Practical AI

Mike Lewis on the non-technical builder AI teams overlook

Mike Lewis· Chief AI Architect at TiER1 Performance
·~55 min·English·Changelog
AgentsBusiness StrategyAI Company
TL;DR

AI adoption isn't a ladder to climb; the payoff comes from finding the one person who understands the work cold and letting them build tools that make the real problem vanish, measurably.

01Core Mental Model

Ignore Everything Outside Your Sphere

Mike's trick for cutting AI noise is a hard rule: if a development doesn't touch how he works with clients day to day, he ignores it, no matter how big it looks at the edge of the field.

I do not concern myself with things outside of my sphere of influence.

— Mike Lewis, Practical AI
Key Insight
The filter is only as good as the person running it. Mike can apply it cleanly because he entered AI as an outsider, a portrait painter with no industry dogma, so he has no reputation to defend and no pet theory to protect. For most leaders the hard part isn't the rule; it's admitting which of their current priorities sit outside their real sphere of influence.
02The Framework

L0 to L3 Is a Map, Not a Ladder

He borrows the L0-to-L3 labels for AI adoption but rejects the ranking they imply: the levels are four different ways of working, not rungs everyone has to climb.

I don't think an L1 is better than an L0 and I don't think an L2 is better than an L1

— Mike Lewis, Practical AI
Key Insight
Flattening the levels is the whole move. The moment L0-to-L3 is read as a ranking, every program becomes a forced march to L3, which is exactly the mandate Mike says backfires. Treating them as roles instead lets a company staff for fit, not for compliance.
03Where To Focus

The Non-Technical Builder

The valuable move isn't turning users into power users; it's finding the rare non-technical builder who can turn the company's own way of working into durable tools.

I sort of run around evangelizing like this is actually where you need to focus your energy.

— Mike Lewis, Practical AI
Key Insight
The bottleneck was never the model; it was translation. An L1 consumes AI; an L2 encodes the company's judgment into something reusable. That's why the jump is rare, it demands someone who understands the work deeply enough to teach a machine to imitate it.
04Adoption

Don't Drag the L0s Uphill

Telling people to learn AI or lose their job actually slows adoption, so Mike doesn't try to convert the disengaged; he invites the willing and leaves the rest alone.

threat framing actually does slow down adoption.

— Mike Lewis, Practical AI
Key Insight
Mandates feel like progress because they're measurable, a deadline, a compliance rate. Invitation isn't. Mike's claim is that the measurable-looking path is the slower one, which is a hard sell to an executive who needs a number for the board.
05The L0 Myth

Five Kinds of Not Using AI

Not using AI hides five very different people, and the biggest group, the roughly 60% who found the output not good enough, is the one executives should listen to rather than dismiss.

This is where I think every executive's ears should turn on. They should lean forward in their seat and they should realize, do not dismiss these people.

— Mike Lewis, Practical AI
Key Insight
Three of the buckets are noise Mike pushes past; the last two, the job-fearful and the quality-disappointed, are where the lesson is. Lumping every non-user into resistant-to-change throws away the single most useful piece of product feedback a company has, from the roughly 60% who tried AI and found it wanting.
06Team Design

One Builder Per Team

Companies should pick their one builder per team for how deeply they understand the work, not for AI enthusiasm, and keep a technical expert within arm's reach to make what they build scale.

If you have two L2s on a team, they are not necessarily as good as just an L2, a few L1s, some L0's

— Mike Lewis, Practical AI
Key Insight
The scarce resource isn't AI talent; it's people who hold the company's tacit standards. Staffing one builder per team, backed by an expert, treats those people as the constraint to protect, the opposite of handing the keys to whoever is loudest about AI.
07Measurable Value

The $4M Job That Became a Claude Skill

Value shows up not in tokens spent but in work erased: an L2 looked at a $4 million document-conversion project and built a Claude skill for it in about three hours.

you're not going to find out the value of this investment based on how many tokens people are spending

— Mike Lewis, Practical AI
Key Insight
Token spend measures activity; the $4M-to-3-hours story measures a decision avoided. The deeper win is quieter: once the L2 aligned the model, the company's tribal knowledge became documented process, insurance against the hit-by-a-bus expert walking out the door.
08On Job Fear

Your Job Was Always Going to Change

Your job was always going to change with or without AI, so the real risk isn't being replaced by a machine; it's being passed by a colleague who adapts before you do.

you might lose your job, but it won't be because the AI replaced you. because Joe who's willing to use AI will replace you, you know.

— Mike Lewis, Practical AI
Key Insight
Notice the sleight of hand: Mike agrees your job is disappearing, then redefines disappearing as changing, which has always been true. The threat he takes seriously isn't the machine; it's the colleague who adapts first. That reframes AI from an existential risk into ordinary competitive pressure.