Two-Sided AI Adoption
Pair top-down executive buy-in with a bottom-up tiger team of excited power users.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 90%
Most enterprise AI deployments fail because they are mandated top-down and stay divorced from how the actual work is done, leaving a workforce that knows it is 'supposed to' use AI but has nobody to learn from. The deployments that work combine executive buy-in (budget, tools, mandate) with genuine bottom-up adoption driven by a small internal team that explores capabilities, applies them to real workflows, and evangelizes through hackathons and knowledge-sharing.
Origin
Shared by Sherwin Wu from OpenAI's own internal rollout of Codex and from observing which enterprise customers succeed vs. get negative ROI on AI.
Core principles
- 01Top-down mandate alone produces a workforce that doesn't understand the technology
- 02Real gains come from employees who genuinely want to learn and evangelize
- 03The tiger team is best staffed with technical-adjacent people (support/ops leads, Excel wizards), not necessarily software engineers
- 04Every function's work is different, so last-mile intricacies must be worked out bottom-up
How to run it
- 1
Secure top-down buy-in
Get the C-suite to commit to becoming AI-first: buy the tools, provide support, and set the intent.
Watch out Buy-in and mandate alone is the anti-pattern — do not stop here.
- 2
Find or staff a tiger team
Identify (or hire) a small full-time internal team of people who genuinely light up around these tools and let them explore the full extent of the capabilities.
Pro tip Look for technical-adjacent people — a support or operations lead who doesn't code but loves the tools — not only software engineers, who many companies don't even have.
- 3
Apply to specific real workflows
Have the team map capabilities onto concrete, function-specific workflows rather than generic 'use AI' guidance, since each function's work is uniquely shaped.
- 4
Evangelize and knowledge-share
Let the team run hackathons, seminars, and knowledge-sharing sessions and publish best-practice documents so excitement and skill spread across the org.
Pro tip Treat these AI power users as your top performers and empower them the same way — hand them the exploration and let them set the standards everyone else adopts.
In the wild
OpenAI always wanted to be AI-centric, but adoption only really took off when Codex let actual employees apply it to their own work; a group of engineers who went deep on best practices then shared documents and ran knowledge-sharing sessions to elevate everyone.
→ 95% of engineers use Codex daily and Codex reviews 100% of PRs.
Common mistakes
Pure top-down mandate
An exec mandate with no bottom-up adoption leaves employees who know they're supposed to use AI (even on their performance review) but have no idea what to do and no peers to learn from, producing negative-ROI deployments and resentment that AI is being forced on them.
Staffing the team only with software engineers
Many companies have few or no software engineers, and the people who get most excited are often technical-adjacent non-coders; restricting the team to engineers misses the best evangelists and stalls adoption.
Is it for you?
Best for
Non-tech enterprise leaders rolling out AI tools across a workforce that isn't AI-native
Not ideal for
A tiny already-AI-fluent startup where everyone is a power user and no evangelism gap exists
From the transcript
“have a combination of both top down buyin”
“find or maybe even staff a full-time team internally that is this kind of tiger team”
“it was like an exact mandate and it's extremely top down and is very divorced from what the actual work looks like”
“find the high performers in AI adoption and empower them”
From the episode
“Engineers are becoming sorcerers”
The future of software development with OpenAI’s Sherwin Wu