Work sampleAI enablement

Running an AI adoption program people actually use

Most AI training produces a good afternoon and no change in how work gets done. I run enablement like a product instead: a framework leaders reuse, sessions that end with something people keep using, a redesign when the real blocker shows up, and adoption counted instead of attendance.

Samuel Bates, AI Solutions Manager

Start with a framework people can reuse

The first workshop was for senior leaders at an internal leadership summit. I built it around a simple four-step framework, so the takeaway was a way of working rather than a list of prompts. It has since reached 800+ people across the consumer and merchant organizations, and it generated follow-on AI projects in several of the teams that attended.

Every session ends with something people keep using

The org-wide series for an operations organization runs sessions of roughly 70 to 100+ practitioners. Each one ships a reusable artifact, such as a skill, a template or a shared team setup, that people keep using after the call ends. Session four was a builder showcase: people from the business demoed production work they had shipped themselves, and I checked each merge record before describing their work to their executives. It is the same check I run as primary PR approver for 5 builders, a role I have held since September 2025.

Redesign around the real blocker

Midway through, a domain owner who had onboarded a group of non-engineers showed me that access and hands-on work in the same session deliver neither. One planned session became four parts:

  • Part 1: get access, and learn five words: repository, branch, commit, pull request, merge.
  • Part 2: work inside a team repository.
  • Part 3: read an existing repository.
  • Part 4: contribute code, opt-in and with a sponsor.

The line that unlocks a non-technical room turned out to be reversibility, not sophistication: everything here can be undone, and you are not an admin.

Recover in front of the room

Minutes after Part 1, the first attendee hit a wall in the access flow. I had tested it as a member with elevated access, which hid a step every newcomer needs. By that afternoon the path was corrected to seven steps with screenshots, and the recap led with the correction and asked everyone to reply with the step they reached. Every access request that day was granted by the next morning.

Count adoption, not attendance

  • A binary conversion artifact per part: Part 1 counts who reached the last access step. Parts 2 to 4, still ahead, will count a first pull request and then a merged change.
  • A survey the leaders run themselves, so the program is steered by the people who own the outcome.
  • A builder showcase, because the strongest adoption signal is someone outside engineering shipping work to production.