My previous post argued that most organizations think they're further along on the AI maturity progression than they actually are, and that the gap between self-perception and reality is itself one of the failure modes the conversation produces. That claim doesn't do anyone any good without a way to actually test it.
This post has a recommended framework for testing. It walks through each of the five stages and gives you three different lenses to evaluate where your organization sits: the observable signals you'd see if the org is actually at that stage, the diagnostic questions to probe with in conversations, and the anti-patterns that mimic the stage from the outside while not being it.
Before walking through the stages, one correction to the previous post. I called stage two "experimentation," and I was wrong to. Experimentation implies a specific kind of rigor: a hypothesis, a method, an outcome you can evaluate against the hypothesis. That's not what's actually happening at that stage in most organizations. What's happening is closer to people picking up a new tool and seeing what it does, building comfort with it, getting a feel for the shape of its capabilities. That's tinkering, and if I'm being honest, is best described as exploration. It's not experimentation. Calling it experimentation lends the activity a scientific seriousness it doesn't earn, and that inflation is itself a small example of the self-perception problem this whole series is trying to expose. I'm renaming the stage to "tinkering and exploration" going forward, and I'll use that naming for the rest of this post and the rest of the series.
My intent in this post is to give an IT or HR leader enough to do an honest self-assessment, not to provide a scoring rubric or a maturity certification. There aren't any of those that I trust, and I don't want to get stuck in the mindset of litigating a structure with weights that may work for my organization, but doesn't work well for others. My focus is on the question of whether the evidence in your own organization matches the stage you've been telling yourself you're at.
This kind of self-assessment may be uncomfortable to do honestly, because most of the answers are going to be lower than where you've been positioning your organization in conversations with leadership, peers, and yourself. That's not a sign that your organization is failing. It's a sign that the gap between self-perception and reality is real, and that you're now doing the work the conversation has mostly been avoiding. The discomfort is the point.
Stage one: awareness.
Observable signals. Leadership conversations include AI as a topic. The board has asked about it at least once. Strategic plans reference AI in general terms. Someone in the organization has been informally designated as "the AI person," even if that's not their actual title. The vocabulary of LLMs and generative AI has entered the room, even if the substance hasn't. Someone has probably attended a conference panel or read a McKinsey report on the topic in the last twelve months.
Diagnostic questions. Can three people in your leadership team articulate, in their own words and without vendor language, what an LLM actually does? Has the organization formally taken a position on AI use, even if the position is "we haven't decided yet"? Does the budget reflect any AI-related line items, or is it still entirely absent? If you asked your CFO what the organization's annual AI spend was, would they be able to give you a number?
Anti-patterns that mimic awareness. A leadership team that uses the vocabulary of AI without any operational grounding in what it does or doesn't actually do. This is more common than it should be. The signal that you're seeing the anti-pattern is when AI gets discussed at the leadership level in terms that would be at home in a vendor pitch deck, with no follow-up about what specific work it would change or what specific outcomes it would produce. Awareness without grounding looks like awareness from the inside but doesn't actually qualify, because the awareness is of AI-as-concept rather than AI-as-capability.
Most organizations are past this stage by 2026, with the caveat that "past it" sometimes means "still at it but talking like they're not."
Stage two: tinkering and exploration.
Observable signals. Individuals in the organization are using AI tools in their personal workflow, with or without formal sanction. Marketing is using something to draft copy. Engineers are using something to assist with code. HR is using something to summarize candidate notes. Finance is using something to format reports. The use is real and it's widespread, but it's individual, ad hoc, and largely invisible to the organization as a whole. There's probably some shadow IT happening on personal devices. The IT function may have rolled out a formal AI tool license, but actual usage is uneven and undocumented.
Diagnostic questions. If you asked five people across different departments what AI tools they use at work, would you get five different answers? Does the organization have visibility into what data is being put into which AI tools, by whom, for what purposes? Has anyone in the organization shared an AI-generated artifact (a draft, a summary, a piece of analysis) in a meeting in the last month? If you asked your team to describe their personal AI workflow, would they be able to, or would the answer be vague?
