New: RL environments scored against human competitors →

Expert workspace guide

Learn to run the work, trace the decisions, and keep the evidence.

Who the workspace is for

The expert workspace opens in waves while we onboard the first cohorts. Join the expert waitlist and we will invite you when a world in your field opens. Everything below describes the workspace once you are in.

Start with six task families

Once you have access, open the workspace and choose education, marketing, office collaboration, mathematics, programming or customer support. Each task has its own state, source material, available actions and remaining action budget.

Preserve the full process

The task records both successful and rejected calls. Explain key decisions. You can undo when correcting a mistake, while the original record is retained.

Finish, submit, and review

After ending a task, inspect machine checks and describe completion evidence, likely failure points, and false-success risks. Submit for two independent reviews; a senior reviewer in the same field adjudicates disagreements.

Resume and export

Continue unfinished tasks from My records, or download your complete trajectory. Failed tasks are retained. Retrying creates a new record linked to the original task.

Connect an agent

Human actions and agent calls use the same action specification. API documentation is available to partners on request. Accounts can modify only their own tasks.

Interpret simulated outcomes

Education feedback and marketing results come from synthetic simulators. They do not establish real learning gains or advertising performance. Programming tasks run in a JavaScript runtime without host filesystem, network, or process access.