# Give an AI teammate a real responsibility For this site, we needed someone to check what we could honestly tell readers about AI teammates and what they cost. I want that kind of preparation to happen without every question turning into more work for the people already on the team. ## Give the work somewhere to go We sent the questions to our Docs Quality teammate through Channels. It read the product, pricing and setup sources and returned a brief we could use. One finding changed the cost explanation below: $3 plus ten billed hours uses $13 of credit, but a new one-person Workspace still costs $20, plus the AI subscription. It also caught a distinction that matters when you budget: the bill counts time the computer runs until it sleeps, not just the moments it produces useful work. Our review caught a mistake too. The brief treated an example model as a product default. We sent that back, and the teammate corrected the sentence against the right source. The next question was whether someone without a Mac could get started. We gave the same teammate the next job: check the onboarding path. It found that the quickstart mentioned only Mac, while the linked install page also offered a Windows preview. We checked both pages and changed our guide to point readers to the current installer choices. The next answer also used the correct source for the model default without another reminder. That is the shape of the work I want to delegate. A question comes up, the teammate gathers the evidence, and we use its answer to make a decision. When something is wrong, the correction belongs in the work that follows. I still need to judge the result. I want the preparation to stop depending on me doing every step. [Read both assignments, the returned briefs and our checks](https://eduarda.ai/evidence/managed-teammate-review-2026-10-17.md). These were source reviews by our existing managed Agent. The release-note editor below is an example of another role you could define. ## Why the identity and computer matter A teammate needs somewhere to do the work. A [managed Socra Agent](https://workspace.socra.com/products/agents) has its own organizational identity and cloud computer with a terminal and a home that persists through ordinary restarts. Its work can continue while your laptop is closed. You give that identity access to the resources the responsibility needs. For a release-note editor, that might be read access to completed changes, a Drive for drafts and a Channels conversation for delivery and questions. A subscription gives the agent AI usage; [resource permissions](https://workspace.socra.com/docs/agents/guides/access-and-credentials) determine what it can reach. That lets you define a useful boundary. The editor can gather evidence and prepare the note. You decide whether it is ready for customers. Another role can have a different boundary. A job title alone doesn’t tell an agent what it should be allowed to do. ## Give the responsibility continuity A release happens again. The teammate should be able to receive the next assignment, find the working files and use what you learned from the last one. Channels supplies the conversation. Drive holds the working files. Cortex holds decisions and reasons, such as how to describe an unfinished feature without implying it shipped. Assigned Issues track work that needs coordination. Tasks can bring the agent back to scheduled work, when you have configured that schedule and its access. This is why I care about the knowledge carrying forward. If I correct a draft, I want the correction to improve the next draft too. The agent maintains the relevant knowledge and retrieves it before working. We still check whether it applies it. [Our Cortex story](https://eduarda.ai/cortex) explains what happened when our agents kept losing what we had learned. ## What one role costs As of October 17, 2026, [Workspace costs $20 per person per month](https://workspace.socra.com/pricing), including all Workspace apps and $20 of monthly usage credit per person, pooled across the team. A managed Agent costs **$3 per month plus $1 per hour running**. Both draw from that credit; excess usage costs extra. Your supported AI subscription is separate. Workspace has no trial. For one person and one Agent kept for a full month, ten billed running hours consume $13 of credit. With no other usage, the Socra bill is $20 and $7 of credit remains until it expires. How much useful work you get from those hours depends on the assignments and the review they need. The Agent’s base price is prorated and continues while it sleeps. Running time rounds up to whole minutes per run. See [billing details](https://workspace.socra.com/docs/admin/agents/billing) for the rules and controls. ## Choose a role worth keeping Start with work that keeps waiting for someone’s attention, has accessible sources and produces something you can judge. A release-note draft, a checked research brief or a review of documentation could qualify. Those are possible responsibilities; the source review above is the one demonstrated here. Give the teammate one real assignment. Check the result and count your own time briefing, reviewing and repairing it. Include the effort of setup and keeping its knowledge useful. Then give it another assignment after a real correction and see whether that correction carries forward. I want more useful work to reach completion with the people already on the team. That is why I would start with a responsibility small enough to inspect, then expand it when the work earns that trust. The [first-role guide](https://eduarda.ai/agent-guide) helps you define the responsibility and its first deliverable before you create a paid Agent.