Stack Fragmentation
The signal is that core IT delivery is getting more fragmented, not more unified, even as customers expect it to behave like a coherent system.
First, let’s start with the managed services market itself. Channel Dive cited new Westcon-Comstor research based on a survey of 500 senior MSP and cloud partner decision-makers, conducted between December 2025 and January 2026. In that data, nearly a quarter of respondents pointed to cloud migration and management as a primary revenue opportunity, and more than 30% said their biggest challenge is keeping data flowing cleanly across platforms. Security and governance pressure is right behind it. That is a pretty direct snapshot of the environment MSPs are operating in: hybrid is the default, and the complexity is measurable.
And that complexity is not staying fixed. New bundled platforms are still entering the core stack, which means more choices, more overlap, and more governance pressure at the point of delivery.
ITPro reports Proton’s launch of Proton Workspace, a privacy-focused alternative to Google Workspace and Microsoft 365. It offers email, calendar, cloud storage, a word processor, spreadsheets, and add-ons like VPN and password management. Priced per user with tier options, Proton sees it as a credible business choice for organizations wary of big tech data practices. The point is not Proton specifically. It is that the core productivity layer is still fragmenting, which gives customers more options but gives operators more environments to govern.
Human-Bounded AI
InformationWeek notes that decision frameworks are changing, as the old “build versus buy” model no longer fits modern IT. Workflows, data, and AI integration evolve quickly, with Gartner predicting many enterprise apps will include task-specific AI agents and hybrid approaches combining packaged platforms with custom work.
The real driver here is not that the tools are getting smarter. It is that work inside organizations is not getting more coherent. AI is landing in environments where the process is fragmented, the roles are unclear, and the handoffs are already messy, so the technology ends up amplifying inconsistency instead of smoothing it.
Forrester’s data makes the point cleanly: prompt-engineering proficiency barely moved, from 22% to 26%, while most employees say AI is increasing workload even as leadership expects more output. That is not a model problem. It is an operating model problem.
And when companies do get AI adoption to stick, the pattern is not autonomy. It is controlled execution. VentureBeat points to Intuit’s AI agents hitting an 85% repeat usage rate, and the key detail is that Intuit attributes that success to keeping humans involved. Not as a fallback. As a design requirement. Reliability comes from a controlled workflow in which people remain responsible for judgment and exceptions, and automation is bounded.
Coordination Tax
AI is that automation is no longer a feature you “turn on.” It is a layer of operations that has to be governed, and the governance burden is going to land somewhere. If it does not land cleanly inside the customer, it lands on whoever is responsible for keeping the business running. That is you.
One proof point is what Front calls the “coordination tax.” In their research, 93% of companies are using AI in customer operations, and yet 71% reported significant AI-related issues in just the past three months. Teams are spending almost three hours coordinating work for every hour they spend actually solving customer problems. That is not an AI maturity story. That is an operating model story. If the customer’s internal coordination is already brittle, adding AI often increases the number of handoffs, exceptions, and “who owns this now” moments. And when those moments hit production, the MSP is the one who gets the escalations, the after-hours calls, and the “why is this broken” meetings. The coordination cost does not disappear. It just moves.
The second proof point is the rise of “shadow AI,” and the fact that the market response is not “better prompts,” it is governance tooling. VentureBeat covers Kilo’s launch of KiloClaw for Organizations, explicitly aimed at pulling personal AI agents back under enterprise identity, access controls, audit logs, and billing visibility. The point is not that Kilo wins. The point is what the product category admits: organizations cannot keep autonomous automation coherent without formal control surfaces. They need someone to define what agents can do, what data they can touch, and how to shut them off when something goes sideways.
So here is the fork for MSPs. Either the MSP becomes the provider that simplifies and governs the automation layer, packaging the controls, the guardrails, the identity, the monitoring, and the human handoffs as a managed service, or the MSP becomes the silent absorber of complexity, eating the coordination tax and exception handling as uncompensated work until margins give out.
Why Do We Care?
The mistake MSPs make is treating AI like another tool in the stack instead of a new layer of operations.
AI doesn’t just produce outputs. It creates exceptions, handoffs, access questions, and failure modes that someone has to manage. And in most environments, that “someone” defaults to the MSP.
If your contracts and service definitions still treat AI as software you support, then you are implicitly agreeing to absorb the coordination, governance, and exception-handling work it generates—without pricing it, scoping it, or controlling it.
That is where the margin erosion shows up. Not in licensing. In labor you didn’t plan for and responsibility you didn’t explicitly take on.
So the fork is simple. Either you define and sell the control layer—how agents are governed, monitored, and escalated—or you inherit that responsibility by default.
And if you inherit it, you don’t get to meter it. You just get to absorb it.
What to Consider
- Audit your current contracts for AI exposure now. Identify every client running M365 Copilot, any AI-embedded SaaS tool, or shadow AI (ChatGPT Teams, Claude, etc.) and map whether your SLA covers the coordination and exception-handling those tools generate. If it does, you’re already absorbing the coordination tax without billing for it. 1
- Treat governance tooling vendors as temporary infrastructure, not strategic partners. KiloClaw and similar products are filling a gap that Microsoft, Google, and others will eventually close natively. Use them tactically to deliver the service; don’t build your practice identity around any single vendor in this space. 1
- Use the KPMG skills gap data as a sales tool, not just an internal warning. When 50%+ of executives admit they expect to be manually managing AI agents for years, that’s a client conversation opener — not a statistic to file away. Lead with it in QBRs to reframe AI governance as a current operational need, not a future consideration
If this trend continues, MSP contracts will add an explicit “AI operations” schedule within 12–18 months: required identity controls for agents, audit log retention, approval workflows for high-risk actions, and a separately priced exception/escalation tier—because customers will demand governance outcomes, not tool access.