ConnectWise Rebuilds
Start with ConnectWise. The company announced that it is dismantling Asio — the platform it has spent the past several years building — and starting over. The replacement is what ConnectWise is calling an AI-native, code-first operating layer that bundles PSA, RMM, security, and automation into a single unified platform. ConnectWise is naming the objective explicitly: the new platform uses agentic AI to autonomously handle tickets and IT service operations, with the company projecting significant efficiency and margin gains for MSPs that adopt it.. ConnectWise has a specific term for where this is heading — they are calling it Predictive IT, a shift from reactive service delivery to a model where the platform anticipates and addresses issues before they surface.
That announcement arrived alongside a separate and unrelated pattern around AI cost. Axios reported a story— Anthropic declined to confirm the specifics — in which an unnamed enterprise incurred approximately five hundred million dollars in charges on Anthropic’s Claude in a single month after failing to set any usage limits. TechCrunch describes an industry-wide scramble among companies that have deployed AI and are now discovering that usage-based token billing scales in ways that outran their financial models. The Linux Foundation launched a new initiative called the Tokenomics Foundation, specifically aimed at developing standards for tracking and managing AI token costs across the industry.
Freshworks surveyed more than twelve thousand IT professionals and put a number to the broader problem. Mid-market companies are losing roughly twenty-five percent of their AI budgets to what the report calls a complexity tax — the friction of moving AI from pilot into actual working production. That translates to approximately sixteen billion dollars in annual waste across the US mid-market. In the same survey, eighty-nine percent of respondents said they plan to increase AI spending, while only fifteen percent reported that AI is currently embedded in their day-to-day workflows.
Ungoverned Agents
The reason the cost problem and the platform problem are arriving at the same time is the same structural condition: AI is entering organizations through the subscription layer, not through IT governance.
Arena, a San Francisco startup that tracks AI agent activity in production, found agents handling code generation, research, and document review at real volume — and recorded an eight percent bluffing rate, meaning agents producing incorrect outputs with full confidence and no failure signal surfaced to the user. These systems have production workloads. They do not have owners.
The adoption path explains why. A part-time tutor pays twenty dollars a month for Notion AI and uses it for transcription, invoicing, and scheduling. A bankruptcy attorney, profiled by the New York Times, has built what he calls a full workforce of AI agents on an open-source platform called OpenClaw, automating operations that previously required staff. Zoom’s ZoomMate enters the enterprise at twenty dollars per user per month, converting post-meeting conversation into automated workflows across enterprise applications. Each of these is a purchasing decision, not an IT decision — and each one adds new agent behaviors, new access permissions, and new cost exposure to a client environment without a corresponding governance structure.
That is the mechanism. The subscription adoption path — one tool, one user, one workflow at a time — creates operational surface area faster than any organization built its processes to track. When usage-based billing is attached to that surface area, costs scale with agent behavior, not headcount. When access permissions are granted at signup rather than governed continuously, the access scope expands with the tool catalog. No individual, small business, or internal IT function was built to manage that accumulation.
When internal coordination runs out of surface area, something external has to supply it.
Roll-Up Warning
Thrive Holdings just committed one billion dollars to acquire local accounting firms and consolidate them under an AI platform called Current. The model is direct: take a fragmented market of independent practitioners, replace the operational overhead with AI-enabled workflows, and deliver the outcome — accurate books, faster turnaround — at a scale no individual firm could match on its own. Current is reporting ninety-eight percent data-entry accuracy and thirty-one percent time savings at the firms already on the platform. The accounting firms being absorbed did not fail. The market moved to a model they were not positioned to run.
The structural conditions are identical in IT services. A fragmented base of independent providers. Clients who need outcomes managed rather than tools sold. An AI operational layer that is now active across the client base, not experimental at the edge. Outside capital has already spotted that pattern in adjacent markets. Lemhi is building directly for the IT services version of that gap — a platform designed to let MSPs package, sell, and manage AI services as recurring revenue. The co-founder’s stated case is not optimistic framing. It is a warning: MSPs who do not own the AI conversation will be displaced by vendors and consultants who do.
That displacement is not a single event. It is already arriving through capital allocations, new entrants, and client relationships that are quietly being reshaped around AI services the MSP is not yet offering.
An MSP that defines what AI can do inside a client environment, instruments the usage, and prices that accountability as a service is selling something neither the vendor nor the roll-up can replicate. The technology is vendor-supplied. The governance of how it operates, what it accesses, and what it costs inside a specific client environment is not. That is the defensible position. The MSP that waits inherits the complexity without the contract language, absorbs the cost events without the margin, and explains the incidents to clients who will eventually find a provider that governed the stack from the start.
That is the choice. Build the governance layer, or absorb the complexity without being paid for it.
Why Do We Care?
The governance void signaled by ConnectWise’s shift— and that Thrive Holdings is already pricing into its roll-up model — is a service MSPs are currently delivering informally and billing for zero. AI usage governance, spend monitoring, and access scoping have a scope, a delivery cost, and a liability attached to them; they are not support tasks to bundle into a base rate. Before your next renewal or new engagement, name it, define it, and add it as a line item — because someone is going to charge for it, and the first MSP who does frames the market for every client that follows.
What to Consider
Instrument before you price. Before adding AI governance as a line item, audit what AI tools are currently running inside each client environment — which agents have access to what systems, what the current monthly token spend is, and whether any usage limits exist. This audit is billable as a one-time assessment. It also surfaces the liability exposure you’re currently absorbing for free.
Treat the Freshworks 89%/15% finding as an account segmentation tool, not a market data point. The gap between planning AI spend and having it embedded in workflows is your prospecting filter — every client sitting in that gap is accumulating ungoverned exposure without an internal function designed to manage it. Sort your accounts by that delta, prioritize the widest gaps, and lead with the liability question, not the productivity pitch.
Price the liability, not the labor. The $500M Claude bill is an extreme case, but the mechanism — no usage limits, no spend visibility, no access controls — exists at smaller scale in most mid-market client environments right now. An MSP who sets spending caps, scopes agent access, and documents that governance in the contract is not selling hours — they’re selling indemnification from a class of cost event that clients don’t yet know they’re exposed to. That’s a different pricing conversation than managed services has historically had.
If this trend continues, AI governance will become a standard managed services line item before AI productivity becomes a reliably measurable client outcome — because the cost, access, and liability exposure will arrive faster than the ROI proof.