AI employee job description · Jon Weberg

AI employee job description: how do you write one, field by field?

November 01, 2026•9 min read

An AI employee job description is a 10 field page that gives an AI one handoff to own: when it runs, what it reads, what it writes, what it decides alone, when it hands to a person, what it never does and how its work is measured. Jon Weberg's INSTALL method uses one page per hire, and the INSTALL book ships eight of them.

You have written prompts, and you have watched one work for a week before it went quiet, haven't you? A prompt describes a task, so the AI does a task; a job description describes a handoff, so the AI owns a result on day 30 the way it did on day 1. It is the core document of Hire AI.

TL;DR

  • Ten fields: title, mission, runs, reads, writes, decides alone, hands off, never does, measured on, log.

  • Every job description starts from one handoff, the join between two people or teams that nobody owns.

  • The hands off field names the signals that stop it, and any thread below 70% confidence goes to a person.

  • A new AI employee sends nothing in week one and earns send rights after 2 clean Fridays.

  • Measure the rate on the handoff, not a count of tasks, and read it in a weekly log.

What is an AI employee job description?

An AI employee job description is the standing instruction set that defines one AI role's mission, inputs, permissions, limits, escalation triggers and success measures, so the AI behaves the same way on every run.

Most published templates agree on the core: Workato's recommended job description structure runs from role and purpose through operating principles and what to avoid, and Sista AI's 2026 template adds boundaries, tools and access, definition of done and optional KPIs. The INSTALL page keeps the measure mandatory, so which handoff would you write on line one tonight, and what is it worth a month?

How is it different from a prompt or a job description for a person?

A person's job description lists duties and a prompt asks for one output, while an AI employee's job description lists triggers, signals, limits and a measure, because an AI reads only the lines you give it.

Prompt

AI employee job description

Starts when

You type it

An event or a schedule

Covers

One output

One handoff, every run

Knows

What you pasted

Every source on its reads list

Stops when

The answer appears

A handoff signal, or confidence below 70%

Judged by

A glance

The measured on numbers each Friday

That page is what keeps an AI sales agent built in GoHighLevel steady after week one, and the week one send rule is the same caution you want from any new hire, isn't it?

What are the ten fields of an AI employee job description?

The ten fields are title, mission, runs, reads, writes, decides alone, hands off, never does, measured on and log, each answering a question the AI would otherwise guess.

Field

The question it answers

Email Triage example

Title

Which handoff does it own?

AI email triage employee

Mission

What is true when the job is done?

Every thread sorted, money with a person

Runs

What triggers it, and when?

Overnight, one pass before your day

Reads

What inputs does it open first?

New mail, sender history, the thread, the Business Brain

Writes

What outputs does it leave behind?

Labels, held drafts, a daily log

Decides alone

What may it choose without you?

The bucket, archiving noise, drafting

Hands off

Which escalation triggers stop it?

Price, refunds, complaints, below 70% confidence

Never does

What is off limits?

Sending in week one, quoting a price

Measured on

Which numbers does it answer for?

Drafts approved without edits

Log

What does one entry record?

Counts per bucket, handoffs, doubts

The eight filled versions, one per revenue job, are in AI job description templates; which of the ten fields is missing from the last prompt you wrote?

How do you write the title and the mission line?

The title names the handoff and the mission states what is true when the job is done, in one sentence you can check at a glance.

Name the title for the join, so "AI second offer employee" tells you where it stands, then write the mission as a finished state, the way Sales Close reads: every qualified no gets the next approved offer. Where on your calendar does your sentence turn true or false? A mission that needs a paragraph is two jobs.

What goes in the reads and writes fields?

Reads lists every source the AI opens before it acts, and writes lists every output it leaves behind, named exactly.

The average worker receives 117 emails a day, as Microsoft WorkLab reported in June 2025, so an inbox job reads each one with its sender history and thread, plus the Business Brain, twelve short sections on what you sell, who buys and your voice; since the AI remembers only what you wrote down, what does it need to know before day one? That list is set out in what an AI employee needs before day one.

What may an AI employee decide alone?

Decides alone lists only choices that are reversible and inside rules you wrote, such as a label, an archive or a draft held for review.

For Email Triage that means the bucket, archiving noise and drafting, and for Sales Close it means which pre approved version to present, never a new price; if your AI could decide only three things alone, which three would let you sleep?

When does it hand off to a person, and what does it never do?

Hands off names the signals that stop the AI and pass the thread to a named person, and never does names the actions it takes under no circumstances.

Email Triage stops on six signals: anything new, a price question, a refund, a complaint, a decision, and any thread below 70% confidence; which of those six would have caught the email you most regret from last quarter? Never does lists what you already keep behind a person, a price, a date, a sale, and the full list of those lines is in what an AI employee should never touch. The whole page for this role is the full Email Triage job description.

How do you measure an AI employee's work?

