Nanites Explores Autonomous AI’s Final Frontier
It’s network management, and MSPs may never be ready for it. Plus: Why Pax8’s Agent Store launch partners bought in early and two forthcoming solutions from Acronis.
There was a time when the idea of AI resolving end user technical issues autonomously struck a lot of MSPs as insane.
That time was December 2023, to be precise, when I first wrote about autonomous AI in an article about then forthcoming functionality from Atera. The very first reader comment on that post read, in its entirety, “This is insane.”
Fast forward to the present day, and the picture looks a lot different. Vendors like Pia and Thread that have been at this for a while are growing steadily and raising funds, a ton of startups are entering the market, and some of the biggest names in venture capital are investing serious (like, really serious) money in businesses built entirely around the growth potential of AI-automated managed services.
Yet as the folks at Pia would be the first to tell you, a lot of MSPs are still adjusting psychologically to letting agents do help desk work without humans in the loop, and I can’t help but wonder how far their adjusting will ultimately go. Autonomous password resets are relatively easy to accept given the low stakes if things go poorly, but when will autonomous network management feel equally safe?
How about never—is never good for you?
It might have to be, I’m starting to think, based on a recent conversation with Alex Cronin, CEO of a very interesting startup called Nanites AI, whose AI-native network troubleshooting solution is nearing general availability. Feed the system an alert or ask a question through its chatbot interface and it will autonomously figure out what’s wrong and how to fix it.
“It executes the same actions as a human engineer, but at machine speed,” Cronin says. In thoroughly documented lab testing, Nanites successfully reduces MTTR of nearly 50 different network performance issues by about 90%.
It would save even more time, moreover, if it actually fixed those issues itself. By design, however, it doesn’t. “It’s read-only. No write commands,” Cronin (pictured) says. “It can propose changes, but the final decision is always human.”
And probably always will be, I suspect, not because the solution’s incapable of repairing networks autonomously but because it’s incapable of doing so with 100% reliability, and neither MSPs nor their clients have much tolerance for avoidable outages involving critical infrastructure like a network, as Cronin knows well.
“Humans forgive humans. They don’t forgive technology,” he observes. “Until AI stops hallucinating, we can never let [Nanites] make live changes.”
Which likely means never making live changes, period, given that OpenAI itself concedes it can reduce hallucinations but never eliminate them. Plus, others add, even if eradicating hallucinations was possible it would make generative AI a lot less useful, assuming you’d like it to generate something original now and again. What we call “hallucinations” when they come from LLMs we call “bad ideas” when they come from people, and there’s a reason “no bad ideas” is the clichéd instruction issued during brainstorming sessions.
Cronin is aware of all this. “LLMs hallucinate by design,” he says. “That’s not going away, so you have to build systems around that.”
Pretty much everyone else in network management agrees right now. The new “AgenticOps” functionality Cisco introduced last week, for example, combines “autonomous action with built-in oversight,” and network management vendor Auvik is in no rush to go fully autonomous either.
“We’re going down that path in certain circumstances,” says Mark Ralls, the company’s president, “but it’s an area where we’re treading very carefully.” As any good critical infrastructure doctor must, he adds. “It’s almost like the Hippocratic Oath. First, do no harm.”
This is a strange moment, I’ll admit, to be questioning autonomous AI when the most viral piece of AI writing in the last week argues that agents will be doing close to everything sooner than you think. But there clearly will be some limits on how far agentic automation goes in managed services for a while, despite the technology’s profit maximizing power. The challenge for MSPs and all the vendors wooing them with agentic solutions will be figuring out how to get as close to those limits as possible without straying beyond them.
(Open) networking as a service
In the meantime, the solution Nanites is building appears poised to help MSPs make money as well as save it at the help desk. Though the system as it exists today supports pretty much any network OS businesses use, its original mission was to help enterprises and service providers simplify management of networks running on open source OSes like SONiC and white-label hardware. From a CAPEX perspective, open networking, as it’s called, is significantly cheaper than the proprietary kind.
