Atera’s AI Co-Pilot is Graduating to Auto-Pilot

Three looks at GenAI in managed services this year and next, plus marketplace value mapping from Forrester and Pax8.

Note: Even channelholics need time off. This particular one will be on vacation next week, but back with a fresh post on the first Friday of the new year. Until then, here’s a supersized post to tide you over.

I recently pointed out how quickly Asio, ConnectWise’s next-gen development platform, allowed the company to add ChatGPT-based scripting functionality to one of its RMM solutions this February. What I didn’t mention is that someone else got there even faster.

About eight days faster, to be precise. Atera announced a nearly identical GPT-based scripting addition to the RMM portion of its unified managed services tool suite on February 1^st. Credit for being (as far as I can tell, anyway) first to market with such a feature probably belongs to the fact that the company had been working on AI-based automation by that point for a decade already.

“Unsuccessfully,” notes Gil Pekelman (pictured), Atera’s CEO.

Vision, he explains, was never the barrier. “We had the framework and we knew what we wanted to do,” Pekelman says. Doing it with home-grown technology, however, led to one dead end after another.

A breakthrough came late in 2022 when Microsoft, a close partner, quietly came calling with something its R&D team was readying, an artificial intelligence platform that would later be known as Azure OpenAI Service. “It actually fit exactly into what we were working on,” Pekelman says.

After years of frustration, Atera had that automated scripting assistant in production within weeks, followed in July by an even more powerful “co-pilot” tool that uses GPT-4 to accelerate easy, time-consuming technical work. Say a user complains of slow PC performance, for example.

“It’ll check your computer and memory and disk and the network and the speed of the internet and even the ISP status,” Pekelman says. It’ll then present summarized results, propose an explanation, and offer up a menu of one-click resolutions.

“It’s like having a junior IT person working for you,” Pekelman says. “It makes everything you do faster and easier.”

Other vendors have introduced AI-based automation tools of their own since then. Kaseya unveiled its Cooper Bots service early in October, for example, and ConnectWise announced its AI companion, named Sidekick, the following month.

Atera, however, is now on the brink of taking co-pilot to the next level with a new feature called auto-pilot, which is currently in beta and due to ship in the first quarter of 2024. Like co-pilot, it uses natural language prompting to collect user input and figure out what’s wrong. But it then resolves the issue without any human intervention.

At present, auto-pilot closes about half of the tickets it encounters autonomously, and escalates everything else to co-pilot. “It’s what we’ve been dreaming about for the last 10 years,” Pekelman says.

And it benefits everyone, he adds. End users get immediate assistance with tickets instead of waiting hours for a technician to call. On average, Pekelman says, auto-pilot cuts response times from one hour to three seconds and improves time-to-resolution from four hours to 10 minutes.

Technicians, meanwhile, get freedom from their least favorite chores (think password resets). “There’s a list of things they hate to do,” Pekelman observes. “Those go away.”

Their boss wins big too. According to Atera, citing data collected during testing, auto-pilot improves operational efficiency ten-fold on average.

Of course, you can forgive MSPs for being, you know, nervous about unleashing an unsupervised, potentially hallucinating digital agent with network and endpoint access on their clients. “Move fast and break things” works fine in a lab but can be dangerous in the real world.

Pekelman, however, insists there’s nothing to worry about. Auto-pilot doesn’t share customer data for training purposes with Microsoft or OpenAI, and uses tunable controls in Azure OpenAI Service to prevent the system from making threats or romantic entreaties. Plus, Atera spent months trying auto-pilot out on several hundred thousand copies of real-world tickets in a safely isolated test environment.

“We worked on it for over a year, again and again and again, until we saw it gets it right,” Pekelman says.

Powerful stuff, he continues, but also just a little scary. Generative AI doesn’t really think, but it sure looks that way when you watch it diagnose tickets. And there’s much smarter AI coming. Earlier this year, Pekelman met Sam Altman and Ilya Sutskever, the OpenAI exec prominently involved in Altman’s brief ouster as CEO last month.

