A chatbot answers. An AI worker works.
Chatbot
↓
AI worker
↓
↓
↓
↓
That's the difference.
What would you like an extra pair of hands with?
Pick as many as you like — the wall below narrows to match. (No filters selected shows everything.)
Yes. They can actually do things.
Tap any card for what it could do, what it would need, and what stays in your hands. Badges are honest: Works today means a Factory worker genuinely helps with this right now; Needs a connection means the ability depends on connecting outside systems — which we haven't switched on yet, and will only ever add with your explicit permission.
Read my emails
and tell me which ones actually matter.
Needs a connection — we'll tell you first
What it could do
With a connection to your inbox, a worker could scan new mail, spot what genuinely needs you and summarise the rest.
What it would need access to
A connection to your email. Email connections aren't live at the Factory yet — today you can paste an email in and your worker will tell you what matters and draft the reply.
What you still control
What it may read, what counts as important, and whether it may do anything beyond summarising.
Draft the replies
and wait for me to approve them.
Works today
What it could do
Give your worker the message and the situation, and it drafts the reply in your tone — firm, friendly or somewhere in between. You approve, tweak or bin it.
What it would need access to
Just the message and any context. Works entirely in conversation today.
What you still control
Nothing is ever sent by the worker — you copy, edit and send everything yourself.
Watch my diary
and warn me when I've created a ridiculous week.
Needs a connection — we'll tell you first
What it could do
With a calendar connection, a worker could look ahead, spot the double-bookings and the no-lunch days, and flag them before they happen.
What it would need access to
A calendar connection — not live at the Factory yet. Today: paste your week in and your worker will happily point out where it's gone wrong.
What you still control
What it can see, and whether it only warns you or suggests fixes.
Prepare me for meetings
before I even remember they're happening.
Works today
What it could do
Tell your worker who you're meeting and why, and it prepares the questions to ask, the points to make and the things you'd forgotten you promised.
What it would need access to
Whatever you can tell it about the meeting. The more you've taught it about your business, the sharper the prep.
What you still control
Everything — it prepares, you decide what to use.
Research something
and give me the answer rather than 47 browser tabs.
Works today
What it could do
Describe what you're trying to decide and your worker helps you structure the question, compare what you've found and reach a defensible answer.
What it would need access to
The information you give it. Factory workers don't browse the live internet yet, so you bring the facts and it brings the thinking.
What you still control
It recommends; you decide.
Chase an unpaid invoice
politely, persistently and without me remembering.
Needs a connection — we'll tell you first
What it could do
Dave could check which invoices have passed their due date, prepare the appropriate follow-up and either show it to you for approval or send it according to the rules you've given him.
What it would need access to
Your invoicing information. Accounting connections aren't live yet — today Dave drafts the perfectly-pitched chaser and you send it.
What you still control
Who gets contacted, what Dave can send and whether approval is required.
Qualify a new enquiry
before I waste half an hour on it.
Works today
What it could do
Paste in the enquiry and your worker checks it against your rules — budget, location, red flags — and tells you whether it's worth your time, plus the reply to send either way.
What it would need access to
The enquiry and your rules for what a good customer looks like (taught once, kept as facts it remembers).
What you still control
The rules are yours; it never contacts anyone.
Follow up leads
that would otherwise disappear.
Works today
What it could do
Teach your worker about your open leads and it keeps those facts, drafts the follow-ups and reminds you — when you ask — who said 'ask me again in March'.
What it would need access to
You tell it about the leads; it keeps them as taught facts you can see and edit.
What you still control
Every message is a draft until you send it.
Prepare quotes
using the rules I give it.
Works today
What it could do
Teach your worker your pricing rules once, and it prepares consistent quotes from each job's details — including the questions to ask when information is missing.
What it would need access to
Your pricing rules, taught once and kept as facts, plus each job's details.
What you still control
You set the rules; you approve every quote before anyone sees it.
Create content ideas
from what's actually happening in my business.
Works today
What it could do
Tell your worker what's going on — jobs finished, questions customers keep asking — and it turns that into content ideas that sound like you, not like an AI.
What it would need access to
A few minutes of you talking about your week.
