Microsoft learned a brutal lesson from Windows Recall, the AI feature it tried to force onto every PC and then had to claw back amid a privacy revolt. At Build 2026, it applied that lesson in the most Microsoft way imaginable: instead of cramming AI into the terminal millions of developers already trust, it shipped a separate, opt-in, open-source fork. Intelligent Terminal arrived June 2 as a standalone MIT-licensed app that drops an AI agent pane into the command line, supports Claude Code, OpenAI Codex, and Google Gemini out of the box, and leaves the stable Windows Terminal untouched. The quiet ambition behind the modest 0.1 version number is to take back the AI-native terminal from Warp, the startup that has owned the category while Microsoft watched.
What Actually Happened
Intelligent Terminal is an open-source fork of Windows Terminal, released under an MIT license and authored by Windows product manager Hamza Usmani. It shipped at Build 2026 as version 0.1, available through the Microsoft Store, the winget package manager, and GitHub, and it installs alongside the standard Windows Terminal as an opt-in tool rather than replacing it. The signature feature is a dockable agent pane, toggled with Ctrl+Shift+period, that brings an AI agent directly into the shell. The terminal automatically detects errors, offers one-click context loading so the agent can see what just failed, and supports background tabs so an agent can grind through a multi-step task while the developer keeps working elsewhere.
The architecture is the genuinely interesting part. Intelligent Terminal speaks the Agent Client Protocol, communicating with agents over JSON-RPC 2.0 through standard input and output, which means it does not bind itself to any single AI vendor. At launch it supports the GitHub Copilot CLI as the default, plus Claude Code, OpenAI Codex, and Google Gemini, auto-detecting whichever compatible agents are already installed on the machine. Microsoft notes that more than 25 agents already supported the protocol as of early 2026. Crucially, the terminal itself makes no cloud calls; it passes shell context to a local agent CLI subprocess, so whatever data leaves the machine is governed by the agent the developer chose, not by Microsoft.
The fork strategy is a deliberate response to recent history. Microsoft explicitly kept the experimental AI code separate from mainline Windows Terminal after the Windows Recall backlash showed the cost of forcing AI on users without real opt-in controls. The company has a graveyard to prove it is serious about not repeating mistakes: AI Shell, an earlier interactive shell with language-model agents, was archived in January 2026, and the experimental Terminal Chat Copilot integration was deprecated. At launch Intelligent Terminal is Windows-only, requiring Windows 11 build 22621.6060 or later, with no local inference mode and WSL2 agent support still on the roadmap rather than in the 0.1 release.
Why This Matters More Than People Think
The terminal is the last unconquered surface of the developer workflow. AI has colonized the editor through Copilot and Cursor, the pull request through automated review, and the browser through chat assistants, but the command line has remained stubbornly human. That matters because the terminal is where the real work lands: builds, deployments, database migrations, and the debugging sessions where a developer is most stuck and most willing to accept help. Whoever owns the AI agent in the terminal owns the moment of maximum developer frustration, which is also the moment of maximum willingness to pay. Microsoft is moving to claim that moment before it becomes Warp's permanently.
The protocol choice signals a strategic shift in how Microsoft thinks about AI lock-in. By building on the open Agent Client Protocol and shipping the default as the GitHub Copilot CLI while still supporting Claude Code, Codex, and Gemini, Microsoft is betting that owning the surface matters more than owning the model. This is the inverse of the Project Polaris strategy, where Microsoft fought to own the model underneath Copilot. In the terminal, Microsoft is content to be the neutral container that any agent can plug into, because the container is what developers open every day, and a neutral container attracts the developers who would never accept a Microsoft-only tool.
There is a trust dividend in the design that is easy to overlook. By making the terminal itself call no cloud APIs and merely shuttle context to a local agent subprocess, Microsoft sidesteps the exact privacy objection that sank Recall. The data-handling story becomes the agent's responsibility, not Microsoft's, which lets security-conscious enterprises adopt the terminal while choosing an agent with the guarantees they need, such as the zero-data-retention agreements that come with GitHub Copilot Business and Enterprise plans. For a company still rebuilding developer trust after Recall, shipping an AI tool whose default posture is to send nothing to Microsoft is a quietly radical act of restraint.
The timing also reframes a developer-economics shift that lands the same week. GitHub moved Copilot to usage-based billing on June 1, and agentic terminal sessions burn through far more tokens than the autocomplete completions developers grew up paying a flat fee for. A terminal that makes running an agent against your shell a one-keystroke habit is also a terminal that quietly raises the metered ceiling on a monthly bill. Microsoft is betting the productivity is worth it, but the broader point is that the command line, long the one free corner of the developer toolchain, is being pulled into the same token-metered economy that now governs editors and code review, and Intelligent Terminal is the on-ramp.
The Competitive Landscape
The clear target is Warp, the venture-backed startup that turned the terminal into an AI-native, block-based, collaborative environment and built a passionate following among developers frustrated with decades-old tools. Warp has had years to refine its agentic experience, its command search, and its workflow sharing, and it enters this fight with a polished product and genuine mindshare. Microsoft's own admission that it spent six months improving mainline Windows Terminal's core features to compete with Warp is a tell: the giant noticed the startup, and Intelligent Terminal is the second front in that campaign, aimed squarely at the AI experience Warp pioneered.
Beyond Warp sit the agent CLIs themselves, which are quietly becoming platforms. Claude Code, OpenAI Codex, and Google Gemini's command-line tools each want to be the primary interface a developer talks to, and Intelligent Terminal's protocol-based neutrality is both an olive branch and a power move. It welcomes all of them while making the terminal, not any one agent, the thing the developer launches. Wave Terminal, iTerm2, and the broader ecosystem of terminal emulators round out a field that was sleepy for twenty years and has suddenly become one of the more contested patches of developer real estate in 2026.
