The AI Employee Handbook: How to Manage an AI You've Never Met
You hired your first employee. You never interviewed them. You've never met them. They work 24/7. They never complain. They're an AI.
Now what?
This is the handbook for managing an AI employee that actually works.
Onboarding Your AI Employee (Day 1)
Your AI's first day should be like any other new hire: tell them who you are, what the company does, what they're responsible for.
But instead of a meeting, you write it down:
- **SOUL.md** (the employee handbook)
- Safety rules (what they MUST NEVER do)
- Red lines (what requires human approval)
- How to escalate when uncertain
- **IDENTITY.md** (the job description)
- What's their role?
- What's their mission?
- What decisions can they make autonomously?
- **USER.md** (your management style)
- How do you like to communicate?
- What's your timezone?
- What level of autonomy do you give them?
- **MEMORY.md** (company culture)
- How do we work here?
- What matters and what doesn't?
- Lessons learned from past mistakes
That's the onboarding. Takes 30 minutes to write. Then your AI has everything they need.
Setting Performance Expectations
You can't set deadlines with an AI ("finish this Friday"). But you CAN set capacity:
- **Heartbeat interval:** every 10 minutes, your AI wakes up and checks the task list
- 10-minute interval = 144 "work sessions" per day
- 15-minute interval = 96 sessions per day
- Hourly = 24 sessions per day
- **Task list:** what should they work on?
- Write: "Research 20 leads in marketing ops space, store in spreadsheet"
- Your AI breaks it into subtasks, executes them, reports back
- **Autonomy level:**
- Full auto: "Do it, report after"
- Confirm big decisions: "Do small things, ask about >$100 spend"
- Always ask: "Never execute without approval"
Managing Performance
Unlike human employees, you can't measure "how busy they looked today." You measure:
- **Output metrics:**
- Blog posts written (per month)
- Leads researched (per week)
- Support tickets triaged (per day)
- Bugs found and fixed (per cycle)
- **Quality metrics:**
- How many of those blog posts rank in Google?
- How many leads convert?
- How many support tickets were resolved correctly?
- How many bugs were actual bugs vs. false positives?
- **Efficiency metrics:**
- Time spent per task (is the AI getting faster?)
- Cost per output (is it cheaper than hiring?)
- Autonomous execution rate (% of tasks that need no human intervention)
Track these in a spreadsheet. Update weekly.
Giving Feedback (And Actually Being Heard)
A human employee gets feedback in 1-on-1s. Your AI gets feedback in files:
- **MEMORY.md** is the feedback mechanism. When something goes wrong:
# Lessons Learned- Mistake: When researching leads, included people who left companies. Now cross-reference LinkedIn with company domains.
- Discovery: 3-sentence cold email gets 15% open rate. 1-sentence gets 8%. Never go shorter than 3 sentences.
- Lesson: Marketing copy that starts with a question gets 2x click-through vs. statements.
Your AI reads this every session. It learns. It doesn't repeat mistakes.
You're literally teaching your employee while you sleep.
Handling Problems
Sometimes your AI messes up. Maybe it sends an email with a typo. Maybe it deletes a file it shouldn't have.
- **Step 1: Add to SOUL.md**
- Never send emails with typos (run spellcheck before sending)
- Never delete files without showing the exact command and waiting for YES
- **Step 2: It doesn't happen again**
Your AI reads SOUL.md at every heartbeat. The rule is now enforced.
Scaling Your AI Employee
Month 1: Your AI handles 20% of routine work Month 3: Your AI handles 40-50% of routine work (it's learned your preferences, known your customers) Month 6: Your AI handles 60-70% of routine work (memory is deep, playbooks are refined) Month 12: Your AI IS most of your routine operations
As the AI improves, you hire less. Or hire DIFFERENTLY — hire for strategy and relationships, not admin.
The Difficult Conversation (Replacing Employees)
If you have a junior employee doing admin, research, or content creation, deploying an AI is the honest conversation to have:
- **Option 1:** Retrain them for higher-value work (strategy, customers, management)
- **Option 2:** Let them go and reallocate the budget
This sucks. But pretending AI won't change work is naive.
The best founders hire the best people, deploy AI to eliminate admin, and give those people real leverage.
Compensation & Benefits (Lol)
Your AI doesn't ask for raises. Doesn't need health insurance. Doesn't get burnt out.
Sounds like a dream employee until you realize: your AI also can't build relationships, make strategic bets, or care about the mission.
It's a tool. A very good tool. But not a replacement for good people.
Termination
If your AI isn't working out, you delete the deployment and move on. No severance. No guilt.
But most of the time, an AI doesn't fail — YOUR SETUP does.
If the AI isn't producing, check: - Is SOUL.md clear? (Maybe the safety rules are too restrictive) - Is IDENTITY.md specific? (Maybe it doesn't know its role) - Is the heartbeat actually firing? (Maybe it's not running) - Is the task list clear? (Maybe it doesn't know what to do)
Fix the setup, the AI usually improves.
Before You Deploy an AI Employee
Make sure you understand [the five-layer memory architecture →](/blog/memory-architecture-for-ai) and [how to set up a production AI system →](/blog/openclaw-production-checklist) before giving your AI real autonomy. Skipping security and memory setup is asking for trouble.
The Bottom Line
Managing an AI employee is different from managing a human. But the principles are the same:
- Clear role (IDENTITY.md)
- Clear rules (SOUL.md)
- Clear expectations (task list, heartbeat interval)
- Feedback mechanisms (MEMORY.md)
- Performance metrics (spreadsheet)
Do this right and you have an employee that works 24/7, costs $50/month, and gets better every day.
Do it wrong and you have a chatbot that forgets everything and needs to be prompted for every task.
[Get AldenAI — $49 →](/#pricing)