Driving AI Adoption Through Hard Constraints

Core principles

• 01Manifestos and executive decrees don't move the needle; hard constraints do, because they force behavior change.
• 02Every transformation splits people into three groups — catalysts, converts, and anchors — each needing a different response.
• 03AI-native is a fundamental culture change, not a tooling change, and cultures thrive on density, so misaligned holdouts dilute it.
• 04Accelerating only one part of a system (e.g. engineering) just moves the bottleneck; total output rises only when the slowest part is fixed.
• 05Executives are usually badly disconnected from actual on-the-ground adoption.

How to run it

1. 1

   Impose hard constraints

   Replace vague encouragement with binding constraints that make the old way impossible. Examples: benchmark team sizes against peers and cap each function at some fraction of that size; forbid new headcount until the team proves AI can't do the job; refuse to review a PRD unless it arrives with three prototypes.

   Pro tip Pair constraints with the softer scaffolding — a named owner, clear budget, incentives in career ladders and performance reviews — but know the constraint is what actually forces adoption.

   Watch out Communication, ownership, and rewards alone won't produce change without the hard constraint underneath them.

2. 2

   Triage people into catalysts, converts, and anchors

   Identify your catalysts (self-driven experimenters leading the charge), your converts (willing but needing structure, permission, a clear plan), and your anchors (dragging feet, silently creating friction). Each group needs different handling.

   Pro tip Give converts exactly what they need to move: decrees, permission, budget, and a clear outline — they'll adapt, they just won't self-start.

   Watch out Converts needing structure is not a character flaw; don't treat them as anchors.

3. 3

   Set a hard deadline for anchors

   Rather than working passively with anchors indefinitely, the companies furthest along set a hard date: make the transformation by X or we define a plan to exit you. Because AI-native is a culture change and culture thrives on density, you can't leave 20-30% operating in a different culture.

   Watch out Fewer than 10% of companies take this hard stance — but those are the ones getting the most adoption and results. It feels harsh; the CEO lens is that density is required for the whole company to succeed.

4. 4

   Get to the ground floor and measure real adoption

   Assume you (as an executive) are disconnected from reality — decrees rarely produce natural adoption. Talk to end users directly and measure actual usage. When a promising experiment stalls in the org, the CEO often has no idea until they stumble into it.

   Pro tip Best-in-class companies like Shopify measure actual adoption and usage to stay close to the ground.

   Watch out ~90% of the time, when you ask an end user how many teammates use a new tool, the answer is 'me and one other person' — the manifesto's teeth are missing.

5. 5

   Find and attack the slowest part of the system

   Treat AI adoption as a system whose output is capped by its slowest part. Diagnose which part is throttling — often IT, legal, or procurement setting the pace, or PMs becoming the bottleneck once engineers speed up — and attack it ruthlessly instead of over-accelerating an already-fast function.

   Pro tip Product is the output of design + PM + engineering; speeding only engineering (biggest, most expensive headcount) just relocates the bottleneck and doesn't raise shipped-product output.

In the wild

Common mistakes

Is it for you?

From the transcript