The instructive historical parallel is Visual Studio Code. A decade ago Microsoft was a proprietary-tooling company that developers tolerated rather than loved, and VS Code changed that by being open source, extensible, free, and genuinely good, until it became the default editor for a generation. The Intelligent Terminal playbook is recognizably the same: open the source, embrace rival tools through an open protocol, ship it free, and win the surface by being the most welcoming option rather than the most locked-down. If it works, Microsoft repeats the VS Code trick in the one developer surface where it does not already dominate.
Hidden Insight: Microsoft Is Selling Restraint as a Feature
The most counterintuitive thing about Intelligent Terminal is what it refuses to do. It does not run its own model. It does not call the cloud. It does not replace the tool you already use. It does not even ship as a feature of Windows Terminal proper. In an era when every company is racing to embed AI more deeply and more invisibly into everything, Microsoft shipped a tool whose entire design philosophy is opt-in minimalism. That is not timidity; it is a calculated reading of where developer trust actually sits after two years of AI being forced into products that did not need it.
This restraint is itself the product strategy. The fork keeps experimental code away from the millions who depend on a stable terminal, which means Microsoft can iterate aggressively on version 0.1 without risking the trust of its mainline users. The open protocol means developers never have to bet on Microsoft picking the winning model, because they can swap Claude for Codex for Gemini as the leaderboard shifts. The local-subprocess architecture means the privacy story is in the developer's hands. Each of these choices trades short-term control for long-term adoption, which is exactly the trade Microsoft refused to make with Recall and paid for dearly.
The bear case, however, is that Microsoft has a documented habit of abandoning exactly these kinds of experiments. AI Shell was archived in January 2026. Terminal Chat was deprecated. The graveyard of half-finished Microsoft developer tools is deep, and a 0.1 release authored by a single product manager, Windows-only, without WSL2 support or local inference, looks precarious next to a funded startup whose entire company is the product. The risk developers are weighing is not whether Intelligent Terminal is good today, but whether it will exist and be maintained in eighteen months, or whether it joins AI Shell in the archive once the Build 2026 news cycle fades. A startup that lives or dies by its terminal will out-ship a side project every time, and developers know it.
There is a structural tension that compounds the abandonment risk. Intelligent Terminal arrived one day after GitHub moved Copilot to usage-based billing on June 1, and agentic terminal sessions consume far more tokens than ordinary completions. The tool that makes it trivially easy to run an agent against your shell is also the tool that makes it trivially easy to run up a metered bill, and Microsoft has an obvious incentive to default to the Copilot CLI that bills back to Microsoft. Skeptics point out that the open-protocol neutrality is real today but could quietly erode if the economics of steering developers toward the in-house, revenue-generating agent prove too tempting to resist.
What to Watch Next
In the next 30 days, watch the GitHub repository. Because Intelligent Terminal is open source, its health is publicly legible in a way most products are not: the commit cadence, the number of external contributors, the speed of issue triage, and whether Microsoft staffs it with more than its single founding product manager will tell you whether this is a funded bet or a hackathon project. Watch the star count and the fork count too, because for a developer tool those are the earliest signals of whether the community is adopting it or merely admiring it from a distance.
Over the next 90 days, the roadmap items to track are WSL2 support and local inference. WSL2 is where a vast share of Windows developers actually do their Linux work, and an AI terminal that cannot reach into that environment is missing the workflow it most needs to serve. Local inference matters for the privacy-and-cost story, because a terminal that can run a small model without any cloud round trip would be a genuine differentiator against Warp and the cloud-bound agent CLIs. Whether Microsoft ships these quickly will reveal how seriously it is investing past the launch keynote.
On the 180-day horizon, the question is whether developers actually switch their daily driver. Terminal choice is among the stickiest decisions a developer makes, often unchanged for years, and an AI pane alone is rarely enough to pry someone away from a tool wired into muscle memory. Watch for the qualitative signal in developer surveys and community threads: are engineers reporting that they replaced Warp or their stock terminal with Intelligent Terminal, or are they trying it once and drifting back? Adoption of a terminal is measured in habits, not downloads, and habits take a full quarter or two to reveal themselves. The download spike from a Build keynote tells you almost nothing about who is still launching the app in October.
Microsoft shipped an AI terminal whose boldest feature is everything it deliberately refuses to do.
Key Takeaways
- Open-source MIT fork of Windows Terminal, shipped as version 0.1 on June 2 via the Microsoft Store, winget, and GitHub, installing alongside the stable terminal.
- Agent Client Protocol over JSON-RPC 2.0 lets it support Claude Code, OpenAI Codex, Google Gemini, and the default GitHub Copilot CLI without vendor lock-in.
- The terminal makes no cloud calls, passing context to a local agent subprocess so data handling is governed by the chosen agent, a direct answer to the Recall backlash.
- Over 25 agents already support the protocol, and the terminal auto-detects whichever compatible agents are installed on the machine.
- Windows-only at 0.1, requiring Windows 11 build 22621.6060 or later, with WSL2 support and local inference still on the roadmap.
Questions Worth Asking
- Given that Microsoft archived AI Shell in January 2026 and deprecated Terminal Chat, what evidence would convince you that Intelligent Terminal will still be maintained in eighteen months?
- If the terminal stays vendor-neutral today but defaults to the Copilot CLI that bills back to Microsoft, how long does open-protocol neutrality survive contact with usage-based billing economics?
- Would an AI agent pane actually change how you work in the terminal, or is terminal choice so wired into habit that no feature short of a workflow you cannot live without will move you?
