What Fleet surveyed
Fleet, an open-source device management vendor, published its Apple in the Enterprise report on October 7, 2026. It surveyed more than 250 enterprise IT practitioners who manage Apple devices. Channel Insider and Computerworld followed with their own coverage.
This is a vendor-run survey of self-reported answers. The public summaries don't give field dates, recruitment method, wording of the questions or margin of error. Read the numbers as a snapshot of admin sentiment. They are not measured incident rates, and they are not a census of all Mac administrators.
The headline numbers
AI is reaching production.
- 86% allow AI-written output to reach production devices after some form of review. Only 14% keep it out of production entirely.
- In the past 30 days, 68% used AI for troubleshooting or log analysis, 62% to write scripts and code, and 61% to author MDM profiles or configurations. These groups overlap, so the percentages shouldn't be added together.
- 57% regularly use an LLM assistant, ahead of the MDM console at 54%.
Safeguards are uneven.
- 36% review and test all AI-generated changes before deployment, and 28% combine human oversight with version control, according to Channel Insider.
- "Some form of review" is therefore a much looser standard than testing everything. Version control adds a separate thing: a record of what changed.
Recovery is slow.
- 69% can't roll back a bad configuration change within an hour, and 43% need more than a day. Another 26% recover the same day, but only with manual work.
- The summaries don't say how Fleet grouped the recovery-time answers, so don't treat these as neatly separate buckets.
- These figures measure how quickly admins say they can reverse a change. They don't count actual failures. They don't prove AI caused an incident, and they don't quantify downtime.
The Intune angle
Only 8% of respondents manage Apple devices exclusively. Most work across mixed environments. Channel Insider reports that Microsoft Intune was the most commonly cited primary Apple management platform at 36%. Jamf Pro followed at 20% and Omnissa Workspace ONE at 18%.
Fleet CEO Mike McNeil separately told Computerworld that 36% of respondents manage Apple devices through Intune. He called it a Windows-first platform and argued that tooling built around one platform's assumptions tends to be weakest at the edges of another's. That is a vendor competitor's opinion. The survey doesn't test it, and it says nothing about Intune's actual rollback behavior.
A separate caution: McNeil's 7.6% figure for exclusively-Mac shops refers to Fleet's own customers. It is not the same as the survey's 8%.
Why rollback is hard
McNeil told Computerworld that risk comes down to two factors: how much privilege the code runs with, and how hard recovery is. He pointed to scripts running as root, where reverting a configuration can't undo what the script already did. He also said iOS is exposed to a bad restriction or network profile, especially when recovering remotely.
He said Windows and Linux have the most scripting-heavy management and so the most room for script-level mistakes. These are interview opinions, not survey findings. The survey did not observe root-level misuse, compromised devices or failed remote recoveries.
His broader point is one admins will recognize: the management channel is a high-value target. AI speeds up authoring, but human review capacity doesn't grow with it. Fleet's CEO put it as errors, or anything malicious in a script, reaching production faster without a gate.
Fleet's proposed fix, and its limits
Fleet advocates treating device configuration like application code. Per McNeil's description to Computerworld:
- Profiles, scripts and policies live as files in a Git repository.
- A change arrives as a pull request.
- A second person reviews it.
- Automation applies it to devices.
- To undo it, you revert the commit.
Fleet says 28% of respondents already keep a human in the loop and route AI-generated changes through version control. McNeil said the pattern works with any tool that has an API.
There are two caveats. First, Channel Insider notes the report acknowledges its survey did not directly test whether this approach improves recovery times. Second, McNeil himself notes that a revert fixes what the configuration says but doesn't undo a script that has already run. He argues for declarative configuration over imperative scripts where possible. Fleet also sells a product built around this workflow, so weigh the advice accordingly.
Practical takeaways for admins
These are general operational implications, not steps the survey evaluated. The sources give no Intune-, macOS- or Apple MDM-specific rollback procedure.
- Gate AI output. Require human review before AI-drafted profiles or scripts ship. Authoring is now fast, and review is the bottleneck.
- Keep a change record. Know who changed what and when, so that remediation doesn't start with archaeology.
- Limit privilege. Be wary of AI-written scripts that run with elevated rights, since those are the hardest to reverse.
- Prefer declarative over imperative where the platform allows it.
- Rehearse the revert. If your recovery plan is manual, find out how long it takes before a bad day, not during one.
- Stage rollouts. This is a general best practice rather than a survey finding. Pilot rings limit how many devices a bad change can reach.
Other findings
78% use or are piloting declarative device management. Channel Insider rounds this to nearly 79%. 31% say keeping up with Apple's release cadence is the hardest part of Mac administration. Computerworld adds that budget and headcount ranked last, at 5%.
Bottom line
The survey doesn't show that AI-written configurations fail more often than human-written ones. What it does show is a gap that is hard to defend: many admins let AI output reach production, and most say they can't reverse a bad change within an hour. For MSPs and mixed Windows-and-Mac shops, the sensible response doesn't depend on Fleet's product. Add review gates, keep a change history and test your rollback path before you need it.