Main Line AutomationRadnor, PA · Est. 2026

What we automate

AI assistants for the office.

A helper for your staff that drafts replies, sums up long threads and looks things up in your own documents. It is set up to name the document it drew on, so a person can check each draft against the source before it is used.

How it runsFour steps · two need a person

  1. Chat or emailA question, or a blank pageYou start
  2. Your documentsIt looks in your documentsAutomatic
  3. Draft with its sourceA draft comes back to checkAutomatic
  4. Your reviewA person reads itYou read

You read every draft before anyone else sees it.

The problem

The answer is in a binder somewhere, and one person knows which one.

An assistant can find it and draft the reply. A person still reads it first.

Departs
A question, or a blank page.
Arrives
A draft or an answer, with the document it names as its source.
Your part
You read every draft before anyone else sees it.

Every office has a binder: the price sheet, the returns policy, how this exact situation was handled last time. One person knows where it is, and everyone else asks that person.

Then there is the blank page. The reply to a long email thread, the summary of a meeting, the letter you have written many times with small differences. Starting is the slow part.

An AI assistant helps with both, within limits. It can search every document you gave it at once, and it can produce a usable first draft. It can also be wrong while sounding entirely sure. That is why it is set up as a helper for your staff, and not as something that answers customers by itself.

How it runs

From a question to a draft a person has read.

Four steps, in order. The line under each one says whether it is automatic or waits for a person.

  1. Step 01

    A question, or a blank page

    Someone in your office asks the assistant something, or asks it to draft a reply or sum up a thread. Your staff use it. Customers do not talk to it.

    Started by your staff.

  2. Step 02

    It looks in your documents

    It searches the material you chose to give it: your handbook, your price sheet, your standard letters. It is instructed to answer from those and to say when the answer is not there.

    Automatic.

  3. Step 03

    A draft comes back to check

    The answer or draft arrives with the document it says it drew on, so it can be checked against the original. If it names none, treat it as a guess. Nothing has been sent anywhere.

    Automatic. It is a draft and nothing more.

  4. Step 04

    A person reads it

    Someone who knows the work compares the draft with the source, corrects it and decides whether to use it. Only then does it go anywhere.

    Waits for a person: every draft is read before it is used.

What you get

An assistant, its instructions and a list of what it reads.

All of it is set up in your name unless the written plan says otherwise. If we part ways, what was built for you stays with you.

  • An assistant set up in an account your business owns, working from documents you chose.
  • A written list of what it was given to read and what was deliberately kept out.
  • Its standing instructions in plain English: what to do, what to refuse and to show its source every time.
  • A one-page guide for your staff: how to ask, how to check a draft against its source and when not to use it.
  • A set of real questions from your office, tried before handover, with the wrong answers recorded beside the right ones.
  • The logins in your name and a number to call.

The limits

An AI draft can be confidently wrong.

It can state a wrong price, a policy you do not have or a document that does not exist, in the same sure tone as a right answer. So every draft is read by a person before it is used, and we do not promise that its answers are accurate.

Where it stops and waits for a person

  • Every draft, every time. Nothing the assistant writes reaches a customer, a record or a decision until a person has read it.
  • Every figure, date, name and quotation. Each is checked against the source it cites, because a wrong one looks exactly like a right one.
  • A question your documents do not answer. It is instructed to say so. If it answers anyway, the missing source is the warning.
  • What it may read. Adding a folder of documents is your decision, made on purpose.

What it will not do

  • Be right every time. We do not say it will. The same question can get a different answer on another day.
  • Send, file, post or pay. It drafts. People act.
  • Summarize a recording nobody agreed to. In Pennsylvania, recording or transcribing a call or a private conversation needs the prior agreement of everyone on it. An assistant we set up is not pointed at recordings made any other way, and the staff guide says so.
  • Answer your customers unsupervised. It does not give anyone legal, medical, tax or financial advice either.
  • Decide anything about a person. Who is hired, approved, treated or turned away is never left to it.
  • Read what you did not give it. Confidential, patient or legal-matter material stays out unless the right agreements are in place and you have chosen to include it.

How to read a draft. Open the source it cites. Check every number, name and date against it. If the draft cites nothing, treat it as a guess.

Questions

Questions about AI assistants.

Cost, upkeep and the rest of the questions

How often is it wrong?

Often enough that every draft has to be read, and we will not put a number on it. It depends on the question, on the state of your documents and on the AI model underneath.

What we can describe is how it goes wrong. It fills a gap with something that sounds right, or it cites a real document for a thing the document does not say. That is why it is set up to name the document behind each answer, and why a person compares the two.

Is what we give it kept private?

That depends on which AI product is used and on which plan, and the terms differ. We set the assistant up in a business account that belongs to you. Before anything is loaded, we go through what that provider commits to in writing about your material, and you decide what goes in.

Staff pasting client details into a free personal chatbot account is a different matter, and those commitments do not cover it. Patient, legal-matter and similar regulated material stays out until the right agreements are in place.

Can it answer our customers directly?

Not the way we set it up. An assistant that answers the public with nobody reading first will sooner or later tell a customer something untrue in your name. It drafts for your staff, and your staff decide what to send.

Will it keep working if we leave it alone?

Not indefinitely. The companies that make AI models retire older ones, and an assistant built on a retired model stops working until it is updated. Your documents change too: an assistant reading last year’s price sheet gives last year’s prices.

Someone has to own both. That can be you, with the guide, or us, if you choose upkeep.

Next to this one

Work that needs rules, not AI.

Each is its own small job, built and handed over before the next begins. See all six.

Counted, not drafted

Weekly numbers

Figures are counted by fixed rules, not drafted by AI. If a number has to be right, a rule should produce it.

How this one works

Start here

Tell us what your office looks up, and what it writes again and again.

The first conversation is free, with no obligation. If the work is worth automating, you will hear what it would take. If it is not, you will hear that too.

The first conversationAbout half an hour, no chargeBy phone or video, with no obligation.
Emailtykennedy@mainlineautomationllc.comIt goes to Ty Kennedy.
WhereRadnor, on the Main LineServing the Main Line, Philadelphia and its Pennsylvania suburbs. No walk-in office: we talk by phone, by video or at your place of business.
Not ready to talkRun the free tallyFive minutes, in your browser, to see how many hours the repeat work really takes.