“They were talking about the inevitability of a super intelligence,” Pekelman says, and promising it will usher in a magical era of easy answers to enormous problems humans have never solved.

Unless, of course, it wipes us all out instead.

Tipping toward security

Atera and ConnectWise are moving quickly (yet safely, they insist) on AI. Kaseya, as I’ve reported before, is progressing more cautiously.

So is N-able. Close to a year after some of its peers, the company added generative AI-based scripting to its N-central RMM solution last week. That’s its first step into AI-powered automation and a reflection of where it sits on the speed-versus-safety spectrum, according to Chief Technology and Product Officer Mike Adler (pictured).

“The balance is going to be on security,” he says.

If anything, he continues, N-able’s first AI functionality enhances safety as much as it does productivity. Technicians are already using ChatGPT to write scripts, Adler notes. They’re just doing it in the least secure way possible—outside N-central using private OpenAI accounts.

“The risks associated with the information you give GenAI are significant,” Adler observes. “[You] might put in credentials, information about your customers, networks, IP addresses.” That’s catnip to attackers and not the kind of data you want ending up in training models. N-able’s new GenAI functionality allows partners to capitalize on earlier OpenAI investments without exposing their clients to harm by essentially integrating N-central with existing ChatGPT subscriptions.

“Now you’re generating code directly from within the [RMM],” Adler says, without leaking data along the way. “Security, at the end of the day, has to be a primary thought process for every MSP.”

Though N-able is new to ChatGPT, Adler emphasizes, it’s far from new to artificial intelligence.

“This is not the beginning of AI in our RMMs. This is the introduction of generative AI into our RMMs,” he says. “Our technologies have used machine learning and AI in various forms to help technicians discover issues, suggest remediations, those kinds of things” for years.

There’s much more to come, according to Adler, who while declining to get into specifics, provided some pretty suggestive hints.

“Using the datasets that customers have, can we build predictions? Can we help them understand the work that might be forthcoming? I think that’s a place that we can help MSPs be more successful,” he says. Same goes for suggesting untapped opportunities to boost productivity.

“Generative AI and AI in general could help recommend automation and orchestration ideas that we witness organizations doing repetitively on a regular basis,” Adler says.

Oh, by the way…

It won’t be official until January, but N-able is the latest name-brand vendor to add MDR, joining SonicWall and WatchGuard among others just in the last few weeks.

“It’s definitely something that we see strong demand for in the marketplace, and frankly more and more need, more and more desire, more and more ability of MSPs to take this on to help grow their businesses,” Adler says. Same here, Channelholic replies.

Co-toddler

Back to AI. Leaning on vendors to get started there will probably save you time, but it’s not mandatory. At least one MSP I know is taking a DIY approach instead, starting with an infant co-pilot.

“We’ve set up our own GPT server that we’re going to start pumping a lot of our data in,” says Brian Weiss (pictured), CEO of San Luis Obispo, Calif.-based ITECH Solutions. “We treat it like a little toddler that we’re training to know everything about our clients and our business, so it can be like our little assistant by our side guiding us through our day.”

And that’s a small piece of the ambitious AI work underway at ITECH. With help from two outside development firms, the company is building a custom PSA solution tailored to its workflows using Microsoft Dynamics software.

“It makes sense to make our PSA live in the Microsoft world because it’s in the same dataverse as most of the data that we need access to as a single source of truth for our clients,” explains Weiss, who looks forward to querying that database for fast answers to everyday questions.

“It’s a lot quicker just to ask the AI to find something for you versus trying to search for it in a documentation system,” he says.

Coming eventually as well is a GPT-powered bot that will use software from Rewst, the hyperautomation vendor we’ve written about here a few times, to gather details about technical problems automatically, either inside a self-serve customer portal or the customer’s Microsoft Teams tenant.

“That’s going to allow us to triage faster,” Weiss says.

