MIT Sloan Says Platforms Must Onboard Agents — Not Just Users. Most Aren't Ready.

Marshall Van Alstyne dropped a line at the 2025 MIT Platform Strategy Summit that should keep every SaaS executive awake: "You are going to launch a marketplace, you're going to have to onboard and get critical mass for agents. You're going to have to design interfaces for agents. You're going to have to sell to agents."

This isn't speculative futurism. Gartner forecasts 40% of enterprise applications will embed task-specific AI agents by end of 2026 — up from under 5% the prior year. Microsoft reports 80% of Fortune 500 companies already deploy AI agents in some capacity. IBM's 2026 predictions describe "super agents" with control planes managing tasks across environments. The agents are arriving. The platforms they need to interact with are still designed for humans clicking buttons.

Van Alstyne — an MIT IDE digital fellow and Boston University professor — identifies the core architectural challenge: platforms built for human users assume human interaction patterns. Search bars, dropdown menus, pagination, visual dashboards. Agents don't use any of that. They need APIs, structured data formats, programmatic access to every function, and governance interfaces that define what they're allowed to do autonomously versus what requires human approval.

This is the "agent-readiness" gap. Your CRM has a beautiful dashboard. Can an agent read, write, and reason over it programmatically? Your project management tool has drag-and-drop boards. Can an agent create tasks, assign them, track dependencies, and escalate blockers without a human intermediary? Your marketplace has a checkout flow. Can an agent evaluate options, negotiate terms, and transact within defined parameters?

The MCP (Model Context Protocol) movement addresses part of this — standardizing how agents connect to tools. Red Hat integrated MCP into OpenShift AI 3.0. CData calls 2026 "the year of enterprise-ready MCP adoption." But MCP is a connection standard. Agent-readiness requires something deeper: redesigning platform workflows, permissions, and governance for non-human users.

Van Alstyne surfaces the hardest question: "What happens when agent decision ability exceeds its formal authority?" An agent might be smart enough to negotiate a better vendor contract but lack the organizational authority to approve it. A coding agent might identify a production bug and know the fix but lack deployment permissions. The gap between capability and authority is where governance becomes critical — and where most platforms have zero infrastructure.

The hidden cost dimension makes this urgent. Geoffrey Parker, speaking at the same MIT summit, warned that AI-generated code is accelerating technical debt. Companies reporting big productivity gains from AI coding tools are discovering a delayed cost: poorly integrated, hard-to-maintain code that compounds over time. The same pattern applies to agent-platform interactions built with shortcuts. Quick API wrappers and screen-scraping integrations work for demos. They create technical debt at scale.

The companies that will win the agent era aren't just building agents — they're building agent-ready infrastructure. Platforms that treat agents as first-class users with structured onboarding, governed permissions, audit trails, and purpose-built interfaces will become the operating system for the agent economy. Platforms that don't will watch their agent-enabled competitors move 10x faster.

At Seven Olives, we build agent teams that interact with your platforms through governed, auditable interfaces — not brittle screen-scraping or ad-hoc API wrappers. We also consult on making your existing platforms agent-ready: designing the permission models, API surfaces, and governance infrastructure that turns your software into agent-native infrastructure. The platforms that adapt first will capture the agent economy. The rest will be the equivalent of businesses that never built a website.