Skip to main content
Back to the villageDONGGU’S FIELD NOTESLIBRARY / PERSPECTIVE
The library · perspectives

IDEAS GATHERED IN THE FIELD

When large-enterprise AX splits adoption from training, both sides fail.

A unified adoption-and-training model — how non-developers come to use AI safely.

Donghyun Kang2026-05-078 min read

I built the GS Caltex VOC AI dashboard on my own. The people reading the output every day were the CS team, who do not write code, and I watched from up close what they trusted and what they routed around. This piece starts there.

Four kinds of misalignment that keep showing up on the ground

Shadow use

A non-developer hits a wall inside the official tool, opens a personal AI account, and pastes the company doc in there.


The governance tool was installed, but no training arrived to change the behavior. A tool sitting there does not stop the workaround.

Trust collapse

A hallucinated answer from a non-developer-built app gets shared as input for an internal decision.


They were given the permission to build, but not the training to vet the output. Being able to build something and being able to judge what it produced are two different skills.

Permission vs. understanding mismatch

Permissions are laid out in five tiers, but training is one-shot and identical for everyone.


Either the people with the access lack the skill, or the people with the skill lack the access. The two tracks move on their own, and nobody ends up able to do the work at their tier.

Outsourced self-service

A non-developer starts building, hits a block, and ends up with the mentor building it for them.


The tool is in place, but when you get stuck the path is "hand it off" rather than "learn." AX ends up as outsourcing in a new shape.

This is not an accident; it is a product of structure.

The four mismatches above are not a willpower problem. They keep happening because adoption and training run on different teams, different timelines, and different KPIs.

01

Ownership split

The tool sits with IT or Digital; the learning sits with HR or L&D. The two teams answer to different KPIs, so the policy gate and the learning progress fall out of sync on top of the same user.

02

Timeline split

Adoption runs as a project (start, finish, close). Training runs as ops (repeat, refine, refresh). The day adoption wraps is the day training begins, so the policy set during adoption no longer matches the stage learners are at.

03

Measurement split

IT measures the adoption rate; HR measures the completion rate. If those two numbers never land in the same report, "adoption is a success but usage is a failure" has no owner.

Permission and learning belong on one track.

Map the two separated tracks 1:1 onto the same stages, and policy stops being a gate and starts being a learning signal.

The permission tiers and the learning track move together across the same five stages.
01

Policy is a learning signal, not a gate.

When a policy violation happens, it should trigger the next learning step, not a hard block. Blocking alone pushes people into shadow use.

02

Permission and training answer to the same KPI.

If the adoption rate and the completion rate are not reported side by side, neither has an owner. Put the two numbers on one dashboard.

03

Champion is a role, not a title.

A champion is not someone inside IT or HR — it is a role spread across every stage. Learning gets absorbed into ops when a peer teaches a peer.

04

Adoption and training share one timeline.

The end of the adoption project is not the start of the learning. The two run together across the same stages.

8 things to check inside your company tomorrow

  • Are the permission tiers mapped onto the same stages as the learning track?
  • When a policy is violated, do you only block, or does the next learning path actually fire?
  • Do the adoption rate and the completion rate land in the same report?
  • Is there a champion role at every permission tier?
  • Is hallucination-checking baked into the permission tier for apps built by non-developers?
  • Is the tool-adoption end date sitting on the same line as the learning start date?
  • Is there a written learning path for the moment someone gets stuck inside the official tool — the one that blocks shadow use?
  • Is the 90-day active-usage rate inside the KPIs you actually measure?

── About the author

Donghyun Kang — Forward Deployed Engineer at Wishket and organizer of Beyond Deploy, an FDE community. Solo-designed and shipped the GS Caltex VOC AI dashboard, and runs mentoring and advisory work to help non-developer employees use AI tools safely.

If you want to talk about unifying large-enterprise AX adoption and training.