Key takeaways
- Four locks keep AI value stuck: no one can see the real work, and people aren’t allowed, able or rewarded to build fixes.
- Training, tools, consultants and mandates each work on one lock. The locks only open together.
- Shadow AI (unapproved tools people bring to work) is not fringe behavior. It is what unmet demand looks like.
Four locks
AI made building fixes far cheaper, but the payoff is conditional: it arrives only where the people who know the work can build. In most companies, four structural locks hold that condition back. Each makes sense to the people inside it, which is why pep talks don’t open them.
Can’t see it
The real process lives in heads, inboxes, spreadsheets and workarounds, not in any system. Leaders cannot see it and vendors cannot design for it. Systems record transactions, not behavior; consultants interview people, who describe what they remember rather than what they do. As long as the work is invisible, nobody can prioritize the right thing to fix.
Not allowed
Building has belonged to IT queues, vendors and consultants. The people who hold the customer history, the exception and the tested-and-failed way cannot turn what they know into working solutions. Governance often blocks without offering a useful path, because it was designed when building required specialists. The context dies in translation, and consultants leave with what they learned.
Not able
Directing AI well is a learned capability, and nobody owns building it in each person inside their real work. Courses don’t transfer and tools don’t teach. This is the learning gap people usually talk about. It is real, but it is still only one lock of four.
Not rewarded
Whoever streamlines their own job can fear looking replaceable. Utilization metrics reward hours, not results. Rollouts that feel like surveillance teach people to hide their work. Wins stay hidden, and the most capable people take their ideas somewhere they are allowed to use them.
Shadow AI is a signal
Microsoft’s 2024 Work Trend Index found that 78% of AI users bring their own AI tools to work. That is not fringe behavior. It is the majority’s workaround, and it is what unmet demand looks like when the approved path is not good enough.
Why the usual fixes fail
- Training aims at skills, and daily work doesn’t change.
- Tools aim at access, and behavior barely moves.
- Consultants deliver a project and leave with the context.
- Mandates raise compliance, not capability.
Each fix picks one lock. The locks only open together, which is why we believe opening them takes a new category of software, not a feature of an existing tool.
What opening them looks like
Visible
With permission, Lucy, an AI coach, sees how work really happens, so she can coach it. Leaders see team-level patterns.
Safe to allow
Every build uses approved tools, inside IT’s guardrails.
Able
Coaching on real work, at the moment of need.
Rewarded
When builders share a win, it’s measured and credited, so leaders can reward it.
Leaders still decide to allow and reward building. Lucy, an AI coach in the flow of work, makes that decision safe, cheap and visible. How that coach behaves is the subject of the next article.
Next: She asks before she answers.
Sources and notes
- Microsoft and LinkedIn, 2024 Work Trend Index: 78% of AI users bring their own AI tools to work.
- Boston Consulting Group, “AI at Work 2025: Momentum Builds, but Gaps Remain,” June 26, 2025 (more than 10,600 workers in 11 countries).
