You open your editor. Write some code. Ask the AI for help. It autocompletes, suggests a refactor, maybe explains a function. You accept or reject. Close the laptop.

That's been the loop for two years. Every AI coding tool — Cursor, Windsurf, Copilot, whatever JetBrains named their thing this month — competes on the same axis: how good is the autocomplete, how fast is the chat, how well does it handle context. They're all fine. Roughly the same. Differentiation at the margins.

The conventional wisdom says this is where the competition stays. Better models, faster completions, bigger context windows. Incremental forever.

Early March 2026 broke that assumption. Nearly a month later, the dust has settled enough to see what actually happened — and what it means.

The Gap Nobody Talks About

Here's the thing nobody says out loud: an AI that helps you write code is useful. An AI that actually does work — watches your Slack, reacts to GitHub pull requests, handles PagerDuty incidents at 3am — is a different category entirely.

The first is a tool. The second is infrastructure.

Cursor decided it wants to be infrastructure. Four weeks in, we can start evaluating whether the bet makes sense.

Two Releases, Three Days Apart

Cursor 2.6 shipped on March 3. The headline: MCP Apps. MCP (Model Context Protocol) is a universal plug standard for AI tools — think USB-C, but for connecting AI agents to external services. MCP Apps let MCP servers render interactive UI components — charts, diagrams, whiteboards — directly inside the agent chat. Ask about your codebase's performance and get a live Amplitude chart. Ask about a component and get a Figma diagram. The agent doesn't just answer — it shows.

Alongside that, Team Marketplaces launched: admins on Teams and Enterprise plans can publish private plugins — bundles of MCP servers paired with "skills," which are instructions that teach the agent how to actually use each service. Over 30 public plugins shipped at launch — Atlassian, Datadog, GitLab, Glean, Hugging Face, monday.com, PlanetScale.

Cursor Automations followed on March 5. Always-on agents triggered by GitHub PR events, Slack messages, PagerDuty incidents, or cron schedules. Fully MCP-native. No custom glue code. Your agent watches for a PR, runs a review, posts a comment, opens a Linear ticket — without you touching a keyboard.

Why MCP Apps Are the Real Move

MCP Apps look cosmetic at first glance. They're not.

Remember how VS Code won the IDE wars? Extensions. Millions of them. A marketplace. Third-party integrations for everything. Cursor just pulled the same maneuver — but for the AI agent layer. The marketplace is the strategic move. The interactive UI is the demo that sells it.

When an agent "knows how to use Datadog," it means someone wrote a plugin that tells it what queries to run, what metrics matter, how to interpret results. That institutional knowledge — the kind that normally lives in one senior engineer's head — now lives in a plugin. Shareable. Versioned. Your org's admins govern it.

That's a platform play. Not a feature update.

The Price Tag (There's Always One)

Team Marketplaces need Teams or Enterprise pricing. Solo devs won't touch that layer for a while.

Automations look impressive in demos, but require real setup — connecting Slack, configuring triggers, deciding what the agent is and isn't allowed to do on its own. You're now making security and permissions decisions about an autonomous agent. That's not a one-click experience. That's a whole new category of "what could possibly go wrong."

MCP proliferation is a real concern too. Every tool ships MCP servers now. The protocol won — nobody disputes that. Whether every MCP implementation is well-designed is a separate conversation entirely. More plugins means more surface area for things to break.

And Cursor's $9.9 billion valuation on 360,000 paying users? That math works until it doesn't. About $27,500 per user in implied value. For context, that's roughly the GDP per capita of Poland.

What This Actually Means For You

Solo developer on free or Pro: most of this doesn't affect you today. MCP Apps work, but the compound value lives in teams sharing curated plugins. You get the UI. Teams get the ecosystem.

On a team: the question shifted. It's no longer "which AI writes the best code?" It's "which tool fits into our actual workflow?" A tool that lives in GitHub, Slack, Linear, and PagerDuty — and acts there autonomously — doesn't compete with other IDEs. It competes with your DevOps platform.

The IDE Wars Are Over

Every tool writes decent code now. That game ended in a draw. The new competition is about who owns the workflow layer — who your agents call when something breaks at 3am, who orchestrates the response, who files the ticket.

In early March, Cursor raised their hand. A month later, the question isn't whether the direction is right — it obviously is. The question is whether 360,000 paying users can support a $9.9 billion bet on becoming the operating system for AI-assisted development.

The editor was always just the front door. Cursor bet the real money on everything behind it. Four weeks in, nobody else has even walked through it.