What Mac Admins Learned About AI This Year (And Why It Matters for Your IT Team)

MacAdmins 2026 ran close to 60 sessions across five rooms simultaneously for three days in State College. Most of it was exactly what you'd expect from this community: practical, high-quality Mac admin content. A handful of talks went further. They offered a different frame for the job itself, not just better tools or sharper scripts, but a different way of thinking about what the work actually is. Here are five worth knowing even if you weren't there.

AI Stopped Being a Slide and Started Being a Script

Kandji's Weldon Dodd opened his session, Formula 1 and Device Management: MCP is Team Radio for Your IT Stack, with a number worth sitting with. In a six-year running survey with tech writer Jason Snell, the share of Mac admins calling AI "very much" useful in their job jumped to the top of the response list this year. 84% now report regular use in some form. That's not early-adopter territory. That's most of the room.

His real argument was more useful than the stat. MCP works as a translation layer, turning a plain-language request into the structured API calls a tool like Jamf actually needs. His framing: using an AI agent through MCP is like ordering food at a restaurant instead of cooking it yourself. You don't manage the kitchen. He backed it up live: generating patch-status dashboards from a prompt, pushing changes across a device fleet, cross-referencing vulnerability scores against his own calendar to flag which device owners he had meetings with that week. His sentiment on AI progression is one that many likely share: capability jumped further in the last six to twelve months than he expected, and it's now doing things he "didn't know it could do a year ago."

Gabriel Jordan's session, Applying AI to Log Analysis, made a narrower version of the same case with logs instead of dashboards. His pitch wasn't that AI replaces reading logs. It's that logs are genuinely hard to parse manually because they're missing context, and an AI model closes that gap once the data's clean. The catch is real: logs are full of hostnames, usernames, and IPs, so he built a sanitization script before uploading anything anywhere, and found Claude specifically handled messy IPv6 formatting better than ChatGPT did. In his case study, the model traced a ProTools freeze to an Exchange error loop responsible for 84% of a 616,000-line log file. Not the direct cause, but the thing quietly starving the system underneath it. His summary was exactly right: AI "enhances existing skill sets." It doesn't replace knowing what question to ask.

Put those two next to each other and the takeaway isn't "AI is coming for IT admin." It already arrived, quietly, in dashboards and log triage built by people who needed it to work.

Onboarding Speed Is Now a Number You Can Compare

Ryan Ball from Alectrona, in Mac Onboarding with Alectrona Patch, walked through onboarding math that's hard to argue with. Their Patch tool installed 10 apps on a fresh Mac in an average of 3 minutes 9 seconds, against 15-plus minutes for the same install list using a traditional scripted approach. The difference came from downloading everything concurrently instead of one app at a time, verified against checksums along the way. His stated philosophy is worth adopting wholesale: old software or an old OS is risk, full stop, so the default should always be the latest version.

The point underneath the benchmark matters more. Manual onboarding scripts accumulate technical debt fast. The answer isn't a cleverer script. It's removing the manual step from the process entirely.

Standing Credentials Are a Design Flaw, Not a Housekeeping Problem

Justin Gianelloni and Matt Parker's session, The Secret-Free Fleet, consisted of some sharp architectural thinking. Their starting complaint was that endpoints holding standing API credentials are inherently risky, and most identity providers only issue high-privilege "god mode" keys. One compromised laptop can mean a compromised admin credential. Their fix, custom middleware called Hookshot, inverts the model entirely. The endpoint requests elevated access on demand, gets validated by a signed certificate, and never holds a standing credential in the first place.

Their framing is worth carrying into other conversations: standing secrets are "an architectural problem, not a hygiene problem." No rotation policy fixes a design that shouldn't require the secret to exist on the device to begin with.

When Relationships Matter More Than Tools

Nathan Beranger's session, Support as Relationship: How Human-Centred IT Builds Trust, wasn't a product walkthrough or a technical deep dive. The basis is that the hard part of the job was never the technology. It's whether people trust you enough to flag something small before it becomes something broken. His specifics were plain and repeatable: acknowledge a message fast even with no fix ready, never make someone feel dumb for asking, be upfront when you don't know something and then actually circle back. His line for it: "Respect is not one big gesture, it is a hundred smaller ones."

The Q&A afterward had two details worth keeping. On language, he pushes his team toward "colleague" over "user" wherever the workflow allows it. It's a small shift that changes the tone of every interaction that follows. He shared a balanced view on AI-assisted support using an internal AI assistant strictly for a first-pass triage. The automation buys speed, but it doesn't substitute for the human follow-up that actually builds trust.

Nathan is a Foqal ambassador and works at Bridgeable, a Foqal customer, so we're not exactly a neutral party here. But we'd have included this talk regardless. It's the one from this year's conference that stands out in the midst of AI-everything, and the worksheet we built with him is simple and effective when actually put into practice.

The worksheet breaks down six moments that come up constantly in IT support and gives you language to use in each scenario. Get the worksheet here.

What Ties These Together

Provisioning, security, and log triage are all getting faster and more automated by the month. Three of these five talks prove that directly. The fifth is the reminder that the part of IT that isn't technical, whether someone trusts you enough to come to you early, doesn't get automated at the same pace. It still gets built one interaction at a time. The strongest sessions this year treated that as a fact to design around, not a tension to resolve.