Writing
Notes on AI employees
64 pieces for operators on one-process builds and the parts that still need a person. RSS
· Foundations
AI employee FAQ
The questions people actually type, answered the way I answer them on a thirty-minute call.
· Foundations
What an AI employee actually is
An AI employee is a named worker for one written process, not a chat box with a job title.
· Build
A 7-day AI employee sprint
One process, from the first call to something running, in seven days. Pick, map, build, hand over.
· Use cases
AI employee for collections
Collections is a reminder with a gate. The employee can nudge and record a promise. It cannot threaten, settle, or keep going after someone asked to stop. Money is a hard gate. Measure recovered cash and complaints, together.
· People
Slack, email, and the AI employee
If the pile is already in Slack or the inbox, that is the workplace. A new tab is how the employee dies in procurement.
· Foundations
AI employee vs chatbot
A chatbot answers. An AI employee finishes a job and leaves a record you can audit.
· Money
Pricing a process, not a seat
A seat price asks how many logins. A process price asks what work leaves the building.
· Foundations
What changed about AI agents in 2026
More agents sat in real workflows in 2026. Governance did not catch up. Owners still matter.
· Build
How to pick the first process
The first AI employee should take a pile you can name in hours, not a strategy project you cannot finish in a quarter.
· People
From pilot to production AI employees
A pilot proves a demo. Production proves a Monday. Those are different projects, and only one needs an owner.
· Industry
AI employees for sales CRM teams
Look up the lead, write the first note, flag the ones worth a call. Do it inside the CRM you already have.
· Foundations
What an AI employee cannot do
It cannot invent policy, close harm alone, or sit outside the law. Drafting is not deciding.
· Build
Map the workflow before you build
If you cannot walk a stranger through the steps, you are not ready to connect a model to your tools.
· Money
CFO questions before you buy
Finance should ask about the process, the miss, the run cost, and the stop rule. The rest is decoration.
· Use cases
AI employee for onboarding
Onboarding is a checklist with a finish line. The employee chases tasks, files the docs, and flags the stuck ones. A person still grants access and signs the money. Measure people who actually start.
· Foundations
When not to hire an AI employee
If the process is folklore, the stakes have no gate, or nobody owns misses, do not hire yet.
· Build
Connect to tools you already use
The point is to take work off the team, not add another tab they will forget to check.
· Money
When $5,000 is enough
$5,000 is enough when you want one live process, not a transformation story.
· Industry
Selling a company that runs on agents
Buyers open cost, uptime, and whether you can explain the system. Agents without owners look like magic.
· People
Multilingual AI employees
If you cannot review it in the language you send, you do not have a multilingual employee. You have a lottery.
· Build
Escalation that does not lose context
People should receive a case they can decide, not a chat log they have to reconstruct.
· Money
Hidden costs of AI employees
The build is visible. Tokens, reviewers, monitoring, and broken connectors are the costs that arrive later.
· Use cases
First-pass CV screening that stays honest
A first pass pulls facts against a written role. It does not hire. It does not invent a skill. Keep the criteria public to the team, the person on the gate, and a log you can replay.
· People
When the model is wrong
Wrong output is not a vibe. It is a state on the work: escalated, owned, and still attached to the facts.
· Foundations
One process, not a platform
Hire one written process. A platform is a year. A queue is a week.
· Industry
What banks will ask about your AI employee
Banks ask cost, uptime, logs, and whether they can plug in before lunch. Not which model you like.
· Build
Testing an AI employee before go-live
Go-live is not a feeling. It is a test set, a miss rate you can explain, and a person who still owns the weird cases.
· Money
Starter vs The week vs several workflows
Three prices. Three scopes. The mistake is stuffing three processes into the cheapest line.
· Use cases
After-hours support without a night shift
Nights need a queue, not a theatre. Close the repeats. Book or take a message. Wake a person when the protocol says so. The morning team should inherit state, not a pile of chats.
· Foundations
Human in the loop is the job
The person on the gate is not a fallback. They are the job you actually hired.
· Build
Handover docs your team will actually read
If the write-up needs a second notebook, it will not be open at 2am. Day 7 is a page and a call, not a binder.
· Money
Hours saved are not the whole story
A shorter timesheet is not proof the process got better. It is one line on a longer sheet.
· Industry
AI employees for recruiting agencies
A recruiting AI can shortlist. It cannot hide you from LL144, the EU AI Act, or the EEOC.
