AI Transformation: Stop Staring at L0 — the Real Value Is the Non-Technical Builder
Companies spend their energy dragging non-AI users in, but the real leverage is the person on every team who understands the business and can turn tacit knowledge into working tools.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
L0 is not one kind of person, it's five
Mike breaks "people who don't use AI" into five groups: performative users (logged in, sent one chat, went back to work), the uninterested, the too-busy, the afraid of losing their jobs, and the disappointed with quality. He stresses that the last two can't be ignored — the afraid often mistakenly believe AI will learn their job and then replace them, and the quality-disappointed make up about 60% of the people he's encountered, and often have legitimate reasons. He met a marketing copywriter who said flat out "humans do it better," and she was right, but no AI expert ever sat down and listened to her.
— Mike LewisThreat-based rollout actually slows adoption down
Mike read 200 pages of research literature, much of it about learning itself, not about AI. The conclusion: threat framing slows adoption. If you back people into a corner and say "learn this or there's no place for you here," they learn far worse than if you say "let's take a look together, if you're interested." He concedes some roles genuinely need AI-proficient people, but thinks that in ten years, looking back, we'll find that not every role does. He gives the example of his wife: she's a nurse, and AI in her work currently just gets in the way.
— Mike LewisThe PC and AI adoption curves almost overlap
Mike saw a chart in his research: the adoption curve when PCs first entered the office and the AI adoption curve are nearly identical over the same time span. His read is that people panicked back then too about colleagues not learning to use personal computers, but it solved itself, because everyone ended up with one. Same with AI — people went home and signed up for ChatGPT themselves, because it's free. But he says this won't happen with CRM or SmartSheet — nobody runs home to open their own SmartSheet account.
— Mike LewisL2 selection isn't about technical skill, it's about company DNA
In Tier One's process for evaluating L2 candidates, it almost never starts with assessing technical ability. What they care about most is company DNA: whether this person understands the work in some "uncanny" way, whether they know which cell of the spreadsheet is worth fighting for, whether they know how the company wants work delivered. Mike says the people currently in the driver's seat are the "AI excited" — the ones shouting about nano banana and methos class models in the hallway — but they tend to be easily distracted, not focused enough, and don't know how to align an agent's output with the standard the company's best SME would sign off on.
— Mike LewisEvery team needs just one L2 — more makes it worse
Mike says plainly: one L2 per team. They found this isn't one-plus-one-equals-two — two L2s on the same team aren't necessarily better than one L2 plus a few L1s and a few L0s; it's a case of too many cooks in the kitchen. And every L2 needs an L3 within arm's reach to review the work: an L2 can make output uncanny, but only an L3 can make it scalable, durable, compliant, and legal. This structure has another benefit: you don't need to buy as many expensive licenses, and training resources can be concentrated on qualified candidates.
— Mike LewisL2s turn tacit knowledge into documented process
Mike says a good L2 doesn't just build tools that handle repeatable work — they're also converting tacit knowledge into documented process. Because once an AI model is aligned, you have documentation — maybe it's in code, but it is documented. This maps directly onto the big problem every industry has right now: an aging workforce, and the person who "can't get hit by a bus" leaves and the plant stops. The tools an L2 builds are themselves a defense against that knowledge loss.
— Mike LewisOne Claude skill killed $4 million of work
Mike tells a client story: one of the world's top four pharma companies had thousands of documents that needed converting from one format to another. It was a $4 million job, they knew exactly what each conversion cost, and it required SME involvement. An L2 took one look and said "this feels like a Claude skill," and built it in three hours. Drag a document in, and what comes out is almost exactly what the end of the process wants. Everyone's jaw hit the table: this was supposed to cost $4 million, and now with Claude they'd already paid for, it can run a thousand documents concurrently and finish in a few hours.
— Mike LewisDon't chase faster — chase new work you couldn't do before
Mike says "faster isn't necessarily better," especially when a system has many parts — you're just creating whiplash and anxiety for everything around it, because the other parts aren't ready to go that fast. Cheaper sometimes doesn't matter either; that's not what a lot of companies are even talking about. What he actually cares about is "newly emergent types of work": what can you now do with a language model plus a clever harness that the human brain was never good at before? For him that means rethinking the product, the department, or the function. He puts the whole "what will the future look like" question outside his sphere of control and stops worrying about it.
— Mike LewisIn their own words · checked verbatim
So threat framing actually does slow down adoption. So if you back someone into a corner and say, Learn this, or there's not room for you here, they will not learn it as well as if you came to them and said, Let's figure this out together if it interests you.
Mike Lewis11:45
It's terrifying. Let's back up. Let's back up. Name another technology that we would say that's true, but can you imagine you just go company wide? Hey, by the way, everyone, we want you to geek out about Smart Sheets and be great at Smart Sheets by the end of this year, or AWS, or just It's ridiculous.
Mike Lewis14:20
Right now, the wrong people are in the driver's seat. We call them the AI excited. The AI excited are the ones who are getting all the attention in the company. They're running down the hall screaming words like nano banana and methos class models, and everyone's just like, Well, they must be the person who should be building things
Mike Lewis25:40
It was a $4,000,000 job. They knew exactly how much it would cost to convert each one, and it requires SMEs. And an L2 took a look at it and said, This feels like a Claude skill.
Mike Lewis37:10
you're not gonna find out the value of this investment based on how many tokens people are spending or, you know, like, you're gonna find it when you start intentionally forming teams with a couple of strategic people in there who know how to spot an AI opportunity and just make the headache vanish.
Mike Lewis39:00
At tier one, we say, like, faster's not always better, especially when it's a it's it's a system with a whole bunch of parts, and you've just created whiplash and angst for everything around that fast piece.
Mike Lewis45:30
there is massive, massive value in the people who are willing to lean against the brick wall and push and push and push. And believe it or not, you'll start to find ways to use these tools that are hugely informative in spite of the fact that they're so burdened by necessary governance and, you know, rules and restrictions
Mike Lewis52:20
Figures
| Share of the L0 "quality-disappointed" group | about 60% | 21:50 |
| Time the L2 took to finish that project with a Claude skill | 3 hours | 37:40 |
| Pages of literature Mike read for his research | 200 pages | 8:50 |
| Number of L0 categories Mike found in his research | 5 | 16:40 |
Glossary
- L0/L1/L2/L3
- AI proficiency tiers: L0 doesn't use AI, L1 is a user who can't live without it, L2 is a non-technical builder, L3 can scale it across the enterprise.
- non technical builder
- Someone who understands the business but doesn't write code, and can turn tacit knowledge into working AI tools.
- Claude skill
- A reusable task package for Anthropic's Claude; drag in a document and it processes it along a preset workflow.
- super user habits index
- A 24-item habit checklist Mike built to identify the behavioral patterns of AI super users.
- solution archetypes framework
- A tool Mike developed early on to categorize enterprise AI deployment scenarios.
How to listen
Transformation leads pushing enterprise AI adoption, internal AI platform builders, and managers who want to know who should get the training budget.
The show intro and Midwest AI Summit promo from 0:00-3:00 can be skipped.