Seen from one angle, deterministic workflows of the kind traditionally associated with RPA applications are an investment by MSPs in consistency and quality. If you have a proven way of onboarding new users, for example, you want everyone at the help desk to follow it every time there’s a new user to onboard.
Seen in relation to rising adoption of AI-based service desk automation tools, however, deterministic workflows are a kind of hedge against the unpredictable and potentially dangerous consequences of letting an agent do something important on its own. MSPs, in this view, may increasingly trust AI to triage and diagnose tickets (32% of them use AI operationally now, according to new data from Nerdio last week, and 46% are piloting or planning to adopt within a year), but they don’t trust it to improv what happens next autonomously.
Nikhil Sehgal, CEO and co-founder of service desk automation startup Neo Agent, suspects most MSPs fall into that latter camp, and he has a message for them: get over it.
“Generally, given the right context, the right guardrails, and the right skills, these systems are highly, highly effective,” he says.
To be clear, Neo Agent’s software doesn’t improv anything, but it does lean harder in the direction of fully autonomous versus totally scripted than many MSPs are ready for. Which hasn’t stopped the company from growing in the last year from 20 to 200 partners, one of whom is Sehgal’s dad, a U.K.-based MSP who was Neo Agent’s very first user.
“I initially built Neo as just a small prototype to help my father become more efficient on his service desk,” says Sehgal (pictured), a tech entrepreneur and one-time algorithmic trading specialist at TD Cowen.
“I was so engrossed in the machine learning field,” he recalls of those days. “It became obvious that there would be new ways for MSPs to automate work on the service desk.”
Obvious enough to inspire the creation of Neo Agent just weeks after the launch of ChatGPT late in 2022. “Initially, it was running on GPT-3,” says Sehgal of the company’s first product, which helped technicians find the fastest route to closing new tickets based on evidence in PSA data from earlier ones. It didn’t catch on, for reasons Sehgal figured out early last year. It was tackling the wrong problem.
“We were telling them, ‘Hey, this is how you solve the issue,’ but they already knew that,” Sehgal says. “It’s just the job takes a while to do.”
So Sehgal and his team rewrote Neo Agent to execute workflows, at machine speed, rather than simply describe them. That version of the program wasn’t an immediate hit either, however, mostly because it was better at ad-libbing workflows than users were OK with.
“They would say, ‘one time Neo does something and the next time it just decides to do something else,’” Sehgal explains. “You want the system to think for itself, but not too much.”
Which is exactly the “happy medium,” in Sehgal’s words, that Neo Agent now offers via what the company calls “Neo Skills.” Like their equivalent in ChatGPT and Claude, Neo Agent’s skills perform tasks independently based on pre-defined guidelines rather than rigid step-by-step instructions. The result, according to Sehgal, is a tool with multiple advantages over RPA solutions.
For one, Neo Agent’s flexible enough to think its way around a ticket’s “nuanced edges,” just like a good human technician. Most RPA solutions, for example, require you to create separate user onboarding flows for each of your clients based on minor differences in their environment. Neo Agent can handle those same variations through a single set of instructions.
It also, according to Sehgal, adapts to surprises like an API change in your RMM solution more flexibly than deterministic scripts. “If you have a flow set up in an RPA tool that relies on interacting with that system and it runs into an error, the whole thing breaks,” he says. “With an agentic system, you can recover.”
Best of all though, Sehgal adds, Neo Agent creates skills automatically in a matter of minutes instead of the hours an RPA solution typically requires to accomplish the same goal. “You’re not defining every single rule that must happen,” he says. “You’re building an agent, giving it access to the tools you use, like PSA, RMM, M365 environments, and defining in natural language how you would like these agents to behave.”
Midsize MSPs with 15-50 technicians and 3,000-5,000 endpoints under management find that value proposition especially attractive, Sehgal continues. “They typically don’t have the time or resources to hire expensive automation engineers to operate a complex product,” he says. “They need to see efficiency gains from automation very quickly, and we can do that for them.”
At usage-based rates that are arguably the trickiest thing about Neo Agent. Each of the system’s subscription plans comes with a monthly supply of “credits,” rather than tokens. “One credit is equivalent to Neo taking some useful action inside an MSP environment,” Sehgal says, like replying to someone through your customer portal or running an RMM job. Or it could be something harder to do and possibly more expensive as well.
“An action that creates and executes PowerShell code is a complex action, and that can sometimes cost two credits depending on the situation,” Sehgal notes.
The 10,000 credits included with Neo Agent’s “Starter” plan, which costs $1,300 per month, are enough to automate about 3,000 tickets, the company estimates. Resulting in serious ROI for most MSPs, Sehgal adds.
“They should be able to at least double their endpoint to tech ratio with a good implementation of Neo,” he says.
Assuming, that is, they’re willing to let the system do its thing unsupervised. According to Sehgal, more and more MSPs are all the time.
