Poolside Observation: It’ll Take Software to Train the Next Generation of Technicians. We Don’t Have It Yet.
Why every MSP on the planet will eventually need AI-driven synthetic practice software and token budgeting smarts. Plus: MSPs have no more excuses for ignoring their clients’ Apple devices.
OK, I lied.
I said in my last post that I’d be taking the week off. And I did in fact just finish a gloriously indulgent vacation (photographic evidence above) involving a lot of eating, snoozing, and sunbathing. But I’m not taking the week off exactly, because my vacation reading sparked a few wee thoughts worth spending my long trip home to Seattle getting down into words.
Now let’s stipulate upfront that your idea of vacation reading may not be the same as mine. But no kidding, I really did read this wonderfully nerdy academic paper poolside last week, months after first coming across it and wishing I had time to give it a thorough look. It covers a lot of ground and I encourage you to skim read the whole thing. But the nub of its argument is that the rapidly growing sophistication of agentic AI is creating a:
“…collision of two racing cost curves: an exponentially decaying Cost to Automate (cA), driven by compute and accumulated knowledge, and a biologically bottlenecked Cost to Verify (cH), bounded by humantime and embodied experience. This structural asymmetry widens a Measurability Gap (Δm) between what agents can execute and what humans can afford to verify, and determines the verifiable share of deployment (sv) that separates truly productive agentic output from merely plausible output.”
In other words, as AI becomes increasingly capable of labor only humans could perform before, the role of humans shifts from doing work to validating that the work agents do on their behalf is accurate and effective. Which is a conundrum, of course, because the more agents come to dominate certain tasks, the harder it will be to find humans qualified to assess whether the agents’ output is any good.
And obviously, when I refer to “certain tasks” here, I’m thinking mostly about help desk tasks, the work that service desk automation systems are getting better and better and better and better at all the time. As I’ve written before, the ability to transfer lower-value administrative chores from people to bots is both a blessing (MSPs using service desk automation software are experiencing 5% to 25% technician productivity improvement, according to Omdia) and a curse, in that people are going to play an indispensable role in Level 3 work at a minimum for a long time to come and no one (like, no one) seems to know where tomorrow’s Level 3s will come from once AI makes human Level 1s and 2s obsolete.
Hence the “biological bottleneck” identified in the research paper. We can’t, or shouldn’t, adopt AI that we can’t validate, we can’t validate AI without deeply experienced humans, and we’re slowly beginning to burn through our supply of deeply experienced humans.
The closest I’ve ever come to an answer to this dilemma is when Joshua Skeens, CEO of mega MSP Logically, speculated that he’ll eventually need to hire bright, young people with L3 potential (referred to as “innate aptitudes” in the research paper) and spend a year plus a whole lot of money training them to do L3 work.
OK, but how. The paper vaguely yet repeatedly suggests “AI-driven synthetic practice,” as in using software to do help desk work on a kind of AI-powered simulator.
Which brings me to the fairly obvious poolside thought that came to mind reading this paper. We’re going to need Level 3 technicians for some time to come. On-the-job training isn’t where we’re going to get them. Training newbies to hold L3 jobs is the best available answer to that problem so far. And “AI-driven synthetic practice” solutions are what we’ll use to provide that training.
Except that there aren’t a whole lot of AI-driven synthetic practice solutions for MSPs out there at the moment. Empath’s software could potentially do the job, as could Virti’s with some customizing. But it seems like vendors ought to be moving as swiftly as possible to create more such solutions given that the total addressable market for them is basically every MSP on earth, and probably every corporate IT department to boot.
More reasons to get smart about token spending sooner versus later
Ready for another good beach read? Try this. It’s a terrific take on the token economics issue I discussed in a somewhat different context a few weeks ago. My point back then was that producing strong, steady margins on AI services is going to be hard for MSPs given that upstream AI companies increasingly employ usage-based billing schemes that make costs hard to predict and downstream end users want predictability in their IT spending.
And to say that AI companies are increasingly charging for consumption is to understate things. In just the six weeks since that first post of mine went live, GitHub and Anthropic have both gone all-in on usage-based billing.
For a very good reason, of course. The data centers that host AI software are extremely expensive and the LLM makers spending absurd amounts of money to build more of them can’t afford to sell tokens for less than they cost to manufacture anymore.
As it happens, that same dynamic plays into a phenomenon Anton Leicht discusses in the blog post I linked to up above. While token costs have been dropping in aggregate over time, they’re actually rising for the newest, most powerful models. To cite the latest example, Gemini 3.5 Flash may well do what used to take a developer days “in a fraction of the time, often at less than half the cost of other frontier models,” per Google’s claim, but it costs more than earlier versions of the Flash family, not less.
