The AQ Maturity Model

People use AI personally, ad hoc. Value exists, but it lives and dies with whoever happens to be a power user.

• A few employees use AI on their own to draft emails, summarize documents, or debug code
• Usage is invisible to leadership — no one is tracking who uses what, or why it works
• Value is real but personal: it disappears the moment that employee is out, changes roles, or leaves

AQ1 organizations mistake individual enthusiasm for organizational capability. Because a few people are visibly excited about AI, leadership assumes the whole company is "adopting AI" — when three people are quietly good at it and everyone else hasn't changed how they work at all.This is a Capture problem: nothing that lone power-user knows has been written down anywhere. If they left tomorrow, AQ1 gets worse.

Start with visibility, not tooling. Get the handful of effective users to show their work — screen recordings, shared prompt libraries, informal office hours — so what's in their heads starts moving into other people's hands.

Teams start sharing prompts, templates, and workflows with each other. Useful patterns spread — slowly, informally.

• Teams maintain shared prompt libraries or internal docs of "things that work"
• New team members get pointed to existing patterns instead of starting from zero
• Workflows are still ad hoc — copy-pasted between people rather than built into any system

Team-level sharing creates the illusion of an organizational system, but it's still bounded by who happens to be in the same Slack channel. Two teams solving the identical problem in isolation is common here, and usually invisible until someone compares notes by accident.This is an Operationalization gap: knowledge is captured, but only within team boundaries — it hasn't become a system anyone outside that team can run.

Requires actively connecting teams — a shared knowledge repository, a cross-functional working group, or simply someone whose job is to notice duplicate effort and consolidate it.

AI agents execute cross-functional processes, not just single tasks. Knowledge starts moving across team boundaries.

• Agents pull information across systems — CRM, support tickets, internal docs — to execute a process end to end
• A question raised in support can trigger a workflow that updates product, engineering, and sales in the same motion
• Handoffs between teams start happening through defined interfaces — agents, shared data, structured workflows — instead of meetings and Slack pings

Connection without documentation is fragile. Agentic workflows at this level often depend on one engineer's mental model of how the pieces fit together — which means the "connected" system has a bus factor of one, just like the knowledge it's supposed to be operationalizing.This is a Governance & Trust gap as much as a Capture gap: cross-boundary execution has outrun cross-boundary validation.

The workflows themselves need to become the documented playbook — written down, versioned, and owned, not just running in production because someone built them once.

Institutional knowledge becomes the company's operating playbook — documented, structured, and executable by design, not by accident.

• New hires are onboarded against a documented playbook, not tribal knowledge passed down informally
• Decisions that used to require one specific senior person can be made correctly by referencing the operating system that captured their judgment
• AI agents execute the playbook directly — the documentation isn't just a reference for humans, it's structured well enough for agents to act on

Operational maturity can calcify into rigidity if the playbook isn't revisited. An organization can become excellent at executing yesterday's best practice while the market moves on — mistaking "we have a system" for "we have the right system."This is a Feedback gap: the playbook exists, but nothing routes new signal back into it, so it goes stale while looking complete.

Requires building the feedback loop — a structured way for execution results, human and agent, to flow back into the playbook and update it, rather than the playbook being a static artifact reviewed once a year.

Humans and agents continuously improve the organization's knowledge system together. The playbook gets better every week, not just once a year.

• Every execution — a closed support ticket, a completed sales cycle, an agent run — generates a signal that feeds back into the operating system
• This quarter's playbook is measurably different, and better, than last quarter's, and people can point to why
• Human judgment and agent execution improve each other in a visible loop: agents surface patterns humans hadn't noticed; humans correct edge cases agents get wrong, and that correction gets captured too

The risk at AQ5 isn't complacency — it's over-automating the feedback loop itself, trusting the system to self-improve without enough human judgment left in the loop to catch when it's optimizing for the wrong thing. High AQ doesn't mean ceding judgment; the two are supposed to scale together.This is a Governance gap resurfacing at the top of the model: the discipline that was optional at AQ1 becomes existential when the system is trusted to improve itself.

Sustaining this level is a discipline problem, not a technology problem. It requires treating knowledge capture and feedback loops as permanent infrastructure, not a project with an end date.