The state of AI in IT

Hard data from 300 senior IT decision-makers on trust, governance, adoption and the pain points that keep growing.

Source

Survey conducted by GLG among 300 senior IT decision-makers, across North America, Europe, and APAC, and skew senior: director-level IT, CIO, and VP IT, plus CFO, COO, CISO, and compliance leaders across sectors at companies of 500 to 5,000 employees, with 45% running IT budgets over $50M.

The trust ladder

IT will hand an AI agent the low-stakes, reversible work and keep a hand on everything that can break the fleet. Trust tracks blast radius, not capability.

Would you let an AI agent do this?

The ladder. Share who would let an AI agent do each action unsupervised ("do it"), ranked high to low. Base: 300 for every row.

Bar chart: share of 300 IT leaders who would let an AI agent do each action unsupervised, ask for approval first, or never allow it. Reroute or escalate a ticket based on SLA risk 72.0% do it; auto-resolve a Tier-1 ticket 63.7%; quarantine a non-compliant endpoint 56.7%; re-apply configuration after drift 44.0%; deprovision access on an HRIS terminate trigger 40.3%; provision access on an HRIS start trigger 35.3%; lock or wipe an idle device 29.0%; generate and submit an audit evidence package 24.3%; patch third-party apps when a CVE drops 22.3%; deploy OS patches on a Friday night 20.0%.

The enforcement gap

Teams are drafting AI policies faster than they can put anything in place to make those rules stick. Companies have written the rules, but haven't enforced them.

46.0%

have a documented AI governance policy with stakeholder sign-off.

18.4%

Only 18.4% actually enforce it in software: 12.7% "tooled" plus 5.7% "continuous."

31.3%

are "ad-hoc": guidelines exist but are not enforced. 4.3% have not started the process yet.

16.0%

Only 16.0% are very or fully confident they could list every AI tool, copilot, and agent in use, including personal-account and browser-based, within 48 hours. 43.5% are not or only slightly confident.

22.7%

Only 22.7% would let an AI system auto-revoke an AI agent's access after an unauthorized action, without human approval. 68% want to approve first; 9.3% say never.

Bar chart: Govern is the capability IT rates most important and is least satisfied with. 77% say it matters, 26% are happy with how it works today, a 50-point gap.

Adoption reality

AI in IT is past the experiment, but the pilot-to-production wall is real.

52.7%

run AI in production today; 30% are in pilot; 5% still exploring.

Pilot outcomes

43.0% are still piloting, 37.7% scaled to production, 11.0% stalled mid-pilot, 3.7% killed before completion. ~15% of pilots stall or die.

Tenure

39.7% have invested in AI for 1–2 years, 33.0% for 2–4 years. This is not a first rodeo.

Timeline

41.3% expect AI to materially change how their team operates within 12–24 months; another 30.3% inside 6–12 months.

Readiness

Self-rating: executive sponsorship 28.3%, but team skills just 1.3% and data quality 3.7%. The buy-in is there; the skills and data are not.

The pain point

The pain is governance, integration, and audit, on top of a stack that keeps growing.

Top pain points today

Bar chart of top IT pain points today: AI-pilot governance 60.7%, tools don't talk 46.0%, audit prep eats my month 40.3%, tickets pile up faster than we close them 40.0%, shadow IT 40.0%.

17.0%

What hurts most this quarter: AI-pilot governance leads at 17.0%, ahead of integration 13.0% and patching 11.3%.

52.7%

Tool sprawl: 52.7% log into 7 or more separate IT tools in a typical week.

The ideal scenario

What would you stop doing if the perfect platform arrived tomorrow?

Bar chart of what IT would stop doing if the perfect platform arrived tomorrow: manual ticket triage 70.7%, spreadsheet IT reporting 62.0%, manual on/offboarding 61.0%, manual patching 58.0%.

Read more stories

1/6

The Rethink: Sophie Danby and William Goddard

Pull up a chair, not a slide deck

IT industry experts Sophie Danby and William Goddard spent years noticing that the best part of any industry event was the bit that happened after it ended. Now, with SuperOps as founding sponsor, they're building a whole event series around it.

You can't steer a stopped bike

As AI takes over the routine work that once defined whole jobs, there's unease everywhere. Ryan Schmierer talks about why the tools aren't the threat, why the real danger is freezing up, and what it takes to start moving.

Carrie Green, CEO, The CanIT Collective: If there's a table you want to be at, and there's no seats, grab a chair or bring your own

Closing the distance for Canada's MSPs

Carrie Green spent a career learning that community, not the next tool, is what moves an MSP forward. Now she builds it for a living.

Amy Babinchak with her book, 20 Questions Every Owner Asks Before Selling Their MSP

Amy Babinchak on exits, AI, and an industry growing up

After 23 years running her own MSP, Amy Babinchak coaches owners through the sale. On her new book, AI-driven early exits, and why the industry must grow up.

Trust fall

What should IT leaders put in place before rolling out agentic AI? We put one question to 13 OGs of IT.

In Conversation: Arvind Parthiban and Dor Eisner at a podcast table

Why security and operations can no longer run apart

For a long time, IT operations and cybersecurity ran on separate tracks. Different tools, different teams, different conversations. AI has collapsed that separation.