Every desk gets an agent. Every agent answers to you.

Ads, quotes, orders, invoices, the phones. We build your company its own system and put an agent on each desk. Agents draft and check. Your people approve anything that spends money or reaches a customer.

Each light is an agent. AI where it helps, plain rules where it does not.

The labels are examples of jobs agents do. Some are live for a client and some are sample builds: each demo below says which.

Sound familiar?

Here are the same three businesses with every agent switched off. Pick a job, or just watch. They come back on one at a time.

  1. The ad bill goes up every month and nobody can say what it bought.

    An agent reads what people actually searched for and flags the terms that spend money and never sell. You approve the block. Bids and budget stay yours.

    Uses AI. Runs our own marketing today. See the full demo

  2. A customer asks for a quote and it takes a day to pull the prices together.

    Ask for the quote in the AI tool you already use. It is drafted inside your system with your prices and your stock, and code checks the margin. You approve before the customer sees it.

    Uses the AI tool your team already has. Live for a client today. See the full demo

  3. The phone rings while everyone is busy, and that customer calls someone else.

    An agent texts back in seconds, asks what is wrong and offers two open times. Your dispatcher confirms the booking.

    Uses AI. Sample build, not yet live for a client. See the full demo

  4. An order comes in, and someone types it all again to buy the stock.

    By morning an agent has drafted the purchase orders, one per vendor, and marked what can ship from stock. A person presses Create.

    Uses AI. Live for a client today.

  5. Invoices sit in someone’s inbox and nobody checks them against what we agreed.

    An agent reads each invoice and lines it up against the purchase order. If a price is higher than agreed, it flags the line and says why. A person decides.

    Uses AI to read the PDF. Live for a client today. See the full demo

Watch it work, desk by desk

An ad account, a quote, a vendor invoice, a missed call. The orange light is an agent doing the job by itself. Some agents use AI, some follow plain rules, and each demo says which. The cursor is one of your people, and the last move is always theirs.

Shown with sample data. The systems behind them are real.

115 releases in 17 days, for one company

A US hardware reseller ran its sales, purchasing and finance on an aging portal, a CRM subscription, spreadsheets and email. The team was working in the new system on day 10, and the first AI agent went live on day 13. Every dot is a release that went live.

  1. Day 1: 0 releases
  2. Day 2: 2 releases
  3. Day 3: 9 releases
  4. Day 4: 1 release
  5. Day 5: 0 releases
  6. Day 6: 6 releases
  7. Day 7: 11 releases
  8. Day 8: 7 releases
  9. Day 9: 4 releases
  10. Day 10: 11 releases
  11. Day 11: 1 release
  12. Day 12: 1 release
  13. Day 13: 18 releases
  14. Day 14: 16 releases
  15. Day 15: 13 releases
  16. Day 16: 6 releases
  17. Day 17: 9 releases
  1. Day 1 First line of code.
  2. Day 2 First release to production.
  3. Day 3 Vendor portal live, with purchase orders signed online.
  4. Day 6 Their own CRM live, next to the one they pay for.
  5. Day 9 Staff connect the AI tools they already use.
  6. Day 10 The whole team moves in. The old portal is frozen.
  7. Day 13 First AI agent live: it reads and checks vendor invoices.
  8. Day 14 Second AI agent live: it drafts purchase orders. Also the morning brief, bank matching and the vendor inbox, on plain rules.
  9. Day 16 An assistant inside the portal for every member of staff.

Day 10. Everyone at the company was working in the new system, with their data moved over and checked.

Day 13. The first AI agent went live, once it had passed its test set.

About 31,500 automated tests run before any change goes live. That is what makes shipping several times a day possible.

What they have now

A staff portal
Orders from the CRM and the online store, purchase orders, vendor invoices, payments, tracking, stock, the product catalog and a finance page.
A vendor portal
Vendors log in, sign purchase orders, send invoices and upload price lists. Nobody retypes them.
Their own CRM
Contacts, deals, quotes and reports, running side by side with the old one until they are ready to switch.
AI agents at work
They read and check vendor invoices, draft purchase orders and read price lists from PDFs. A person approves what they propose.
Routine work on plain rules
The morning brief, vendor reminders, bank matching and the vendor inbox run on tested rules, with no AI at all.
Quotes from any AI tool
Staff create customer quotes from the AI assistant they already use. Nothing is sent until someone approves it.
Help built in
An assistant inside the portal answers how-to questions, with a help page for every feature and short narrated videos for the main tasks.

A senior engineer, directing a team of AI agents

We build your system the way we ask you to run it: agents write, test and review the code, and a person approves every release. You talk to the engineer who is building yours.

No agent marks its own work.

  1. Write the rules

    An agent writes down how your business works, with test cases.

  2. Split the work

    A lead agent breaks it into small tasks.

  3. Build with tests

    Several agents build at once, each with its own tests.

  4. Block what is untested

    A gatekeeper agent refuses anything without proof.

  5. Try to break it

    An independent reviewer that did not write the code attacks it.

  6. Approve and ship

    The engineer approves. It goes live and is checked in production.

No agent spends, pays or sets a price on its own

Not on the ad account, not on a vendor bill, not on a customer quote. These are rules in the code, checked by tests. They are not instructions we hope the AI follows.