Further AI functionality will automatically flag service issues that come up frequently. “What are we spending human hours on that could be scripted?” Weiss asks. Or better yet fully automated, a la Atera’s auto-pilot.

“We don’t even need a human involved at that point,” Weiss observes.

Like everyone else quoted in this post, ITECH is taking a safety-first approach to all this. “It’s part of the reason we set up our own private GPT servers,” Weiss says. “I want to use the technology, but I want to have full control of who can access the data.”

With help from Rewst, everything AI does will be meticulously documented as well, he adds. “You’re never going to get that from a human.” And the resulting audit trail will make explaining and then fixing any mistakes AI is responsible for relatively easy.

“If a human makes a mistake, it could be a can of worms sometimes to figure out exactly what went wrong,” Weiss notes.

All this in-house effort is preparing ITECH to start charging clients for AI project work too. “Once we do this internally for ourselves, there’s no reason why we shouldn’t take all that experience we just gained and start duplicating that,” says Weiss, who expects to build a developer team for those projects soon.

“Otherwise, we’re going to be stuck in a commoditized world.”

Measuring marketplace value

Or maybe locked out of deals altogether. The rise of “product-led growth” has many SaaS vendors thinking they can get by without partners so long as they have great software and an easy signup experience.

If only it were that simple. Yes, buyers increasingly want self-serve online purchasing and provisioning. But as we’ve noted multiple times here in recent months, what they really want aren’t products but solutions with all the software, hardware, and services required to achieve specific business outcomes. Subscribing to something on Slack.com, while undeniably convenient, won’t get them there.

What will get them there and keep them coming back, according to cloud distributor Pax8, is a solution-friendly online marketplace offering mix-and-match applications and partner services in a single interface with unified delivery and billing. Per my writeup back in June, the company is hard at work on precisely such a platform.

A report from Forrester published earlier this month suggests they might be on to something. Among other things, the study cites three specific ways marketplaces can benefit vendors and their partners:

1. “Higher lifetime value from increased customer appeal” — marketplaces aren’t just an easy way to shop. They’re an easy way to shop for exactly what you want, and get it all at once in one place. Businesses like that, resulting in more revenue for both vendors and partners over a longer customer lifespan.

“This is not just a marketplace for marketplace purposes,” says Chief Strategy Officer Ryan Walsh (pictured) of Pax8’s forthcoming site. “It’s also a customer experience and loyalty enabler.”

2. “Longer-term value creation” — Sellers aren’t alone in getting long-term value from marketplaces. Buyers do the same via recurring services for existing solutions in areas like training, management, security, backup, and reporting, plus ongoing assistance designing and implementing new solutions.

“The value creation of yesterday was, ‘I’m just going to buy it for you.’ That’s not good enough anymore,” Walsh notes. “Everyone in this ecosystem has to rethink how they’re adding value.”

3. “The option to move away from relying solely on your product margins” — Good luck building a business around the profits you’ll make reselling software and infrastructure as a service. Microsoft pays meager margins on its cloud software. Salesforce pays nothing. Marketplaces help IT providers turn selling low-margin products into a high-margin opportunity through services.

“The way that progressive and modern partners are making their money is the value-add attach,” Walsh says.

Pax8’s marketplace of the future remains under development at present. Partner-focused tools for functions like quoting deals and identifying cross-sale opportunities will arrive across the first half of 2024, Walsh says. Vendor-focused features, including self-serve profile and campaign creation tools, will begin appearing late next year and continue rolling out in 2025.

Also worth noting

• The mobile app for Ingram Micro’s Xvantage platform, previewed in September, is now generally available.

• Cisco has acquired Isovalent to amp up its cloud networking and security game. Not a shock given how much cloud, networking, and security are on the company’s mind.

• Stellar Cyber is teaming up with BlackBerry and SentinelOne on XDR.

• Good news for a niche audience. Channelholic readers in Western Canada who like AWS now have an infrastructure region of their own.