“From my experience, I think the mindset is shifting,” he says. “A lot of MSPs are very much willing to let agentic systems handle mission-critical workflows.”
Duke flies solo
IgnitHQ is seeing a mindshift move toward trust in AI too.
“We have AI doing a lot more than I thought it was going to be doing,” says founder and CEO Yaron Baitch. And not because Duke, the young MSP platform maker’s AI engine, is proving more capable than he expected but because users are proving more willing to let it process tickets on its own start to finish than he expected.
“We are seeing a very large percentage of tickets that would have been touched by a human that are not being touched by a human,” Baitch (pictured) says.
I first profiled IgnitHQ in June, weeks after it began shipping its all-new, AI-native, and surprisingly extensive managed services line-of-business suite. A little over 20 MSPs have migrated to the new platform since then, and another 25 or so are in the process of joining them. Collectively, those partners have processed some 10 million tickets so far, autonomously much of the time.
“Don’t quote me on an exact number, but we’re seeing a very large percentage of tickets that would have been touched by a human not being touched by a human,” Baitch says, adding that it typically takes a little while for MSPs to reach that stage.
“Generally, after about a month of really active, heavy usage with somebody watching, they start to trust it more and more,” Baitch says. So far small and midsize MSPs are more apt than others to develop that trust.
“Which I love to see, because now they’re able to compete in the market against bigger ones,” Baitch says. “They can go after bigger projects without having to increase headcount and still be able to get better results from their ticketing.”
At present, only about a quarter of IgnitHQ’s partners are letting Duke close tickets, when it can, without technician involvement. Another 50% won’t let Duke do anything meaningful without human-in-the-loop approval, and a final 25% pretty much insist on doing everything themselves.
“They still don’t trust AI,” Baitch says. Some of them, he predicts, never will. “There are going to be some that are always the holdouts.”
Straight from the Sunny source
I quoted IT By Design CEO Sunny Kaila on the AI-in-managed-services landscape a couple of months ago, but the best way to get his experience-laden take on the matter is directly from him. Which you can now via the latest episode of MSP Chat, the podcast I co-host. And then, when you’ve finished doing so, you can go here to browse our growing library of other insightful interviews with industry thought leaders.
Rewst appears to be uncaging automation
I made a whole lot of comments about generic RPA tools earlier in this post, you may have noticed, without once mentioning the most familiar maker of RPA tools for MSPs, Rewst.
That’s because Rewst hasn’t been a generic RPA tool maker since June when it shipped a radically new version of its product that builds automations for users on demand in response to natural language prompts. The goal, as I wrote at the time, was to democratize RPA by allowing people who’ve never met an automation engineer, let alone trained to become one, build their own workflows unassisted. The question three months later is, has it?
According to Rewst, yes.
They can’t prove it yet, mind you, but they do have circumstantial evidence. Until a couple weeks ago, when the new version of the platform went GA, there were about 110 MSPs, which is roughly 10% of Rewst’s partner base, running an early access edition of the system. During a single 30-day span, the company says, those users created an average of 150 new workflows per day, which is about twice as many as other users averaged over the same span.
And when I say “other users,” I mean all other users, combined. All 90% of Rewst partners not in the early access cohort collectively produced about 75 automations per day during that 30-day period. The 10% using the new tool collectively created 150. Which doesn’t establish conclusively that amateurs are now building automations for the first time, but sure implies something about the system has gotten a whole lot simpler.
“Our theory is if people are creating more automations on the platform, it must be easier,” says Aharon Chernin (pictured), Rewst’s CEO.
The next three questions about the new platform, of course, are how soon do other Rewst partners adopt it, do they see the same giant uptick in automation building, and are some of those automations created by people formerly unqualified to automate? In the meantime, a few other things make the early access results interesting:
1. AI is becoming Rewst’s primary interface. About 80% of early access automations were built from directly inside ChatGPT, Claude, or another LLM via Rewst’s MCP server rather than the system’s traditional interface. That mirrors a trend Syncro predicted earlier this year and that TD SYNNEX reported more recently. It also suggests that large numbers of MSPs are embracing chatbots as their primary pane of glass, something that felt a little out there when I first speculated about it 15 months ago, at a far faster clip than lots of people expected. Including Rewst.
“We thought it’d be something that would happen six to twelve months from now,” says Chernin, adding that Rewst is shifting development resources from the system’s web interface to its AI one in response.
“It’s become really clear where we need to spend our time,” he notes.
2. AI is becoming Rewst’s primary customer. Well, not exactly. MSPs are still the ones buying and paying for Rewst’s software. But as far as Chernin can tell, partners aren’t asking Claude to build an automation using Rewst. They’re asking Claude to build an automation, period, and Claude is choosing Rewst as the simplest way to do so.
