The developer's nightmare used to look something like this: Claude Code running in one terminal, OpenAI Codex in another, git branches tangling like holiday lights, and the nagging worry that two AI agents might try to edit the same file at the exact same moment. Not ideal, to put it mildly.
A small San Francisco team thinks they've found a cleaner path forward. 21st—a three-person shop that went through Y Combinator's Winter 2026 batch—launched 1Code with a straightforward pitch: what if you could run multiple AI coding agents side by side without the chaos?
The premise is deceptively simple. Founders Serafim Korablev and Sergey Bunas built 1Code as open-source infrastructure that treats AI coding agents like orchestra members rather than solo performers. Each agent gets its own workspace, its own git worktree to avoid conflicts, and the whole operation runs from a single control panel. Whether it catches on may say as much about where developer tools are headed as it does about 21st's execution.
Orchestration as the New Table Stakes
Strip away the marketing gloss and 1Code is essentially a traffic controller. The platform lets developers run Claude Code and Codex—or other models, as support expands—without the agents stepping on each other's toes. Each session runs isolated in git worktrees. Background tasks spin up in sandboxed cloud environments powered by E2B, a containerization service.
The interface prioritizes speed. Developers can switch agents mid-task, preview changes across desktop and mobile viewports in real-time, and stage commits visually rather than through command-line gymnastics. It's keyboard-first by design, which will either delight power users or leave everyone else hunting for mouse-friendly alternatives.
Where 1Code gets interesting—or perhaps overly ambitious—is its integration layer. The platform plugs into the Model Context Protocol, allowing teams to summon agents directly from Notion, Linear, GitHub, Slack, Sentry, or PostgreSQL. Type "@1code" in a Linear ticket, and theoretically the system picks it up, writes code, runs tests, and opens a pull request. The company's higher-tier plans automate code review cycles, letting agents push fix commits without manual intervention and respond to CI/CD failures autonomously.
Whether that level of automation feels liberating or terrifying probably depends on how much you trust your agents—and your test coverage.
Beyond the Desktop Client
21st didn't stop at building a desktop app. On March 16, the company released an Agents SDK aimed at developers who want to embed these capabilities into their own products. The SDK bundles sandboxed execution environments, credential proxies (so API keys never touch the sandbox), server-sent event streaming, and observability tools.
According to the company's launch materials, the SDK works with TypeScript, Next.js, Node, and Python. Developers define agents in code, deploy via CLI, and manage instances through an API. The persistent E2B environments keep filesystem and git state intact across sessions, which matters if you're building something that needs to remember context between tasks.
It's a broader platform play than the initial 1Code desktop app suggested. But then again, startups rarely raise money by solving just one problem.
Pricing: Free to Expensive
The business model breaks into three tiers, each telegraphing a different customer profile. The open-source version runs free on macOS, Windows, and Linux under an Apache 2.0 license. Teams can self-host, modify, and deploy it internally—a nod to the open-source ethos that still carries weight in developer circles.
Pro costs $20 monthly and adds a web app, PWA for mobile use, background agents, and live browser previews. The idea is you start a task on your phone during a commute, then pick it up on your desktop later. Whether developers actually work that way remains to be seen.
Max, priced at $100 per month, targets teams and power users. It includes automated PR reviews, Linear task orchestration, CI/CD fixes, and direct support. That's where the automation features live—the ones that either save hours or introduce new failure modes, depending on your perspective.
Early Traction, Quiet Launch
1Code first appeared on Hacker News in mid-January, drawing the usual mix of enthusiasm and skepticism that greets Show HN posts. By February, developer newsletters and tech press in France and Denmark had picked it up, mostly highlighting the visual diff workflow and version control approach.
The company's most recent changelog, dated early February, graduated its Kanban and Tasks features out of beta and added support for Claude Code's Opus 4.6. It's the kind of incremental progress that matters more to daily users than casual observers.
As for traction? The team lists three people on its Y Combinator profile. Funding hasn't been disclosed beyond the standard YC check. The product site mentions usage by development teams but conspicuously lacks logos or adoption numbers—a red flag for some investors, a sign of early days for others.
The Crowded Orchestration Space
21st isn't alone in betting on multi-agent coordination. Anthropic rolled out a multi-agent code review feature for Claude Code around the same time 1Code launched. OpenAI released its Codex app for Mac in early February, positioning it as a command center for managing agent teams across the full development lifecycle.
Competing tools have proliferated. Conductor offers a Mac-native app with parallel agent workflows and local-first architecture. Agent Swarm went the open-source route with MCP-powered orchestration and self-hosting options. Google's Gemini CLI spawned community extensions like Maestro for multi-agent setups. Meanwhile, startups like Orchestre focus specifically on MCP-based coordination for production environments.
The pattern is unmistakable: as AI coding agents mature, developers need infrastructure to wrangle them. Running one agent is manageable. Running three or four simultaneously without stepping on your own toes? That requires tooling.
1Code's angle is treating that coordination layer as open-source infrastructure with optional managed services, rather than locking teams into a proprietary platform. It's a classic developer-first strategy—give away the core, charge for convenience and scale.
The Execution Question
Whether 1Code becomes the default way developers juggle multiple agents hinges on questions the company is still answering. How quickly can it add support for models beyond Claude and Codex? Does its Git-first workflow resonate with teams already drowning in CI/CD complexity, or does it add one more thing to manage?
Perhaps more fundamentally: do developers actually want to run multiple AI agents in parallel, or is this a solution to a problem that sounds worse than it is in practice?
The answers will likely emerge slowly. Developer tools adoption follows a peculiar path—Hacker News buzz, word-of-mouth in Slack channels, gradual infiltration into team workflows. Success often looks less like explosive growth and more like quiet, persistent usage that compounds over months.
For now, 21st has built something technically coherent and priced it in a way that lets teams experiment cheaply. The open-source version removes the biggest barrier to trial. The rest depends on whether Korablev and Bunas can turn early curiosity into habitual use—the kind where developers forget they ever worked another way.
That's always the hardest part.
