Putting AI Agents to Work: How BMS Runs Its MDM with an Agent Team, and How Your Organization Can Too


 🎙️Podcast: Building AI Agent Teams

Don't forget to subscribe 😀

A while back I caught myself doing something I never expected to do: handing out work to a team that isn't on any payroll. Each morning I check in on the BMS MDM project, see what my agents posted to the shared board overnight, and decide what needs my attention. It feels less like using software and more like running a small staff.

Here is how that came about. At Benchmark Service (BMS), I built BMS MDM, a device management platform for Apple fleets. Keeping a platform like that healthy takes a lot of steady, repetitive work, and a single AI chat window was not the right tool for it. It lost context, blurred priorities, and had me explaining the same things over and over. So I tried something different: an AI agent team.

How the Team Works

An agent is an AI assistant with one defined role. A team is several of them coordinating through a shared task board. One agent posts what it has done and what should happen next, and another picks it up. Because every task and hand-off is written down, nothing depends on anyone remembering what happened.

Each agent needed a name so the board shows who posted what, so I used Star Trek bridge roles like: an XO, a Science Officer, and a Navigator. It started as a bit of fun, but it earned its place, because each title matches the agent's job. A glance at the board tells me who is responsible for what, and if something looks wrong, I know which role to look at.

The agents take on the repetitive parts of platform administration: moving devices onto the platform remotely, deploying apps, setting up new devices automatically when they enroll, handling wipes, locks, and reassignments as staff come and go, running regular health checks, and troubleshooting remotely with escalation to Apple when needed. They also watch for security advisories and assess each one against how a client's systems are actually set up, and they keep one shared knowledge base up to date. I stay in charge of the decisions. Agents prepare, check, and recommend, and I review before anything important changes.

Here is a small example of why the structure matters. While recounting devices during a migration, one agent compared our tracking spreadsheet against the server's actual enrollment log, and the two didn't agree. Some rows were out of sync, including a device filed under the wrong person. An assistant working alone would most likely have trusted the spreadsheet, because that is what it was handed. Having a second agent check the first one's work caught the mismatch before it caused any trouble. The same habit helped in another case. A support representative told me by phone that one of our Macs was enrolled in our management server. The team checked the server and found the Mac had never actually checked in. That gave me real evidence to move the support case forward, instead of trading one person's word against another's. The agents don't take anyone's word for it, not even a vendor's. They check against the records.

It Isn't Only for IT

Nothing about this structure is specific to IT. Any operation with routine, repeatable work can use it. A scheduling agent can track requests and availability while another drafts confirmations and reminders for you to approve. One agent can gather the figures for a weekly report while a second checks it for gaps before you read it. Follow-ups and paperwork work the same way: an agent keeps a list of open conversations and drafts the next message for each, or assembles first drafts from your templates, ready for your review.

How to Start Small

If you want to try it, start smaller than you think you need to:

  1. Pick one repetitive process that takes up time each week.
  2. Break it into two or three roles, such as one agent to gather information and one to check it.
  3. Set up a shared place for the agents to record their work, such as a simple task board.
  4. Decide which actions need your approval, and keep those with a person.
  5. Run it for a few weeks, review the results, and adjust the roles before adding more.

What to Watch For

Agents are only as safe as the access you give them. Give each one the minimum it needs, keep sensitive data out of places it doesn't belong, and keep a person responsible for decisions that affect clients, money, or security. Review the board regularly. A team that runs unsupervised will eventually do something you didn't intend.

How BMS Can Help

BMS helps small teams set up AI in a practical way: choosing a process worth handing off, defining clear agent roles, and building in a human check. If your team has routine work that eats into your week, I'm glad to talk it through.

Which routine task in your own operation would you hand to an agent first? Let me know in the comments or send me a message.

Comments

Post a Comment

Thanks for your comment.

Popular posts from this blog

The Process of Troubleshooting

Image
🎥 BMS Resolution Framework The True Cost of IT Downtime—And How Great Support Solves It Research shows that workers lose an average of more than 22 minutes per day as a result of IT issues. This adds up to 91 hours, or more than two weeks per year! Do the math on how much that costs you or your organization. Those numbers are likely much higher for individuals and organizations without access to proper IT support, and small organizations are affected the most. This statistic alone makes a pretty solid case for a proper, effective, and comprehensive IT support solution. And that’s not even considering security issues, which can damage your reputation and cost significant money in lost business. Prevention vs. Inevitable Glitches A solid issue prevention solution involving proper backup, virus protection, and system administration—which I’ll be talking about in a future blog—is the best way to avoid issues. However, problems will come up with all IT systems no matter how well managed. ...
Read more

Pipeline Threat

Image
🎙️Podcast: Pipeline Threat Don't forget to subscribe 😀 New Threat Reality The threat landscape facing small and medium-sized recruiting firms in Japan has crossed a fundamental threshold. For years, cybersecurity was framed around targeted manual intrusions—expert hackers spending weeks or months probing a specific company’s defenses. Today, the democratizing power of artificial intelligence, paired with state-backed infiltration tactics, has turned cybersecurity into a game of high-volume, automated exploitation. Recruitment agencies are uniquely exposed to this new reality. Operating in a high-trust, high-data-volume environment, recruiters serve as the digital doorway between job candidates, external contractors, and major corporate clients. Understanding this threat requires looking past traditional views of "hacking" to examine how automated tools and pipeline weaponization operate, what a breach actually costs a firm, and where this threat is heading in the near f...
Read more

The AI-Driven Threat Surge: Why Apple’s Record Patch Release Signals a New Era in Cybersecurity

Image
The modern digital threat landscape is evolving faster than ever before. For years, organizations and individuals held a comfortable assumption: software updates were routine maintenance, and cybersecurity threats developed at a predictable human pace. That era has officially ended. As artificial intelligence technologies advance, so do the tools used by threat actors. Generative AI and automated vulnerability scanners now allow malicious actors to analyze software code, discover exploits, and craft attack vectors at scale. This escalation in AI-generated and automated threats has triggered a global response from tech giants, security firms, and regulatory bodies worldwide. A defining example of this shift arrived in mid-September 2026, when Apple released a record-setting volume of security patches across its entire ecosystem, including iOS, iPadOS, macOS, watchOS, and device drivers[ cite: 5]. Driven in part by industry disclosure initiatives like the Mythos initiative—which has acce...
Read more