IT's AI shopping list is inverted: 46.5% want automation, 71% of their AI tools are invisible

Automation earns its keep against an inventory you can see. These respondents do not have one. Fourteen AI tools per enterprise against four that IT can name puts roughly 71% of the estate outside any system that would feed an agent ^[5][1]. Point automation at that and you get quick action on a partial map, which is worse than slow action on a complete one.

Which is why the top line reads differently than the coverage implies. 46.5% naming AI-driven automation the leading future investment is a plurality, so 53.5% of these directors and above put something else first ^[2][2]. The two categories behind it, vulnerability remediation and device visibility, are four tenths of a point apart ^[3][7]. That is a population wanting several things at near-equal intensity, reported as a population wanting one thing.

Bradley Chambers, who writes the Apple @ Work column that covered the report, lists what would have to exist first: device management, network filtering, education, and telemetry ^[8]. Only the last produces the data an AI feature would consume, and it is the hardest line item to sell upward. He also notes that none of this arrives with extra headcount ^[9], which sets the real ceiling on how much new tooling a team can absorb in a year.

The 5,000-engineer coding rollout is the same gap expressed in money. A full-year 2026 token budget gone in four months means consumption ran at three times plan, with eight months left to fund from something else ^[6][4]. That is not a discovery you make on a dashboard you never built.

Provenance is worth stating plainly. The percentages are Fleet's own report ^[1], and the column relaying them is exclusively sponsored by Mosyle, which sells an Apple device management platform and cites more than 45,000 customer organizations ^[10][11]. The write-up describes the respondent screen but gives no fielding window and no margin of error ^[5]. Vendor research is not disqualified by being vendor research, but an ordering this tight, with no published error bar, is thin ground for a product roadmap.

What the evidence actually supports is narrower and duller than "IT wants AI." It supports a market where the first sellable thing is discovery: finding the ten tools nobody logged, metering what they spend, attaching that to the device record. AI that manages Macs comes after the Macs report what is running on them. Anyone shipping the second without the first will be demoing to buyers whose data cannot answer the demo's questions, and 78% of their employees have already started without them ^[4].