You measure an AI employee on the rate at its handoff, read in a weekly log, never on a count of tasks completed.

Jon Weberg writes these pages the way he ran a 22 closer floor, by handoff and rate. Email Triage answers for drafts approved without edits, and the AI second offer employee answers for revenue per declined buyer and days from the no to first contact, read at the Friday check in. Sales Close is the written form of Frank Kern's result, $41,253 in 3 days from buyers of an $8,600 program who said they could not afford it, through one $2,500 offer with no new ad spend, after which Kern hired Jon four more times, money that had waited on the declined and lost buyer seam; when a buyer says no on your calls, how many hours pass before anyone sends the next offer?

What are the most common mistakes in an AI employee job description?

An AI employee job description fails in five common ways.

  1. A task instead of a handoff.

  2. No measure for Friday.

  3. Money inside decides alone, such as a discount.

  4. No week one send rule.

  5. A log nobody reads, so drift surfaces in week six instead of week two.

Which of those five would your current setup fail first?

What does a complete AI employee job description look like?

A complete AI employee job description fills all ten fields for one handoff, as Email Triage does below.

Field

Email Triage

Title

AI email triage employee (INSTALL: Email Triage, the first hire)

Mission

Every thread sorted, obvious replies drafted in your voice, nothing about money or a complaint left without a person.

Runs

Overnight, plus one pass before work begins.

Reads

Mail since the last run, sender history, the thread, the Business Brain.

Writes

One of four bucket labels per thread, held drafts, handoff summaries, a daily log.

Decides alone

The bucket, archiving noise, draft replies, acknowledging known questions.

Hands off

Anything new, price, refunds, complaints, decisions, any thread below 70% confidence.

Never does

Sends in week one, quotes a price, promises a date, closes a sale, makes a call.

Measured on

Drafts approved without edits, threads handed off, buying questions answered the same morning.

Log

Daily: counts per bucket, drafts held, handoffs, anything it was unsure of.

Next Friday you read one page of log, approved drafts climbing since Monday, one line changed, and the same behavior every morning after, so how long would your first one take to write tonight?

Frequently asked questions

How long does it take to write an AI employee job description?

About half an hour for the first one, once you know the handoff. The thinking is choosing the handoff and the measure, and no code is involved.

Is a job description the same as a prompt?

No. A prompt asks for one output now. A job description defines a standing role: what it reads, what it decides, when it stops and how it is judged.

Can ChatGPT write the job description for me?

It can draft the wording. You decide the handoff, what it decides alone and the never does line, because those carry your money and your clients.

Do I need one job description per AI employee?

Yes. One job description covers one handoff, so two handoffs means two AI employees, each with its own measure and log.

How often should the job description change?

Read the log every Friday and make one change, so the page improves weekly without rewrites.

Where do I find job descriptions for sales, support and marketing?

The INSTALL book ships eight, one per revenue job: Lead Gen, Follow-up, Sales Close, Support, Retention, Marketing, Content and Operations.

Write your first one tonight

Write the ten fields for your inbox tonight, run it a week with every send held, and read the log on Friday; all eight finished pages come in INSTALL, the $7 book. When the handoff you need covered crosses several systems, have it built for you: Install My AI Seam Architect.

Custom HTML/CSS
Jon Weberg

Jon Weberg

Jon Weberg is a revenue consultant, three time author and international keynote speaker who finds and closes the seams between a company's teams, the handoffs where revenue is lost. He is the founder of Profitalize and the originator of The Thirteen Seams, the AI Seam Architect and the INSTALL method for hiring AI employees. He led the campaign architecture for the Klotho Neurosciences and Greenland Mines merger announcement, which reached an investor audience of 5.1 million, served as Fractional President of a $70 million a year company, where he directed the sales, marketing and media buying directors and built the custom CRM it runs on, and has worked with Frank Kern and John Crestani.

LinkedIn logo icon
Instagram logo icon
Youtube logo icon
Back to Blog
Jon Weberg, founder of Profitalize
Written by

Jon Weberg

Founder, Profitalize · AI growth seam architect

Jon finds the handoffs where revenue leaks between your teams and installs the AI employees that own them. As Fractional President of AI Acquisition he directed its Sales, Marketing and Media Buying Directors and built the CRM a $70M a year company runs on. He is a three time published author, an international keynote speaker at Affiliate World, Traffic and Conversion Summit and Affiliate Summit, and has spent 13 years in direct response.

  • $1,790,467 in 28 business days · Tax Strategists of America
  • $41,253 in 3 days · Frank Kern
  • $270M referral revenue · NowLifestyle

About JonCase studiesLinkedInYouTube

Your next step

Find your first seam this week.

Forty five minutes on your own numbers, or the $7 book that has your first AI employee running by Friday.

Forty five minutes. We start installing your AI Seam Architect and price your three biggest seams before we hang up.