“The MSP isn’t using the product. Claude is,” Chernin says. Which (per an earlier post) means Rewst has to think about keeping agents satisfied these days at least as much as it thinks about keeping partners satisfied.
“When an MSP says, ‘Hey, make me this automation,’ the agent’s not going to pick your product unless you actually make the agent’s life easier,” Chernin observes.
That’s something for anyone rolling out MCP servers or headless versions of their software to consider.
Determinism isn’t dead
Neo Agent, as we saw before, foresees MSPs relying less on deterministic scripting and more on AI in the future. So, it’s worth noting, does Rewst. Indeed, that’s a core design principle in the new version of its product.
“Rewst can be as deterministic or non-deterministic as you want it to be. You just need to tell your agent what you want,” Chernin says. If that’s an old-school RPA flow, ask for it. If it’s something more like Neo Agent, ask for that instead.
“If you can describe it and define it, you can build it,” Chernin says.
Still, Chernin doesn’t expect deterministic workflows to disappear anytime soon. “There’s always going to be tasks that require determinism,” he says, highlighting compliance-related ones in particular.
“When an auditor comes in and says, ‘Show me your processes for doing X, Y, and Z,’” he observes, you’d better have an answer other than “we don’t have processes.”
Over on Business of Tech
Host Dave Sobel has a question about the new FinOps service from TD SYNNEX I wrote about recently:
“When the distributor builds the product, sets the meter, and hands the MSP the brand, who owns the outcome, and who keeps the margin?”
Want to hear my answer? Then join us this Wednesday, October 28th, at 3:00pm ET for my latest live appearance on Sobel’s show right here.
I’m pretty sure he’ll want to discuss Pax8’s new persistent agent too. Should be a good one. Don’t miss it!
Let’s talk
When not writing this newsletter, I’m chief analyst at Channel Mastered, a consultancy that helps vendors build, grow, and optimize thriving MSP channels. Wrestling with a channel challenge? Hit me up here: rich@channelmastered.com.
Also worth noting
Zero Trust AI Control, from ThreatLocker, automatically enforces externally managed policies governing what AI agents can access, execute, and connect to.
Cortex XDR One, from Palo Alto Networks, automatically unifies prevention, detection, response, and investigation of agentic threats.
Singularity Agentic Detection and Response and Singularity Autonomous Security Operations Center, from SentinelOne, are designed to help security teams defend AI environments and automate security operations.
The first of those tools is now available in early access to members of SentinelOne’s PartnerOne program, along with new certifications and an Authorized Deployment Partner Program.
Veeam’s new Sovereign Service Providers Program helps customers meet data sovereignty requirements with help from local service providers.
94% of organizations surveyed by Barracuda report cybersecurity skills gaps. 56%, on the other hand, say they’ve reduced security headcount due to AI.
Shield Blueprints, from Mailprotector, is designed to help MSPs standardize and scale email security deployments without proportionally scaling up administrative workloads.
The latest AI-powered addition to Keeper Security’s browser extension is designed to warn users before they enter credentials on phishing sites, using on-device analysis to preserve privacy.
Last Mile Agents, from Cyware, are designed to help participants in threat intelligence sharing networks identify threats, prioritize vulnerabilities, and determine appropriate responses without deploying Cyware’s full platform.
CyberBOT, from Optiv, is a free cybersecurity education program for schools, nonprofits, and community organizations.
Bishop Fox’s Emerging Threat Testing, Attack Surface Discovery, and Attack Surface Testing are now standalone offerings.
ScalePad and Pax8 say they’ll collaborate on joint product development and initiatives to help MSPs uncover shadow IT and AI. ScalePad software is now on the Pax8 Marketplace too.
Speaking of Pax8, GTIA’s ASCII community has named it ASCII Cup Vendor of the Year for 2026.
Gartner expects overall IT spending to climb 15% globally next year, but IT services spending to rise a comparatively anemic 4.9%.
Criminals have noticed how valuable memory is these days too. According to research from Overhaul, thefts of data center equipment, chips, and computing components have increased 34% since 2024.
TD SYNNEX will distribute tools, training, and enablement for autonomous mobile robots in North America and Europe via a distribution agreement with KUKA Group.
Cognizant and Google Cloud have expanded their alliance to include a new Gemini Frontier Capability Group combining Google’s AI technology with Cognizant’s engineering and industry expertise.
Dell has added new deskside agentic AI solutions, infrastructure, and ecosystem enhancements to its AI Factory with NVIDIA.
The latest enhancements to Zoom’s Zoom Up Partner Program include expanded enablement resources and more help with AI-powered offerings.
CompTIA’s new AgentOps+ certification is designed to define the workforce skills needed to deploy, manage, and oversee AI agents.
The Channel Marketing Association and Channel Sales Association have jointly launched a strategic advisory program designed to give members of both groups access to AI, leadership, and sales experts.