What you still control
Ideas and drafts only; nothing is posted anywhere by the worker.
Turn one idea into content
for multiple channels.
Works today
What it could do
One good idea in; a post, an email paragraph, a short script and a caption out — each written for where it's going.
What it would need access to
The idea, and what it's been taught about your tone.
What you still control
You edit and publish everything yourself.
Watch for something
and tell me when it happens.
Needs a connection — we'll tell you first
What it could do
Monitoring the outside world needs connections and background running — abilities Factory workers don't have yet. When they arrive, you'll switch them on per worker, and be told before anything costs extra.
What it would need access to
A connection to whatever's being watched — not live yet.
What you still control
What's watched, how often, and what the worker may do about it (usually: just tell you).
Organise a trip
around my actual requirements.
Works today
What it could do
Tell it the dates, the budget, who's coming and what you can't stand, and your worker builds the shortlist criteria, compares the options you paste in and produces the itinerary.
What it would need access to
Your requirements and the options you've found.
What you still control
It plans; you book. It can't spend a penny.
Remember important things
so I don't have to.
Works today
What it could do
Everything you teach your worker is kept as individual facts — preferences, rules, things about your business, how you like things done. You can see every fact it knows, and edit or delete any of them.
What it would need access to
You telling it, once.
What you still control
You can see, change or delete anything it has been taught, any time.
Prepare my day
before I start it.
Works today
What it could do
Tell your worker what's on and it turns the chaos into a plan: what to do first, what to prepare, what can safely slide.
What it would need access to
A quick brain-dump from you each morning (or the night before).
What you still control
It suggests the plan; you live it.
Help organise presents
before someone's birthday becomes tomorrow.
Works today
What it could do
Teach it the important dates and the people, and ask it each month what's coming. It'll suggest ideas matched to what you've told it about each person.
What it would need access to
The dates and a bit about the people.
What you still control
It reminds and suggests when you ask; you do the buying.
Keep a running to-do list
and actually chase me about it.
Works today
What it could do
Teach your worker what's on your list and it keeps each item as a fact. Ask 'what's outstanding?' and get an honest answer — including the thing from three weeks ago you were hoping it had forgotten.
What it would need access to
You telling it what's on the list. (It chases when you check in — it can't yet message you out of the blue.)
What you still control
Your list, your priorities; it never shares it anywhere.
Compare options
and explain the useful differences.
Works today
What it could do
Paste in the three quotes, laptops, or vans and get back the differences that actually matter for your situation — not a spec sheet.
What it would need access to
The options and what matters to you.
What you still control
It recommends with reasons; you choose.
Turn messy information
into something organised.
Works today
What it could do
Notes from a call, a rambling voice-note transcript, a chaotic list — in. A structured summary, action list or table — out.
What it would need access to
The mess.
What you still control
You decide what happens with the tidy version.
Work through a process
the same way every time.
Works today
What it could do
Teach your worker your process — onboarding a client, closing a job, month-end — and it walks through it with you identically every time, never skipping the boring step.
What it would need access to
The process, taught once in plain English and kept as standing instructions.
What you still control
You wrote the process; you can change it whenever.
Move information between systems
I've connected.
Needs a connection — we'll tell you first
What it could do
This one genuinely needs connections to your other software, which Factory workers don't have yet. When connections arrive, you'll enable each one yourself, per worker.
What it would need access to
Connections to the systems involved — not live yet.
What you still control
Which systems, which direction, and what needs your approval first.
Create reports
without me manually assembling them.
Works today
What it could do
Give your worker the numbers and notes, and it assembles the report — structure, summary, the 'so what' — in your format.
What it would need access to
The raw information from you (it can't pull data from other systems yet).
What you still control
You review before anyone else sees it.
Notice something needs attention
and bring it to me.
Works today
What it could do
Within a conversation, workers are good at spotting what you've missed — the unanswered question in a thread, the gap in a plan, the assumption that doesn't hold. Watching things unattended needs connections that aren't live yet.
What it would need access to
The material, in the conversation.
What you still control
It flags; you act.
Ask me for approval
when it shouldn't decide alone.