“The challenge is total cost of ownership,” Cronin says, which includes considerable OPEX expenses associated with provisioning, operating, and troubleshooting.
“It’s very challenging,” Cronin observes. “You have to take all these disaggregated components—hardware, software, and support—and try to marry them up and make them all work.” Nanites is designed, among other things, to simplify that task.
“I believe Nanites is the bridge from traditional to open networking because it abstracts the complexity using AI,” Cronin says. That also makes it a bridge to network-as-a-service offerings built around affordable infrastructure and margin-rich ongoing support, a prospect that has at least some MSPs talking to Nanites months before its solution goes GA.
“We’re working with MSPs large and small,” Cronin says.
So too, I might add, is Flamingo, the company I wrote about here recently that similarly sees combining open source technology with AI as a way to help MSPs make more and spend less.
Who’s in on Pax8’s Agent Store?
Someday, perhaps, Nanites agents will be listed on the agentic AI marketplace Pax8 told the world it was building during its Beyond conference last spring. If they are, it seems, they’ll be in good company. Among the Agent Store launch partners Pax8 announced last October are industry giants like Microsoft and AWS, MSP mainstays like ConnectWise and Rewst, and agentic AI trailblazers like Pia and (recent ConnectWise acquisition) zofiQ.
Do the 30 companies on that list all have detailed roadmaps for what they’ll sell in the marketplace when it opens for business and concrete plans for weaving those agents into larger go-to-market strategies? Not so far as I can tell from recent conversations with three of them. What they do have, however, are six deeply held beliefs:
1. Agents will soon play a part in the vast majority of solutions sold and supported by MSPs.
“That’s our vision, and that’s where we see the market going,” says Gaidar Magdanurov (pictured), president of security vendor Acronis. “Agents will be everywhere in all products.”
2. Pax8 is all-in on the future importance of agents in managed services.
“What [Pax8 CEO Scott Chasin] said at Pax8 Beyond last year about the managed intelligence provider is exactly what we believe as well,” explains James Farrow, co-founder and CEO of Cyft, one of the many AI-native startups for MSPs you’ve read about here on Channelholic in recent months. “It’s just perfect alignment philosophically in how we see the future.”
3. MSPs will eventually transact as routinely on agent marketplaces as they already do on SaaS marketplaces.
“From a first-principles perspective, they exist for the same reason,” Farrow says. “There are so many options out there, and we need them aggregated to make discovery easier.” If anything, he continues, MSPs will probably value agent marketplaces even more than they do SaaS ones.
“AI is harder to evaluate for ROI and value,” Farrow observes. “You need trust that the marketplace has vetted vendors and that they’re not vaporware.”
4. A whole lot of MSPs buy software from Pax8.
Over 47,000 of them, to be precise, according to Pax8 itself. “Pax8 is a huge player,” says Amanda Adams, vice president of Americas Alliances at CrowdStrike. “They own that market when it comes to how they service their clients.”
5. Pax8 is, as far as we know right now, pretty much the only MSP-focused SaaS marketplace operator developing an agent store.
The store it’s building, moreover, will do more than sell agents. It will orchestrate the work those agents do as well.
“Over time, agents will talk to each other,” Magdanurov says. “You’ll mix and match tools into workflows—SentinelOne detects something, Acronis responds, another system reports it.”
MSPs will design these end-to-end workflows, he continues, using Pax8’s marketplace as a sort of centralized point of coordination. Those workflows, in turn, will become the foundation for process-optimizing, productivity-boosting end user solutions, Adams adds.
“Instead of just a marketplace listing and buying products, what the Pax Agent Store is truly doing is delivering outcomes,” she says.
6. Based on the logic of beliefs 1-5, launch partners have concluded, listing on Pax8’s forthcoming agent marketplace makes sense even if many of the specifics are still coming into focus.
“We need to be everywhere customers are,” Magdanurov says. Adams agrees.