Anti-patterns that mimic this stage. An organization that has approved AI tools and counts the approval itself as evidence of progress. This is the failure mode where leadership believes the organization is doing the work of this stage because the licenses have been purchased and the policy has been written, when in fact the tools are sitting underused and the actual hands-on use is happening in shadow accounts that the formal program hasn't captured. The signal you're seeing this anti-pattern is when the metrics being tracked are licensing metrics (seats deployed, accounts provisioned) rather than usage metrics (what people are actually doing with the tools, what work has changed because of them). The licenses are infrastructure. The tinkering is the actual stage. Buying the infrastructure doesn't constitute being at the stage; only the activity does.
Most mid-market organizations are at this stage right now, whether they realize it or not. The tinkering is happening regardless of what the organization has decided to do about it. The question is whether the organization has visibility into what its own people are doing with these tools, or whether it's operating blind to its own activity.
Stage three: operational connection.
Observable signals. Specific operational problems have been identified, evaluated for AI applicability, and pursued through actual implementation. The pursuit might involve a vendor purchase, an internal build, or some combination, but in any case the AI use is no longer general, it's pointed at a defined operational outcome the organization has committed to producing. There are projects with timelines, budgets, sponsors, and success criteria. There are people whose job descriptions include some form of "make this AI-supported process work." Results, when they arrive, are evaluated against the original criteria, not against a vague sense that AI is good.
Diagnostic questions. Can you name three specific operational problems your organization has identified as AI candidates, the criteria you'll use to evaluate the AI solution against alternatives, and the timeline for the evaluation? Has anyone in the organization successfully retired a process or repurposed a role because an AI implementation made the existing work unnecessary? When AI is discussed in leadership meetings, are the discussions about specific outcomes you're pursuing, or about general capabilities you might pursue someday? Do you have a budget allocated to AI implementation that's distinct from your general technology budget, and is that budget being deployed against specific projects with named owners?
Anti-patterns that mimic operational connection. The pilot program that never exits pilot. This is the most common failure mode at this stage. The pilot gets approved, the vendor gets engaged, the technology gets deployed, and then the pilot sits in a state of indefinite evaluation, never quite producing the outcomes that would justify expanding it, never quite failing badly enough to be cancelled. The pilot becomes a kind of theater that lets the organization claim operational connection without doing the harder work of integrating the capability into actual operations or making the decision to walk away. If you have AI pilots that have been running for more than nine months without a clear go/no-go decision, the pilots are probably theater. A second anti-pattern at this stage is the "AI feature" that's actually a marketing repositioning of existing capability, the vendor relabeled their existing product as AI-driven, the organization adopted the relabeled product, and nothing operationally has actually changed.
A minority of mid-market organizations are at this stage. The ones that are usually got here because someone in IT or HR took the time to map AI capabilities to actual organizational needs and then drove the work of pursuing specific solutions for specific problems. The pursuit isn't easy, and most of the energy spent here ends up being on the upstream problem-definition work rather than on the AI implementation itself.
Stage four: systemic redesign.
Observable signals. Processes are being designed with AI capability as a foundational input, not bolted onto existing processes after the fact. Job descriptions are being rewritten to reflect what work the human is now doing versus what's been delegated to AI. Workflows that existed for years because of manual constraints are being questioned because the constraints have changed. The organization is asking different questions about how work should be done, not just which tools should be used to do the existing work. Performance metrics are being reconsidered because the unit of work has changed.
Diagnostic questions. Has any business process in your organization been redesigned from the ground up in the last twelve months specifically because AI made a different design possible? Have any roles been substantively redesigned, not eliminated, but redesigned, because the work the role does has fundamentally changed? Is your HR function actively involved in conversations about what jobs should look like under the new operational substrate, or are they being asked to handle workforce implications after the technology decisions have already been made? When AI capability is discussed in your organization, are the discussions framed as "how do we use this tool" or "how should the work look now that this capability exists"?