Works today
What it could do
Built into the whole idea: Factory workers only draft, suggest and prepare. Nothing goes anywhere without you. As real-world abilities arrive, approval rules come with them — by design, not as an afterthought.
What it would need access to
Nothing. This is the default, not a feature you switch on.
What you still control
All of it. That's the point.
Do repeating jobs
without getting bored.
Works today
What it could do
The fifteenth product description, the fortieth personalised follow-up, the same weekly summary — done to the same standard as the first, with none of the sighing.
What it would need access to
The pattern and the inputs.
What you still control
You check the output; it never publishes or sends.
The difference between "help me" and "do it"
Old way
"Write me an invoice reminder email."
AI worker
"Every Friday, check which invoices are overdue. Prepare the follow-ups. Don't send anything over £5,000 without asking me."
Old way
"What's in my diary tomorrow?"
AI worker
"Every evening, look at tomorrow. Tell me what I need to prepare and flag anything I've forgotten."
Old way
"Research hotels in Rome."
AI worker
"Find me somewhere in Rome matching my requirements, compare the sensible options and bring me the best three."
One is a task. The other is a job with rules. AI workers are built for the second kind — and the rules are always yours.
At work. In life. Both.
But how clever can I make it?
As specific as you like.
The options in the Factory are there to make designing your worker easy. They are not the limit. You can simply tell us what else you want:
- If the Factory can configure it now, we'll tell you.
- If it needs a connection, we'll tell you.
- If it would cost extra, we'll tell you before anything happens.
- If it can't currently do it, we'll tell you that too.
No pretending.
What can't an AI agent do?
An honest answer, because most pages skip this one. What any AI agent can do depends entirely on its permissions, its connections, the tools available to it, the information it has been given, the limits of the underlying AI services, its safety controls and the third-party systems involved.
- It cannot magically access accounts it hasn't been connected to. No connection, no access — full stop.
- It shouldn't make important decisions outside the authority you've given it, and a well-built one won't.
- It should ask the human when a decision matters — and asking is a feature, not a weakness.
- It isn't infallible. It drafts, you decide; that order exists for a reason.
Read more: what an AI agent can actually do, in detail and whether AI agents can be trusted.
So does it just run around doing whatever it likes?
No.
- You decide what it can access. Right now, a Factory worker accesses nothing outside your conversation and what you've taught it.
- You decide what it can do. Today that's chatting, drafting, planning and analysing — it can't send, spend, delete or touch outside accounts, because those abilities don't exist yet.
- You decide what needs approval. As real-world abilities arrive, each comes switched off, with approval rules you set.
- It doesn't get your passwords handed to it in a text box — connections, when they come, are authorised properly and revocable.
- You can pause your worker at any time. Paused workers don't chat, full stop.
More detail: how AI agent permissions work and are AI agents safe?
Beginner questions, straight answers
What is an AI agent?
An AI agent is software you give a job to, rather than just a question. Where a chatbot answers and forgets, an agent can hold a role, remember what you've taught it, work through multi-step tasks and — where you've granted permission and connected the right tools — take actions on your behalf.
What can an AI agent do?
Depending on how it's set up: research and compare options, draft emails and content, prepare quotes and reports, organise messy information, plan projects and trips, remember your preferences and rules, and work through processes step by step. With connections and permissions, agents can also act — sending, updating and monitoring — under rules you set.
What is the difference between an AI agent and ChatGPT?
ChatGPT is a general-purpose chat tool: brilliant, but it starts blank every time and belongs to everyone. An AI agent is set up for a specific job, keeps a memory you control, follows your standing rules and can be limited or extended with permissions. It's the difference between asking a stranger and briefing a colleague.
What is the difference between an AI agent and a chatbot?
A chatbot answers questions. An agent takes responsibility for a job: it works out the steps, remembers the context, uses whatever tools it's been allowed and comes back to you when a decision needs a human. Chatbots respond; agents work.
Can an AI agent send emails?
Some AI agents can, when connected to an email account with permission. Factory workers currently draft emails rather than send them — you stay the sender. Email connections are on our roadmap, and when they arrive you'll switch them on yourself, per worker, with approval rules.
Can an AI agent manage my calendar?
