The Scenario
Traditionally, you open Cursor, have Claude Code write a feature, wait for it to finish, review the result, and iterate. One agent, one path, serial waiting.
But what if you could run 5 agents at once—each in an independent git worktree, implementing the same feature with different strategies—and just merge the best one?
That's what Orca does. Launched in March 2026, it reached 42,000 stars by August, with YC backing. It's not another coding agent—it's an orchestration environment for multiple coding agents, officially called an ADE (Agent Development Environment).
From IDE to ADE
The evolution of dev tools: editors (Vim, Emacs) manage text; IDEs (VS Code, IntelliJ) manage projects; AI IDEs (Cursor, Windsurf) manage single AI interactions. Orca's ADE adds one more layer: managing agent fleets.
In an ADE, the basic unit is no longer "one prompt" or "one file" but "one complete work session of an agent in one worktree." You can run Codex, Claude Code, OpenCode, and Pi simultaneously, each in its own isolated worktree.
This is fundamentally different from opening multiple VS Code windows, where you still switch attention serially. Orca provides unified tracking: all agents' states, diffs, and terminal outputs live in one interface for side-by-side comparison.
Core Mechanism: Worktree Isolation + Mobile Monitoring
Orca's key design is Parallel Worktrees. Git worktrees let multiple working directories of the same repo coexist, each checked out to a different branch or commit. Orca applies this to agent orchestration:
1. You give one prompt 2. Orca fans it out to N agents 3. Each agent works independently in its own worktree 4. You compare results and merge the winner
The elegance: isolation is free. Worktrees share the .git directory and only create working files—5 agents on one repo barely increases disk usage.
The second killer feature is the Mobile Companion. When your agent finishes, your phone gets a notification; if an agent gets stuck, you can send a follow-up from your phone on the subway. Agents become background processes you can manage asynchronously.
Comparison with Existing Tools
| Dimension | Cursor / Windsurf | Claude Code (CLI) | Orca | |------|-------------------|-------------------|------| | Agent count | 1 | 1 | N (parallel) | | Isolation | None | None | Git worktree | | Mobile | None | None | iOS/Android | | Agent choice | Locked to vendor API | Locked to Anthropic | Any agent, own subscription | | Remote | None | None | SSH worktree |
That last row matters. SSH Worktrees let you run agents on beefy remote machines while editing and reviewing locally—important for GPU- or memory-intensive tasks like test suites and data processing.
Orca is vendor-agnostic: use your own Codex and Claude subscriptions; Orca only orchestrates. In 2026's agent ecosystem, avoiding vendor lock-in is a significant choice.
Why Now?
By mid-2026, the coding agent market completed its first shakeout: Claude Code, Codex, OpenCode, and Cursor Agent each found their footing. Single-agent capability is strong enough—but has a hard limit: it's serial. One agent finishes a task before the next begins. If you have 5 independent bugs or 3 implementation approaches to compare, serial waiting wastes time.
The value of parallel orchestration isn't making agents smarter—it's letting multiple sufficiently smart agents work simultaneously. It's like the jump from single-core to multi-core CPUs: once per-core performance hit a wall, multicore became the only way forward.
Open Questions
1. Merge conflicts: If two agents modify different parts of the same file, merging still requires manual intervention despite Orca's "merge the winner" promise. 2. Attention allocation: Can you really review 5 diffs at once? Mobile notifications solve "knowing it's done" but not "understanding what it did." 3. Cost: 5 agents running 10 minutes in parallel equals 50 minutes of API consumption—significant for token-billed users. 4. Vendor dependence: Orca depends on the stability of the agent ecosystem; if Anthropic changes Claude Code's interface, Orca must follow.
The Bigger Picture
Orca represents a broader trend: dev tools are shifting from "assisting humans to write code" to "orchestrating agents that write code."
The endgame poses a question: when agents can independently complete most coding tasks, what is the human's role? Orca's answer: the human becomes a commander—deciding which agents do what, comparing results, merging winners. The mobile companion reinforces this metaphor.
This differs sharply from the traditional image of the software engineer as a "craftsman at the keyboard." In the ADE era, the engineer is more like a project manager. That's not a bad thing—it means leverage grows: your output becomes "how many agents can you orchestrate" rather than "how many lines can you write." Orchestration skill is becoming a new core competency.
Orca is the first tool to productize this orchestration capability. It won't be the last.
---
Project: https://github.com/stablyai/orca Website: https://onorca.dev License: MIT Platforms: macOS / Windows / Linux / iOS / Android