Clark

About Clark

Clark is an AI coworker that operates on its own cloud computer, complete with a browser, terminal, files, and code. You assign a task, close the tab, and return later to finished work - such as sourced research, spreadsheets, decks, audits, or tested code. The tool can run on a schedule, fan work out to parallel specialists, and deliver artifacts with the underlying evidence.

Review

Clark shifts the AI interaction model from a synchronous chat window to an asynchronous workspace. Instead of prompting and waiting, you hand off a real task and get back a concrete result. The cloud computer framing means the agent has access to a browser and file system, which lets it handle multi-step jobs that require external research or code execution.

Key Features

• Cloud computer workspace: Each task runs on an isolated cloud computer with a browser, terminal, files, and code. The agent can navigate the web, manipulate files, and execute commands.
• Parallel specialist fan-out: Clark splits complex work across multiple isolated instances that run concurrently, then merges their outputs into a single artifact automatically.
• Scheduled runs and monitoring: Tasks can be set to repeat on a schedule, so Clark can perform periodic checks or updates without manual intervention.
• Evidence artifacts: Completed jobs return files, screenshots, sources, logs, or URLs. The chat interface shows a full proof of work for inspection.
• Clark Code: For software development, Clark works directly inside real repositories, making changes and handing back diffs or pull requests - similar to tools like Codex or Claude Code.

Pricing and Value

Clark offers a free tier with 2000 credits. Details about paid plans, credit costs after the free tier, or subscription pricing have not been defined publicly at this stage.

Pros

• Tasks run asynchronously; you can close the tab and come back to a finished artifact.
• Returned work includes inspectable evidence - files, screenshots, and logs - so you can verify the output.
• The parallel specialist model lets Clark tackle broad research or multi-faceted projects more efficiently than a single-threaded agent.
• Clark Code integrates with existing repositories, which means you can keep using your own version control and review changes before merging.
• An OpenAI-compatible API allows embedding the agent into other tools or workflows.

Cons

• Clark is not well suited for users who need real-time, interactive assistance or step-by-step guidance, as tasks are handed off and completed asynchronously.
• When parallel specialists merge their outputs, the reconciliation process may surface conflicts or inconsistencies that require manual review - the exact handling of those edge cases isn't fully documented yet.
• State persistence across scheduled runs and between separate tasks in the cloud computer is not clearly specified; it's uncertain whether the workspace is ephemeral or durable.

Clark fits use cases where you want to delegate a well-scoped task and inspect the result later - think deep research, document drafts, prototype code, or recurring monitoring. The evidence trail and repository integration make it a practical option for developers and analysts who need to audit the work. If you prefer a live, back-and-forth interaction or need tight control over every intermediate step, this async model may feel limiting.