when it doesn't. You can tell when the AI is hallucinating versus when it's right. You know which tool to reach for when. You know how to give the AI enough context for it to actually help you. None of this is domain-specific. All of it is learned by doing.
The research backs this up. The Brynjolfsson, Li, and Raymond NBER study of customer-service workers using generative AI found a 14% productivity gain on average, but a 34% gain for less-experienced workers. The same AI produced very different gains for different people. What people bring to the tool matters, not just the model.
Lucy decomposes this layer into ten universal capability areas:
- 01Prompting fundamentalsgetting useful output the first time
- 02Effective iterationgetting useful output when the first try misses
- 03Output evaluation and quality controlknowing when to trust what comes back
- 04Context managementgiving the AI what it needs to actually help
- 05AI-aware task decompositionbreaking work into pieces that AI can move
- 06LLM capabilities and limitationsunderstanding what language models can and cannot do
- 07Tool selection and orchestrationchoosing and combining the right AI tools
- 08Ethical use and risk awarenessrecognizing privacy, bias, IP, security, and transparency risks
- 09Collaborative AI worksharing prompts, reviewing outputs, and working with teammates
- 10AI adoption mindsetrecognizing when AI adds value and keeping critical thinking
These ten areas move a person from "AI is a thing I sometimes use" to "AI is a thing I work with." The universal foundation is the baseline, not a final tier. From there Lucy can help someone build tool fluency, then climb toward solution architecture and agentic workflows (combining tools and agents into complete solutions); specialized and domain depth follows the person's role through that climb.
Role-shaped specialized and domain depth
Once the universal foundation exists, role-shaped specialized and domain depth multiplies it. AI stops being a generic chatbot and starts being a tool that fits your specific job.
If you're in sales, this is when AI starts handling the parts of customer prep that used to eat your morning. If you're an engineer, this is Cursor or Copilot in your editor, suggesting what you'd have written in half the time. If you're an analyst, this is natural-language data exploration with ChatGPT instead of staring at SQL. If you're in marketing, this is multimodal creation that compresses days of design iteration into hours.
Lucy keeps this depth role-shaped rather than treating it as a fixed list of domains. Examples include:
- Code generation
- Agent building and automation
- Presentation design
- Visual content creation
- Research synthesis
- Data analysis prompting
You don't need every specialization. You need the depth that matches your actual work. A salesperson with role-shaped depth can prepare faster, follow up more sharply, and keep the nuance from the meeting that closes the deal. As a rough model, universal skills cover about 70% of the work and value, and specialized and domain skills, applied just in time, cover the other 30%. It is not a score, a training-time estimate, or a share of people.
From foundation to solution architecture
Another way to see it is as a climb: Foundation → Tool fluency → Solution architecture and agentic workflows. Domain depth develops through that climb, shaped by the person's role and goals.
Foundation is the universal capability base. Tool fluency is choosing and using the right tool for each kind of work. Solution architecture and agentic workflows is the top: combining tools and agents, building in rules and checks, and making solutions reusable and easy to maintain. It is a capability people can grow into, not an optional tier reserved for a percentage of the population.
- Foundation The universal capability base
- Tool fluency Choosing and using the right tool for the task
- Solution architecture and agentic workflows Combining tools and agents, with rules and checks built in
Why most rollouts chase the demo and skip the muscle
The conventional AI rollout playbook is built around two beats: "give people the tools" and "show them what's possible" with a flashy vendor demo of someone, somewhere, building something amazing. The middle (the universal foundation, the role-shaped specialized depth) is treated as something that "just happens" once people have access.
It doesn't. The middle is where coaching matters most, where almost no rollout invests, and where actual capability is built. Companies spend billions on AI tools and consulting and training, and the muscle layer, the universal skills behind roughly 70% of the work and value, stays empty. The rollout looks impressive on a slide. The day-to-day barely changes.
Lucy exists to close that gap. Not by running another training course. Not by deploying another tool. By coaching people in the moment, on their actual work, until the universal capabilities become muscle memory, then helping them build role-shaped depth, tool fluency, and solution architecture and agentic workflows as their work calls for it.
What this means for you
If you're an individual reading this: figure out where you actually are. Be honest. If you're still treating AI like a faster Google, the next step isn't a fancier tool: it's the universal foundation. From there, build tool fluency and choose the specialized and domain depth that matches your work; solution architecture and agentic workflows are the top of the progression when your goals call for it.
If you're rolling out AI for a team: stop optimizing for the demo and start optimizing for capability. Build the universal foundation, support role-shaped specialized and domain depth, and make tool fluency and solution architecture and agentic workflows available to the people whose work calls for them. The goal is capability that fits the work, not a population target.
If you're building the AI strategy for your company: the gap between "we have AI tools" and "we have AI capability" is exactly the gap described here. It sits in the Capability stage, between adoption and scale, and closing it is not optional anymore. The companies that close it own the next decade. The ones that don't will keep wondering why their pilot didn't deliver.
Travis Sheppard is the founder of Lucy Labs. He's spent twenty years leading the kind of change that makes companies actually scale, and is now building Lucy, an AI coach, to put that know-how behind every person learning to work with AI.