With a calendar connection, an agent can check availability, spot clashes and propose plans. Factory workers don't connect to calendars yet — today they'll happily plan your week from whatever you paste in, and warn you it looks ridiculous.
Can an AI agent automate admin?
It can take the thinking and drafting out of admin today: organising information, preparing documents, working through checklists, keeping lists and rules. Fully hands-off automation needs connections to your other software, which varies by product — at the Factory those connections aren't live yet.
Can an AI agent help run a small business?
Yes — as a tireless assistant rather than a manager. Quotes, follow-ups, customer replies, content, reports, processes and remembering how you like things done are all real, immediate help. You stay the decision-maker; a good agent makes your decisions faster and better-prepared.
Can an AI agent find customers?
It can help you qualify enquiries, write outreach worth sending, follow up leads consistently and sharpen your offer — which is most of what 'finding customers' actually involves. It can't conjure strangers out of the air, and you should distrust anything that claims to.
Can an AI agent follow up leads?
Yes. Tell it who's in the pipeline and it remembers them, drafts the follow-ups, suggests timing and keeps you honest about who's gone quiet. At the Factory, every follow-up is a draft until you send it.
Can an AI agent research things for me?
It can structure the question, compare options against what matters to you and turn a pile of findings into an answer. Whether it can search the live internet depends on the product — Factory workers currently work with the information you give them, which keeps them honest about sources.
Can an AI agent remember things?
Yes — this is one of the biggest differences from ordinary chat tools. A Factory worker keeps what you teach it as individual facts: preferences, rules, standing instructions and ways of working. You can read everything it has been taught, change it or delete any single fact at any time.
Can an AI agent work automatically?
Some agents can run in the background when given connections and standing instructions. Factory workers currently work while you're talking to them — everything happens in front of you and is saved as it happens. Background work is the kind of ability we'll only ever add with explicit permissions.
Can an AI agent use other software?
Only if it's been connected to it, by you. No agent can magically access accounts it hasn't been given. Factory workers have no outside connections yet; when connections launch, each one is switched on per worker and can be switched off again.
Can an AI agent work on my phone?
Yes. Factory workers live on our platform, not on your device — use the same worker from your phone, computer or tablet, and add it to your Home Screen for one-tap access. Lose your phone and your worker is exactly where you left it.
Do I need to know how to code to use an AI agent?
Not at the Factory. You describe the job in ordinary sentences, answer a few questions and teach it your preferences the same way you'd brief a new colleague. If you can write a text message, you're qualified.
Are AI agents safe?
They're as safe as their permissions. A Factory worker can only talk with you — it has no access to your email, files, money or accounts, so the worst case is a bad sentence, not a bad payment. Any future real-world abilities will be off by default and switched on by you.
Can I control what my AI agent is allowed to do?
Yes, and you should refuse any product where you can't. At the Factory you control what your worker knows, how it behaves, when it's paused and what it's called. As action abilities arrive, each comes with its own on/off switch and approval rules.
Can I create my own AI agent?
Yes. The Factory walks you through it: pick the kind of help you want, answer plain-English questions about your situation, and it designs the worker with you — role, personality, rules and all. Building and trying one is free.
How much does an AI agent cost?
At the Factory: building and trying a worker is free. Keeping workers active starts at £49/month for one active worker, £129/month for up to three, and £249/month for up to six. Each plan includes a monthly message allowance, and nothing ever spends money without you.
Can an AI agent replace an employee?
No, and be wary of anyone who says otherwise. An agent has no judgement, no accountability and no hands. What it genuinely replaces is the hours your existing people (including you) lose to drafting, formatting, chasing, summarising and remembering.
What jobs should I give an AI agent?
The repetitive, the structured and the forgettable: drafting replies, preparing quotes from your rules, tidying notes, planning weeks, chasing lists, turning one idea into five pieces of content. Anything you'd delegate to a capable assistant with clear instructions.
What should I NOT give an AI agent?
Final decisions that need human judgement, anything involving money moving without your approval, legal or medical calls, and anything you couldn't describe clearly to a person. If you wouldn't delegate it to a new assistant on their first week, don't delegate it to an agent.