  1. Agents propose. People approve.

    You decide which actions need a sign-off. Paying, pricing, deleting, ad spend and anything a customer sees always do. Routine reminders run on rules you switch on yourself.

  2. Code does the math.

    AI reads documents and drafts messages. Every dollar amount is calculated by ordinary, tested code, never guessed by a language model.

  3. Everything has a name on it.

    Every action, by a person or an agent, is logged with who, what and when.

  4. New features start switched off.

    Anything risky ships turned off. You turn it on once you have seen it work on your own data.

  5. AI on a budget.

    You set a monthly spending cap for the agents. They cannot go over it, and one switch stops them all.

  6. Outside text is data, never orders.

    A vendor email or a PDF cannot tell an agent what to do. And nobody changes bank details by email, ever.

Whatever you sell, there is an office behind it

Best fit: 10 to 200 people, work that passes between desks every day, and an owner who wants to be involved.

Marketing, sales and any busy office

Ad spend watched, campaigns and quotes drafted, invoices matched, with a person approving anything that spends money or reaches a customer.

  • Run search and social ads, with approval
  • Draft email campaigns that wait for your send
  • Draft quotes with the margin checked
  • Match invoices and bank payments
  • Answer how-to questions for new staff

Our own marketing runs on agents today. Finance and quoting are live for a client.

Plumbing, HVAC, electrical and other service companies

Calls answered, estimates followed up, jobs invoiced and reviews requested, while your techs stay on the job.

  • Text back every missed call
  • Book jobs onto tomorrow’s board
  • Follow up on open estimates
  • Send invoices and payment reminders

Sample builds today. Now booking pilot clients.

Distributors, resellers and online sellers

Orders, purchase orders, vendor invoices, stock, quotes and customers in one system, with agents doing the checking and the chasing.

  • Match vendor invoices to purchase orders
  • Draft purchase orders from new orders
  • Chase late shipments and file the tracking
  • Keep prices and stock current
  • Match bank payments to bills

Live for a client today.

Start with one desk, not a big project

We start where the work hurts most and add desks as you go. Every step has a fixed price agreed before it starts, and you can stop after any of them and keep what was built.

  1. Shadow Week

    Five working days

    Our agents shadow the desks you choose, with read-only access: a sample of the shared inbox, a few months of the records your team works in, such as orders, invoices, quotes, ad reports or bookings, and a walkthrough of how the work gets done today. We talk to you and two or three of your people.

    You get

    • A one-page map of every manual step in the office
    • A working agent, running on your own data
    • A fixed-price plan for the first build

    From you

    Someone who can say yes, read access or exports, and about three hours of your team’s time.

  2. Build Sprint

    About two weeks each

    We build the first part of your own system, connect it to the tools you already use, and switch on the first agents in propose-only mode. It runs side by side with the old way until you are happy.

    You get

    • The software, with logins and roles for your staff
    • Agents, approval rules and a full activity log
    • A help page for every feature, and short videos for the main tasks
    • The code, in an account that belongs to you

    From you

    One decision maker for 15 minutes a day, and one person on staff who tries things out.

  3. Run and Improve

    Month to month

    We host it, watch it, fix what breaks and keep improving it every week. Each month you get a plain report of what the agents did and what your people approved.

    You get

    • Hosting, monitoring, backups and fixes
    • New features and agents as you ask for them
    • AI running costs passed through at cost, under a cap you set

    From you

    A list of what annoys you. We will keep working through it.

Questions owners ask us

What if an agent makes a mistake with money or a customer?

No agent can move money on its own. AI reads documents and drafts messages. Ordinary, tested code does every calculation. Anything that pays, changes a price, spends on ads or reaches a customer waits for a person to approve it, and every action is logged with a name and a time.

We are not technical. Will this work for us?

That is who it is for. You tell us how the work gets done today and we do the rest. Every feature comes with a plain help page, the main tasks have short narrated videos, and an assistant inside the system answers how-to questions.

Do we have to throw out the software we already use?

No. We usually start by connecting to what you have. If replacing a tool makes sense later, the new system runs side by side with the old one until you are confident, and only then do you switch.

Is our data safe?

Your data lives in your own system, not in a shared platform. Agents only see what their job needs, you can see every document an agent touched, and we connect to AI providers through their business services, not consumer chat apps.

What happens if you are not around anymore?

You own the code, the data and the accounts. The software comes with thousands of automated tests and plain-language documentation, so another developer can pick it up. There is no annual lock-in.

Will this replace my staff?

It takes over the copying, checking and chasing, not the people who know your customers. They get that time back for the work only they can do.

How long until we see something real?

At the end of the first week you see a working agent on your own data. For the reseller in our case study, the whole team was working in the new system on day 10.

How big is your team?

Every project is led by a senior engineer who directs a team of AI agents that write, test and review the software. That is why we move in weeks, and why you talk to the person building your system.

Show us the desk that eats your week

Marketing, sales, operations, finance or the front desk. Tell us the one job you would hand off first. We will tell you honestly whether an agent can take it, and whether it needs AI or plain rules.

  • A 30 minute call about how that desk works today. No slides.
  • If it is a fit, we start with one Shadow Week.
  • If it is not, we will say so.

Rather write an email? albert@leantechnologies.net

What kind of business? (optional)
How many people? (optional)
When would you start? (optional)