Meanwhile, as the newest models get to be really powerful, a la Anthropic’s Mythos, both cyber security and national security concerns will limit who gets access to them and when.
Which suggests that we’re entering a have-and-have-not era in AI in which businesses with big budgets, big brands, and political pull get the premium tokens and everyone else will make do with what’s left. Further suggesting that businesses with modest budgets, obscure brands, and no political pull (like SMBs) will badly need MSPs capable of doing two things:
Provide actionable guidance on balancing cost against performance when deciding which version of which model to apply against each of their workloads.
Wring as much impact as possible out of every sub-premium token they have access to and can afford.
Could be wrong, but sounds like another area where a new class of software products, or at least features, could be extremely useful.
Meet Virtua’s excuse eraser for managing Macs
But then again, why spend a lot of time mastering the arcana of a brand new field evolving rapidly in real time when there’s an easier way to burnish customer loyalty and expand your top line? Heck, if you’re reading this on your phone you might be holding the modest little goldmine I’m referring to in your hand right now.
I’m talking about Apple device management. To be sure, managing endpoints is about as commoditized a service as an MSP can offer, but as long as businesses still use endpoints and endpoints still experience downtime and security breaches, businesses are going to need someone to manage their devices, and if it can’t be in-house IT it’ll be an MSP instead.
Yet, according to Justin Esgar, CEO of New York City-based Apple support and solution specialist Virtua Computers, most MSPs studiously blinder themselves to the existence of Macs, iPhones, and iPads in the environments they’re responsible for.
“The PC MSPs just don’t want to talk about it or deal with it,” says Esgar (pictured), noting that one reason he knows this to be the case is how often an MSP calls him in to manage the five Macs in a 100-endpoint office they support.
“They’re leaving money on the table,” Esgar observes. Or worse yet, he adds, subtracting money from their top and bottom lines.
“They’re harming their relationship with their customers,” says Esgar, an Apple Premium Technical Partner with 18 years of experience who also co-hosts the reliably excellent All Things MSP podcast. “To say, ‘We don’t do that because it’s a Mac or an iPhone or an iPad,’ is a really damaging line for an MSP to say to a customer. So they’re either going to lose money straight up by not supporting it, or they’re going to lose the customer and then lose money.”
Maybe a lot of money at that. Apple products currently account for 12.6% of global desktop/laptop market share (23.9% in the U.S.), according to StatCounter, and 21% of global smartphone share, according to IDC. The numbers are even higher in specific verticals, Esgar notes.
“Doctors use iPhones. Truck drivers use iPads. Those devices need proper management,” he says. “These MSPs are often ignoring 10 to 30 percent of a client’s fleet because it’s Apple.”
He knows why they’re not supporting all that hardware, too. “They don’t know how. They don’t have the bandwidth to learn.” And besides, no one needs to worry about Apple devices anyway, right?
“They’re fed a lot of information from Apple like Macs are secure, so it should just be fine, but it’s not,” Esgar says.
Indeed, turns out security incidents involving Apple devices really are a thing. And while software for mitigating that thing is available from Microsoft, via Intune, Jamf, and Addigy, among others, deploying a tool doesn’t necessarily close an MSP’s Apple management skills gap.
That realization is a big part of the thought process that led Esgar to envision an outsourced Apple management service for MSPs that don’t have Apple skills, don’t use Apple tools, don’t wish to change either fact, but do wish to manage Apple devices and make money doing so. Called VirtuaCare, the service provides either white-label or sub-agent Apple management support on a 24/7 basis. You keep the relationship, we handle the Apple stuff, Esgar says.
Partners pay by the hour rather than monthly subscription fees. “We don’t want to lock you into a per-client, per-device thing,” Esgar says. “We know PC MSPs have a lot of clients, and maybe this person has two Macs and that client’s got four iPads, or whatever.”
Partners can buy hours at $200 apiece in five-, 10-, 15-, or 20-hour blocks. The more hours you buy, the more ancillary benefits (like Slack channel access and monthly strategy calls) you get, and the shorter your guaranteed response time.
From Esgar’s perspective, though, VirtuaCare’s most significant ROI, which you get no matter how many hours you buy, is the satisfaction of knowing you’re doing the job you aspire to do for all of your clients all of the time.
“If you’re a PC MSP and the customer says, ‘Jim’s computer isn’t connecting to the network,’ you would never say, ‘We don’t do firewalls.’ That’s not a thing we do as MSPs, right? We want to make sure the whole thing works,” Esgar says. “PC MSPs, Mac MSPs, we need to just start being MSPs.”