· People
Training your team to work with an AI employee
The team does not need a lecture on models. They need a week of reviewing real cases without rubber-stamping the output.
· Use cases
AI employee for appointment booking
Booking is offering a slot that exists, writing it down, and telling the truth when it does not. Escalate clinical and money questions. Measure no-shows and ghost bookings, not chat volume.
· Build
Monitoring an AI employee in production
Pretty charts do not replace a person who can explain why a case failed. Watch the paths a stranger will shout about.
· Money
How to measure ROI on an AI employee
A live workflow is capability. Hours actually leaving a desk is realized. Strategy is a later conversation.
· Industry
AI employees in compliance
Compliance AI should cut the pile, not the person. The miss still needs a name next to it.
· Foundations
AI employee vs copilot
A copilot sits next to a person. An AI employee owns the queue when the person is not there.
· People
Who owns the AI employee
If nobody can stop it, take the exceptions, and explain a miss, you do not have an AI employee. You have a demo with a login.
· Use cases
AI employee for document review
Document review is extract, flag gaps, and leave the weird cases. That loop cut review work by about half at SmartComply. The model was not the hero. The person on the miss was.
· Build
What breaks in week two
Launch week hides the cases you did not sample. Week two puts them on the table. That is normal if you are watching.
· Money
How much does an AI employee cost?
The build fee is the easy number. The real cost is what you keep paying after the handover call.
· Industry
AI employees in lending
Lending AI should chase the incomplete file. The yes still belongs to a person you can name.
· Use cases
AI employee for internal operations
Internal ops is the copy-paste between two tools. Automate that first. Keep a person on money, access, and anything that changes a customer record. Measure cycle time, not “automation adoption”.
· Industry
AI employees for African ops teams
Lagos ops runs on WhatsApp, bad power, and ledgers that have to be right. Build for that.
· Foundations
AI employee vs RPA
RPA repeats a click path. An AI employee reads messy input and still has to stop at a gate.
· People
Write a job description for the AI employee
If you cannot write the job on one page, you are not ready to automate it. Models do not fix a fuzzy role.
· Build
Build vs buy an AI employee
Do not buy a platform to avoid naming a process. Do not build a platform to avoid using the tools you already pay for.
· Use cases
AI voice employee for inbound calls
Inbound voice is pick up, ask the right questions, and book the slot or route the call. A person still takes money, medical, and anger. Measure booked work and repeat-yourself rate.
· Industry
AI employees in insurance
Insurance needs the onboarding pack read once and a log a buyer can open. Not a smarter slogan.
· People
Checklist before you automate a process
If you cannot write the steps, count the pile, and name the exceptions, you are not choosing a model. You are choosing a mess.
· Industry
AI employees in fintech
Banks will not adopt an AI employee they cannot plug in before lunch. Make the API boring first.
· Use cases
AI employee for marketing ops
Marketing ops is a calendar, a brief, and a first draft. The employee researches, drafts, and keeps the week from going quiet. A person still owns claims, legal, and the publish button.
· Foundations
How to explain an AI employee to your board
Tell the board it is one worker, one process, a cost, an owner, and a gate. Not a platform.
· Use cases
AI employee for recruiting
Recruiting AI is a first pass, not a hire. Keep a person on the decision. Write the criteria down. Treat NYC LL144 and the EU AI Act as reasons to go slower, not as footnotes.
· Industry
AI employees in healthcare clinics
Clinics do not need a chatbot. They need the incomplete file chased and the platform up on Monday.
· Use cases
AI employee for sales follow-up
Follow-up is lookup, a first note, and a flag for the leads worth a call. Do not let the employee discount or close. Measure replies that turn into meetings, not open rates.
· Use cases
AI employee for customer support
Support is a queue with a job description. Give the employee the repeats, a state handoff, and a 10–15% escalation target. Leave refunds and medical advice alone on day one.
· Notes
Keeping 6,000 clinics online
What we learned running a health credit platform when downtime is not a slide in a retro.
· Notes
Stop reading the same document twice
How we cut review work in half without pretending a model never gets it wrong.
· Notes
What buyers open first
If someone wants to buy the company, they will look at cost, uptime, and whether you can explain the database.
· Notes
Make the API boring
Banks will use you if they can plug in before lunch. That is the whole job.