Remember the old film “Play Misty for Me”? Well, Juniper may be revising it to “Play Mist for Sales”, at least to channel players. In a June CRN article, their head of partner programs says “Our Mist AI native networking platform—it’s that gateway into predictable as-a-service growth because that for me is the future and what customers are looking for, and our technology just plays so well into that.” They may be right.
I don’t get much direct input from the channel partner side; my Andover Intel comment mailbox is for users of technology only. I do get input from the SMB space, from 128 since we opened up the mailbox in 2023. What I find is that the SMB space may be especially interested in AIops, and have more to gain from its adoption.
SMBs have an exaggerated form of the problem that most tech user organizations face, which is difficulty in acquiring and maintaining support talent for both IT and network operations. In most cases, these smaller companies can’t offer competitive salaries or a career path that even rivals enterprise users, much less vendors and operators. The growing role of the Internet in sales and customer support has made these firms increasingly dependent on network technology, to the point where some say that their network problems are just as acute as those of enterprises. In my 128 contacts among SMBs, 84 are users of cloud computing, and 112 have at least one server cluster that you could reasonably call a data center. Most of these (95) get their technology from a channel partner rather than through direct sales.
Channel partners have their own reason to be interested in AIops, too. Few companies of any size are immune to the pressure to grow revenues in the face of what’s generally rising costs of goods, technology, and people. None are immune to pressure to at least keep profits stable, and so the channel partners are always looking for an upsell. Managed services are one area that’s always raised a lot of interest.
The problem with managed services is that it relies on the provider being able to create a pretty significant economy of scale in service management and operations. Many channel partners are smaller than their larger customers, and even though any tech provider can normally attract and hold specialists a bit easier than a user organization, there is still a question of whether they can generate enough revenue in managed services based on willingness to pay, unless they achieve a higher level of productivity than the workforce they can retain is likely to offer alone. AI augmentation sounds good to them, for sure.
Juniper’s Mist AI platform was originally a LAN management tool, but the company is extending it across their whole product line. LAN management isn’t cited as a particular problem by SMBs in isolation, but QoE issues are often cited, and are seen largely as a result of the interplay of behaviors of all the network services and devices they use. Thus, to manage them they really need an end-to-end, holistic, view.
The problems associated with fault determination, isolation, and recovery tend to be proportional to the square of the complexity of the configuration. Thus, it’s not particularly difficult for a mid-sized business (and even a few tech-centric small businesses) to have more operations issues than they can deal with in-house. They might think of using AI themselves, but almost all my SMB contacts think that getting AI as a product rather than getting an AI-based service is only going to add AI complexity to their overall complexity problem.
Channel players, based on my past experience and even my own experience as one, would see something like Mist as a potential solution to both their problem and their customers’ problem. They could be right, too. I have more Mist AI users in AIops missions than any other solution, and they seem to have a favorable view of it, and a hope that Juniper makes it even broader in what it can do, and what kind of operations it can support.
It’s fair to speculate on whether Mist might expand into the data center and cloud operations space now that the DoJ has settled their suit against the HPE/Juniper deal. The QoE-centric way that SMBs look at things make that fairly likely, but the HPE deal would also raise a challenge for Juniper’s Mist planning, and not just for SMBs. Will even a limited opening of Mist to others make the new company more or less interested in Mist? I doubt it would in any way limit HPE’s desire to leverage Mist as far as it can, but it might raise the question of whether the new combined company would support other vendors’ gear and perhaps even integrate software-platform management.
One thing that’s true for SMBs and less true for enterprises is a vendor-homogeneous data center and network (though not usually from the same vendor). If you broaden AIops to capture more pieces of the stuff that creates QoE, you raise the need to support foreign gear. Enterprises tend to see management as being tied to the platform being managed, including network management. They say that in part because they believe that there is no stronger invitation to finger-pointing and feet-dragging than to take a network vendor a supposed fault that a competitor’s management framework claimed to detect. SMBs, having less vendor mix, has less reason to worry about this, but sticking IT into the picture could change their mind on that. Since SMBs tend to keep hardware longer than enterprises (at least half-again as long, and a third say even twice as long), it’s more difficult to convince them to switch vendors and displace gear that’s running.
Juniper obviously expects a broader Mist focus to help enterprise and even service provider sales, and for those missions breadth of compatible equipment would be even more important. The latter group may be especially critical, because managed service providers (MSPs) might in theory be able to push a Mist platform as a way of improving their management economy of scale and thus raising their profit, if Mist has a wider supported-gear footprint.
One interesting point, raised by an enterprise in a chat, is whether AIops might have an impact on the network or other devices. This comment was made specifically with regard to LLMs/generative AI use of the Model Context Protocol (MCP), but the enterprise who made it pointed out it could also be a factor with even some uses of machine learning (ML).
MCP allows an LLM that’s been generally trained to pull specific data from a collection of sources. This is something enterprises are always interested in because they need their own business data to generate the most useful insights. It’s also critical for AIops because that application necessarily requires injection of real-time data into the model.
The enterprise ran some tests using a self-hosted LLM model and MCP both to query a database and to invoke real-time requests of management interfaces. They found that it worked as long as you dipped into a database for device status rather than initiating polling of the device. Otherwise, complex issues that could involve a wide variety of gadgets could result in the model making so many status requests that it loaded the devices and management systems. You don’t have control of how a model evaluates information, and thus how often it might dip into real-time data. I haven’t heard that issue come up for Juniper or any other vendor’s own AIops stuff, but it’s worth thinking about.
AIops is surely one of the most potentially valuable missions for AI, but enterprises are realizing that it’s not just a simple matter of asking a chatbot “What’s wrong with my network and how do I fix it?” Hopefully, this will drive more innovation in the AIops area, and result in even better business outcomes.