Anti-patterns that mimic systemic redesign. The "AI-powered" transformation initiative that's actually a process improvement initiative with AI vocabulary layered on top. The initiative may produce genuine improvements, but the improvements would have been available without AI and aren't a result of redesigning around AI capability. The signal you're seeing this anti-pattern is when the redesign work isn't actually contingent on AI being part of the substrate, if you removed the AI assumption from the redesign and the redesign still made sense, you weren't doing systemic redesign, you were doing process improvement. Another anti-pattern at this stage is the org chart redesign that's been announced but not implemented, where leadership has talked about restructuring roles around AI capability but the actual roles, reporting lines, and responsibilities haven't changed. Talking about systemic redesign is not the same thing as doing systemic redesign.
Very few organizations are operating at this stage. The ones that are have usually spent multiple years getting here, and the work of operating at this stage is ongoing rather than complete. The path through this stage is longer than most leadership teams initially expect, and most of the value of being at this stage shows up over years rather than quarters.
Stage five: transformational.
Observable signals. AI has been integrated into the actual texture of how work gets done. Humans and AI are operating in genuine partnership toward outcomes neither could achieve alone. The work doesn't look like the work used to look. The roles don't look like the roles used to look. New categories of work have emerged that didn't exist before, and the organization has staffed and developed people into those roles deliberately. The organization can credibly say that the outcomes it's now producing are different in kind from what was possible before, and the evidence for this claim is visible in operational data rather than just in rhetoric.
Diagnostic questions. Can you point to outcomes your organization is now producing that would have been categorically impossible without AI being part of the operating substrate? Are there roles on your team that didn't exist three years ago and that have clear career ladders, performance criteria, and salary bands? When the organization describes its work to outsiders, is AI part of how the work itself is described, or is AI described as a tool the organization uses to do its existing work? If you removed AI from your operations tomorrow, what specifically would you no longer be able to do, and how serious would the loss be?
Anti-patterns that mimic transformational use. Almost everything described in vendor pitches and thought leadership as "AI transformation." Most published case studies of AI transformation are stage three or stage four organizations describing themselves in stage five language. The signal you're seeing this anti-pattern is when the transformation narrative is built on impressive metrics about adoption, deployment, or productivity gains rather than on outcomes that are different in kind from what was possible before. Productivity gains are a stage three signal. New categories of outcome are a stage five signal. The two are often conflated. A second anti-pattern at this stage is the organization that's deeply integrated with a specific AI vendor's platform and confuses the integration with transformation. Integration with a platform is a stage three or four signal at best. Transformation requires that the organization's work itself has changed in ways that aren't reducible to which platform they're using.
This is the stage the AI conversation usually skips to in vendor pitches and thought leadership, and it's the stage almost no organization actually inhabits. The leap from stage four to stage five is the hardest part of the progression, and it's not a leap that tooling alone can produce. Honest self-assessment at this stage usually produces the answer "we're not actually here yet, and the things we thought made us transformational are actually stage three or four signals we'd been mislabeling."
A note on the meta-pattern.
If you walked through this honestly and your answer for each stage was lower than where you'd been positioning your organization, you've now confirmed the self-perception-versus-reality gap from the previous post. That's useful information. It's also uncomfortable information, and the natural instinct is to argue with it, find the asterisk that makes your organization the exception, or rationalize the gap as evidence that the framework is wrong rather than that the self-perception was.
I'd encourage sitting with the discomfort instead. The leaders I've watched move their organizations through the maturity progression most effectively are the ones who started by being honest about where they actually were, even when "actually were" was meaningfully behind where they'd been claiming to be. The dishonest version of self-assessment, where you talk your organization up the ladder without doing the work, produces organizations that flail when they try to act at stages they haven't actually reached. The honest version produces organizations that know what they're doing and don't waste resources pretending they're somewhere they aren't.
This kind of honest self-assessment is also where the IT and HR functions can do some of their most useful work together. Both functions have access to the operational evidence that would confirm or refute a leadership team's self-perception. Both functions are usually closer to the actual texture of the work than the leadership conversations let on. A CIO and a CHRO who walk through this exercise together, comparing their reads on each stage, will produce a more accurate diagnosis than either could produce alone. The next post in this series walks through what this looks like in practice, using a theoretical organization as a case study and tracing what each stage looks like from the inside.
— Chris