Fleet survey finds AI-written Mac changes deploy in seconds but rollback takes days

Fleet’s Apple in the Enterprise survey found that 69% of more than 250 IT practitioners cannot reverse a faulty configuration change within an hour, and 43% need more than a day to undo it. All respondents had admin access to their device management systems. “When one of those changes is wrong, 69% can’t undo it within an hour, and 43% need more than 24 hours,” said Mike McNeil. The results were published Nov. 7, 2026.

The practitioners in the survey manage an average of 2,000 devices each, and nearly half have worked as Apple admins for six to 10 years. Apple device administration now covers identity, security, compliance and app lifecycle on top of a yearly macOS release cycle that forces changes every year.

Sixty-eight percent of respondents relied on AI for troubleshooting and log analysis across the previous 30 days, while 62% had AI write scripts and 61% had it create device management profiles or configurations. Eighty-six percent allow AI-written output into production after review, while 14% exclude it from production environments entirely. “AI can help push changes out in seconds, while a mistake can drain hours of valuable IT time and resources,” said Fleet CIO Allen Houchins.

Teams can push an AI-written configuration profile to thousands of Macs in seconds following a quick review. Reversing a faulty network setting or an app lockout in one of those profiles takes far longer than shipping it.

Just 26% of respondents are able to fix a bad change on the same day, and only by doing the work manually. Fleet says the risk is real even though there are no incident counts, since most teams cannot reverse a bad change within an hour.

Reversing a configuration corrects what it specifies, yet it does not undo a script that has already executed, so Fleet recommends declarative scripts instead of imperative ones (declarative scripts describe the desired end state; imperative scripts list step-by-step commands). Weakly written AI-created mobile device management (MDM) profiles can trigger a series of problems that take days to resolve, especially in large-scale deployments, and absent audit trails make errors harder to fix.

Bad or malicious MDM-delivered changes are especially consequential because MDM is among the limited trusted routes capable of bypassing macOS platform security prompts. Fleet CEO Mike McNeil said the tool created by AI is not what gets pulled back; the configuration it produced is, and most teams roll back changes manually.

Just 28% of respondents combine human review of AI-generated changes with version control to establish a clear record and rollback path. Fifty-three percent of respondents already run declarative device management in production, and 78% are using or piloting it.

Fleet CEO Mike McNeil said Mac admins function as engineers in practice, whatever their official job titles may be. “Mac admins are basically engineers now, whether their job description calls them that or not,” McNeil said. Mac administration is now an engineering discipline, and administrators have progressed further than the industry believes, but the surrounding infrastructure has not caught up.

Mike McNeil said on Oct. 8, 2026 that risk hinges on code privilege and how hard recovery is, and he noted that root scripts can seize complete control of a machine. Windows contends with a larger and older attack ecosystem and more legacy surface area, Linux grants administrators the most freedom and fewest guardrails, and the management channel represents the most valuable target across all platforms. Windows and Linux rely most heavily on scripting and offer the most room for script-level mistakes, while iOS faces danger from a faulty restriction or network profile, particularly during remote recovery.

Fleet’s CEO said AI accelerates authoring while review capacity remains unchanged, so without a gate, errors and anything malicious reach production faster. Computerworld describes a wide gap between how quickly AI-generated output deploys and how quickly it rolls back, and says adopting AI has moved ahead of deployment management. McNeil made the comments on Oct. 8, 2026, and Computerworld published its account the same day.

Only 8% of respondents manage Apple devices alone, and Intune, a Windows-first platform, is used by 36% of respondents to oversee Apple devices. Founder and CEO Mike McNeil told Computerworld on Oct. 8, 2026 that only 7.6% of Fleet’s customers rely solely on Macs, meaning most Fleet clients manage multiple platforms.

Fleet recommends handling configuration like code, keeping profiles, scripts and policies as files in a Git repository, so teams can revert changes by commit. Fleet says writing device configurations as code gives teams more control over changes and makes reversing mistakes simpler.

McNeil said on Oct. 8, 2026 that Fleet’s MDM system supports GitOps using YAML files, and the approach works with any tool that has an API. In McNeil’s recommended rollback process, a change comes as a pull request, a second person reviews it, automation applies it to devices, and teams reverse it by reverting the commit.

McNeil said on Oct. 8, 2026 that IT should move toward “less clicking, more engineering” as AI takes over profile creation, script drafting, and routine troubleshooting. As AI assumes mechanics such as building profiles, writing scripts, and handling routine issues, admins spend less time working out how to make a change and more time deciding what should change, its business impact, and where human judgment matters.

McNeil told Computerworld on Oct. 8, 2026 that administrators are stretched by the annual OS cycle, new frameworks, and deprecations, while budget and headcount ranked last at 5%. Thirty-one percent of respondents named Apple’s release cadence their top IT challenge, while budget and headcount ranked last at 5%.

Declarative Device Management and tighter platform security benefit enterprises, but the pace of Apple’s releases is the issue and the main reason admins require automation and a safety net. Fleet says modern device management must apply consistent workflows, visibility, and control across operating systems as AI adoption grows.

Fleet provides an open device management and vulnerability management platform trusted by more than 1,300 organizations in 90 countries to oversee millions of devices. Fleet says it applies infrastructure-as-code practices to device management so endpoints stay secure and operational.

Fleet says when configuration lives in files, each AI-written profile arrives as a diff, review turns into a pull request, and rollback turns into a single revert.

LLM assistants appear in 57% of respondents’ engineering toolkits, code editors in 46%, and CI/CD runners in 42%. AI assistants see more use than MDM consoles, which 53% of respondents use. Sixty-four percent of respondents already see a defined career path from administrator up to senior and staff engineer.

Microsoft 365 and Google Workspace simplified e-mail compared with running servers two decades ago, but e-mail still requires AI-driven phishing defense, multifactor authentication management, and related tasks that replaced server upkeep; the shift represents two steps forward and one back, and the report calls for three steps forward and zero back. The 9to5Mac column and LavX summary use hosted e-mail as an analogy for how new tools trade old burdens for new ones without reducing total workload.