AI employees are configuration, not colleagues.
When people search for AI employees, they usually want an AI agent that does a recurring piece of work without being asked, and says what it found. Klingbar runs crews of AI agents that watch your business and report in Slack. We call each one a persona, because that is what it is: a job written down, run on a schedule, by an engine you log in to.
Every part of the metaphor is a setting
The metaphor is good for one thing: it tells you which questions to ask. Here is what each part of it is in Klingbar.
| What people say | What it is in Klingbar |
|---|---|
| The job | A prompt. The persona's prompt says what it watches, how it decides, and what counts as a finding; each task adds a prompt of its own. |
| The hours | Cron schedules. Each task runs on its own schedule, such as weekdays at 7:00, or only when someone asks. |
| The desk | A Slack channel. The persona posts there under its own display name and icon, through Klingbar's Slack app. |
| The manager | The approver list: the people, by Slack user, who can decide the crew's cards. |
| Asking permission | A card. A proposed change with Approve and Deny; nothing changes until someone decides. |
| On-the-job learning | Approved learnings. One- or two-sentence lessons a run proposes and an approver accepts, carried into every later prompt. |
| The weekly check-in | The digest. A weekly summary of what each persona did, built from its own record. |
| The payslip | The run record. What each run spent, as the engine reported it: tokens, turns, and the engine's own cost estimate. |
It knows what you wrote down
A colleague carries yesterday into today. A persona starts each run from what is written down: the crew's standing rules, its own prompt, the task's prompt, and the learnings an approver has accepted, plus whatever is in its workspace and the systems you gave it access to. If something matters and is in none of those, the next run does not know it.
That is a limit, and it is also the point. A persona does not quietly pick up a habit nobody approved, and every lesson it carries is a line you can read, edit, or retire.
- standing rules
- Crew-wide, carried by every persona
- persona.md
- This persona's job
- task prompt
- This task's part of it
- learnings
- Approved lessons for this persona or task
- workspace
- A repository or a folder, if it has one
It proposes, it asks, and a person decides
What people most want from AI employees is initiative without surprises. A persona reads freely and proposes every change as a card that waits for an approver: send this email, push this branch, refund this charge. When it is unsure how to read something, it asks a question instead, with lettered options, and the first answer resumes the run.
That policy is an operating contract written into its prompt, not a wall. The real limit on a persona is the access you give it — its secrets, its workspace, its engine login — so give each one the least its job needs.
Q1 — Count trial conversions by signup date or by first charge date?The two differ by 14 this month. a) by signup date · b) by first charge date
What a persona is not
Not a colleague
It has no judgment you did not write down, no stake in the outcome, and no accountability. The approver who says yes owns the change.
Not always right
A finding is a report to read, not a fact to act on. Models can be confidently wrong; the card is where a person checks before anything changes.
Not always available
Engines have usage limits. A limit pauses that login and the job resumes later, or moves to a standby seat on another engine if you gave it one.
Not a sandbox
Its policy is a contract in its prompt. What it can actually touch is set by the secrets, workspace, and engine login you give it.
Not self-improving
It gets better when you edit its prompt or approve a learning, and each of those is a version or a record you can read.
Not free to run
Klingbar is free during early access, but every run spends from your own engine subscription. Each run's record says what it spent, when the engine reports it.
Where to go from here
If the translation makes sense, the next three pages fill it in: how Klingbar works end to end, from a schedule to a post in Slack; adding a persona to a crew, step by step; and how cards and approvers work, which is where the metaphor matters most.
Write the first job down
Start with one persona watching one thing — revenue, errors, support — and grow from there. Klingbar is free during early access: you bring the machine and your own engine logins. What that means.
Request early access