Steven LafortuneUsing AI and getting mature with AI are two different kinds of progress. In the Devolutions 2026 AI Maturity Benchmark Report, 92% of organizations are already using AI. Only 8% said they are not. The useful question is what to do next, and that depends on where you stand.
We grouped 616 IT professionals by a self-reported score from 0 to 100:
- Early-stage (0 to 40): 267 organizations, 43% of the sample
- AI Adopters (41 to 60): 255 organizations, 41%
- Advanced (61 to 100): 94 organizations, 15%
This is the practical companion to the report announcement. The lists below come from the report’s five recommendations: name an owner, build enablement around the tools, let teams experiment before you standardize, measure outcomes, and let governance grow as autonomy grows. What changes by stage is which of those is the bottleneck.
One limit, stated in the report itself: these are behaviors that show up alongside higher maturity. A single practice does not cause an organization to become Advanced. The patterns are still concrete enough to choose a next step.
Early-stage: give AI a name, some time, and a narrow target
Early-stage organizations are already in the work. On average they use AI in about 2 of the 9 areas we tracked, most often content creation (48%) and software development (38%). What is still thin is the organization around those experiments.
Only 14% say leadership actively promotes and invests in AI. About 60% report at least some leadership support. Half have no clear owner. Formal KPI tracking is almost absent, at 0.4%. When people explain why a task they want to automate still is not automated, the top reasons are lack of time or resources (27%), security or compliance concerns (19%), and lack of internal expertise (16%).
1. Name one person who can answer three questions
Where is AI being used, why, and under what rules? Across the full survey, active leadership support was far more common when a role or team was explicitly accountable for AI: 77%, compared with 8% when nobody was accountable.
That person does not need a new department. Among AI Adopters, informal ownership inside IT is the bridge most teams are standing on (34%, versus 22% at Early-stage and 7% at Advanced). Give an existing IT lead the job of collecting what is already happening.
2. Ask leadership for protected learning time
About 60% of Early-stage organizations already have some leadership buy-in. The gap is making that buy-in useful. One Early-stage respondent said leaders were offering AI-engineer training even though most of the IT staff are not data engineers. Others described the same pile-up: thin expertise, worry about data privacy, and no time in the day to learn the tools.
Ask for training that matches the jobs people actually have, and for a few hours that are reserved for practice. The report’s leadership takeaway fits this stage exactly. Leaders do not need every answer. People need confidence and room to learn.
3. Make the use cases you already have safe to repeat
Breadth can wait. Service desk support is in place at 14% of Early-stage organizations, and identity and access management at 3%. Content creation and software development are where this group already works.
Security or compliance is the second-most cited reason a wanted automation has not happened. A one-page rule does more here than another license: which data may go into which tool, and who reviews the output before anyone trusts it.
AI Adopters: turn attention into a destination
This is the midpoint, and several patterns bend here. Active leadership support rises to 49%, and about 95% report at least some support. Informal IT ownership peaks at 34%. IT leadership drives AI in 38% of these organizations, and 15% have a dedicated AI role or team. Together, that clearer ownership is in place for 53%, versus 21% at Early-stage.
The footprint is wider: nearly 4 of the 9 areas, on average. The friction moves with it. Lack of internal expertise is still high (19%). Work that feels too complex or too dependent on context jumps from 5% at Early-stage to 12%. Formal KPI tracking reaches 9%.
1. Put the outcome in one sentence
Attention from leadership is much more common here. A clear destination is the piece that still slips. The report describes this stage as momentum without a shared picture of what “done” looks like. One AI Adopter respondent had expected the win to show up as cost savings. Time savings showed up instead, and they were much larger.
Pick the result you will actually look for. Time saved, documentation quality, and how fast incidents get resolved are the kinds of indicators the report treats as practical. Cost can stay on the list. It is often the slower one to appear.
2. Give the informal owner a mandate, and a review rule
IT is already holding a lot of this together. The pattern that shows up more often at the next stage is deliberate responsibility: IT leadership, or a small dedicated role, plus rules people can follow. An AI Adopter respondent described the practical version: governance, security and IP protection, training, and a policy for when tools can be used and how AI-generated output gets reviewed for accuracy, quality, and compliance.
Write that policy for the tools you already allow. The report also recommends letting teams compare a controlled set of tools before you standardize on one. Locking in a single stack too early freezes the learning.
3. Name what the next workflow depends on
On average, AI Adopters are coming up on a fifth area of use. The report’s scale takeaway is to identify, before that next workflow goes live, which systems it must connect to, what context it needs, and who has time to support it.
This is also the stage where AI agents start to look real. Exploration is still common, and pilots become much more visible. Keep a pilot inside one workflow, with a person who reviews the result. Security remains the leading concern about AI agents in every maturity group, including this one.
Advanced: focus the lead you already have
At this end of the curve, the basics are mostly in place. 91% say leadership actively promotes and invests in AI, and at least some leadership support reaches 100%. A dedicated AI role or team exists in 55% of organizations. IT leadership drives AI in another 34%. Only 2% have no clear owner.
The footprint is wide: more than 6 of the 9 areas, and 3.8 AI tools on average, versus 1.9 at Early-stage. 47% build or customize AI internally. 47% track benefits with formal KPIs. More than 4 in 5 are already piloting AI agents or running them in production.
The constraint changes shape. Lack of time or resources is the top reason a wanted automation still has not happened, at 35%, higher than in either of the other groups. Lack of expertise falls to 4%. Integration challenges rise to 8%.
1. Point leadership at fewer bets
Whether leaders will engage is settled for this group. The report’s outlook is that the open question becomes where that attention goes when use cases compete. Nearly half of Advanced organizations already track benefits formally. Expand the uses that improve time, quality, resolution speed, throughput, or risk. Pause the ones that only add activity.
Competitiveness also becomes a more common reason to invest, rising from 17% at Early-stage to 38% here. Operational efficiency is still the leading driver across the full sample (39%). Proof is what keeps a competitive bet honest.
2. Budget the integration work
Respondents further along describe the hard part as everything around the model: legacy systems, data governance, APIs, identity, permissions, audit, and the hours to do that carefully. Expertise is rarely the blocker anymore. Capacity is.
If a desired automation is stuck, staff the connective work. Adding another tool on top of an average of 3.8 is the slower path.
3. Write the human line before you add more autonomy
Agents are already in pilot or production for most of this group. Greater maturity does not retire the worries. Security vulnerabilities remain the leading concern about AI agents, including among Advanced organizations. Loss of human control ranks second overall (17%) and stays near the top at every stage. Advanced organizations are more likely to report no significant concerns (15%, versus about 5% in the earlier groups), and less likely to point at a lack of transparency.
Put three lines in writing: which actions still need a person to approve them, how outputs are checked, and who is accountable when something goes wrong. Let AI take more of the analysis and the routine execution. Keep context, oversight, and responsibility for the outcome with people.
The move that fits your stage
- Early-stage: name an owner, and protect time to learn on the use cases you already have.
- AI Adopters: write the outcome and the review rule before the next workflow goes live.
- Advanced: fund the integration work, and draw the line where a human still decides.
You can start with the move for the stage you are actually in. Our CEO, David Hervieux, closed the report on the same idea:
You do not need a perfect AI strategy before you start. What you need is a clear way to learn. Give people room to experiment, make sure someone is responsible for how AI is used, and pay attention to what actually improves the work.
Download the report. It is a free PDF, with no sign-up.
We are also walking through the findings live on October 7, 2026. Register for the webinar if you want the discussion, not only the charts.
Questions, or a story from your own stage? Leave a comment below.