An AI teammate’s first job should produce something you need and can check. Choose a bounded assignment with accessible source material, a clear result and a person who can judge it. Review that first result before making the work recurring or granting broader access.
This guide is for a founder or small team deciding where to begin. The examples are proposed assignments, not customer results or promises of autonomous completion.
Which Job Should Come First?
Write down the work you want to take on. For each candidate, answer four questions:
| Question | A promising first assignment | Narrow the job when… |
|---|---|---|
| Can you describe the finished result? | A release brief with changed dates and source links. | The request is “run marketing” or “grow revenue.” |
| Can the teammate reach the facts? | You can share the release plan and approved product notes. | Success depends on facts nobody has recorded. |
| Can you check the result? | You know which dates are authoritative and can review the draft. | You cannot tell whether an answer is correct. |
| Can you limit the consequences? | The first output stays in a shared draft for review. | The first step sends messages, spends money or changes production. |
Choose the smallest useful version of a promising job. A research role might return a sourced comparison of five options. A launch role might reconcile a release brief with the current plan. A quality role might reproduce one reported issue and record the steps. Each gives you an artifact to inspect before you expand the role.
The title alone does not define the job. “Launch lead” becomes actionable when you specify which brief to maintain, which sources to trust, what may change and who approves the customer announcement.
Write a Brief the Result Can Be Judged Against
Here is an example first assignment:
Compare the launch brief with the current release plan. Update changed dates in the draft and link each change to its source. Flag disagreements and missing dates. Post the draft and a short change list in the launch channel. I will review it before anything goes to customers. Do not send email or change the release plan.
This brief names the output, evidence, permitted changes and review boundary. It also gives the teammate a useful response when the facts do not support a confident answer.
Download the editable assignment brief. It is plain text: paste it into a conversation or keep it with the team’s working files. Complete the acceptance criteria before assigning the work.
Give the Role Its Own Access
In Socra, an AI teammate has a separate identity and cloud computer. Share the channels, files, knowledge and repositories its job calls for. Keep unrelated information outside that scope. The identity and access guide explains the product’s boundaries.
A written instruction describes intent. Permissions enforce what an identity can reach. Use both. Asking a teammate to avoid a folder is different from withholding access to that folder.
Before the first assignment, check the identity shown on the work, the resources it can read, the resources it can change and the person who can revoke that access. Follow the supported access and credentials setup; connecting an external tool still depends on that tool’s supported authentication and permissions.
For the launch example, start with the release plan as a source, the launch draft as an editable file and the launch channel for the handoff. Customer email sending does not need to be part of that first job.
Preserve the Reasons Behind the Work
Give the teammate the decisions that would change its answer. “Use the public release date only after approval” is more useful than a large folder of unprioritized notes.
Cortex keeps shared knowledge and its reasons in Modules, each covering a subject. Dependencies connect prerequisite knowledge. A Flash retrieves a Module with those prerequisites before work starts. The teammate still needs to check the current source material and system; stored guidance is not proof that a date, feature or implementation is current.
For personal working notes and files, see context and memory. Put a decision the whole team should follow in shared knowledge rather than relying on one conversation’s recall.
Review the First Result
Use the acceptance criteria you wrote before the run. Keep a record that distinguishes a plausible answer from work you accepted:
| Record | What to capture |
|---|---|
| Result | Link to the actual draft, report, change or other artifact. |
| Factual accuracy | Which claims you checked against which sources; every correction required. |
| Scope | Whether it made only the changes you allowed. |
| Missing information | Whether it flagged uncertainty instead of filling gaps. |
| Human effort | Time spent briefing, reviewing and correcting. |
| Usage | Recorded running time and charges for the work. |
| Decision | Accept, revise or stop, with the reason. |
Do not label a run successful merely because it finished. For the launch brief, open the cited plan, check every changed date and confirm that no customer message was sent. If the brief needs correction, record whether the cause was missing context, access, an ambiguous instruction or execution. Repair that cause before repeating the same job.
One accepted example demonstrates that example. Try another representative assignment, including a case with missing or conflicting information, before treating the role as dependable for a wider class of work.
What Does a First Teammate Cost?
Socra HQ costs $20 per human seat per month, including $20 per seat in shared usage credit. Each managed AI teammate costs $3 per month plus $1 per running hour. Both teammate charges use the same included credit. A supported ChatGPT or Claude subscription is required separately.
These full-month examples use total billable running hours across all listed teammates. They assume no other HQ usage, no taxes or provider charges, and additional running usage enabled wherever needed:
| Human seats | AI teammates | Total running hours | Teammate charges | Included credit | Socra HQ total |
|---|---|---|---|---|---|
| 1 | 1 | 10 | $13 | $20 | $20 |
| 1 | 1 | 17 | $20 | $20 | $20 |
| 1 | 2 | 30 | $36 | $20 | $36 |
Running time is measured from wake to sleep. Each session rounds up to a whole minute, with a one-minute minimum. A computer that is awake can incur running time while waiting; this is not a price per accepted result.
Additional running usage is off by default. Running pauses when available credit is exhausted unless an Account owner enables additional usage. The monthly teammate charge continues while the teammate exists and can exceed included credit. Unused credit does not roll over. Use the pricing calculator for a different team and read the billing details there before deciding.
When Should the Job Become Recurring?
After you accept the first results, define when the work should recur, which changes deserve a notification, and what requires review. Tasks provides scheduled work; the assignment still needs a clear owner and stopping conditions.
For the launch role: compare dates every Monday, update the draft when a confirmed date changes, and notify the human reviewer with the source. Keep customer announcements subject to review. Recheck the assignment when the release plan, access or role changes.
Add another role when there is another defined job and a clear handoff. A research teammate can return source material to a writing teammate, with a person judging the final draft. More teammates also mean more coordination and usage; measure whether the handoff improves the result.
Start With Work You Can Inspect
See how Socra Agents work, check subscription requirements, and use the setup guide when you are ready. Keep the completed brief and the accepted result together so the next assignment starts from what you learned.