“There is only an upside,” she says.
As long as we’re talking about Cyft and Acronis…
…I might as well fill you in on a little news from both of them, starting with Cyft.
When you last heard from the company six months ago, it was completing work on the first of what will in time be at least three AI-native automation solutions for MSPs. Called Sam (as in “service activity memory”), the system is an automated, AI-powered documentation tool that quietly listens to help desk calls and uses what it hears to update tickets, generate detailed documentation, and draft client-ready resolution reports.
Cyft officially launched the product, which sells for $149 per technician per month on annual contracts, last December. At present it integrates solely with ConnectWise PSA, but three more unspecified PSA integrations are due this year.
Longer term, Cyft has ambitions not unlike those you read about here recently in relation to another young startup named Lexful. Both companies plan to use automated documentation as the starting point for collecting data capable of doing far more strategic work for MSPs across their tool stack.
“MSPs rely on human-generated data today, which is inconsistent,” Farrow (pictured) says. “We provide consistent, machine-generated intelligence that can power many downstream outcomes.”
As for Acronis, it’s currently working on two new agentic security solutions. The first, called Acronis Cyber Workspace, is a self-help security chatbot that MSPs can offer their clients. The second, called Acronis Cyber Studio, will let MSPs create and run integrated agentic help desk workflows.
“People used to operate multiple tools, go to different places, click multiple buttons,” Magdanurov says. “Our goal now is to streamline this so you have one workflow.”
That will typically begin with a technician receiving automated word of an incident and issuing remediation instructions via natural language chat. “Then the agentic workflow kicks in and executes all the proper steps, coming back to technicians only if something needs approval, or fully resolving it automatically,” Magdanurov explains.
Both products are due by the end of the year.
Who better to ask about Google’s new partner program than the guy who built it?
Though it’s the only big three hyperscaler not among Pax8’s Agent Store launch partners, Google has plenty brewing in AI, and a brand new partner program designed to help MSPs and others get in on it all. Phil Larson, managing director of partner programs for Google Cloud, discusses that and more on the latest episode of MSP Chat, the weekly podcast I co-host. Check that episode out here and others featuring some of the MSP world’s top thought leaders here.
Also worth noting
NinjaOne added IT hardware and software asset management to its unified IT ops platform for MSPs.
Cisco announced updates to both its AgenticOps solution and its Cisco AI Defense product.
Cisco unit Splunk’s new agentic AI offerings include AI models coming soon and a new MCP server.
Fully 78% of organizations can’t scale AI initiatives safely, according to new AI maturity research from JumpCloud.
Cork Cyber introduced Cork Cyber Score, a new offering designed to help organizations quantify and communicate cyber risk.
Proofpoint announced its acquisition of Acuvity to expand AI security and governance capabilities.
Palo Alto Networks announced the completion of its $25 billion acquisition of identity security specialist CyberArk.
Arctic Wolf launched Aurora Managed Endpoint Defense for MSPs to enhance managed security offerings and drive momentum within the MSP channel.
Illumio expanded its strategic partnership with Armis to enhance security across converged IT and OT environments.
ExtraHop added new visibility and forensic capabilities to its agentic SOC platform.
BitSight unveiled what it calls the industry’s first dark web intelligence solution for supply chains.
Bishop Fox introduced a new pen testing service that combines the elite white hat hackers with a proprietary AI engine.
SpecterOps launched BloodHound Sentry to help organizations operationalize identity attack path management and reduce identity-based risk more effectively.
Reinvent Telecom launched Always On Internet, a connectivity solution aimed at delivering uninterrupted service for reseller partners and their customers.
Dell added support for Nutanix to its private cloud offerings.
Alex Spigel is the new CEO of Choice Cyber Solutions.
Kelly Morgan is the new chief customer officer at KnowBe4.
Hey, I know most of these people! GTIA announced the finalists for its 2026 North America Spotlight Awards, recognizing individuals and organizations for excellence and impact across the technology channel.







