Use cases

See what an AI agent could take off your team's plate.

Most businesses run on a handful of systems that don't share information well: email, spreadsheets, accounting, scheduling, a portal or two. People fill the gaps by reading, retyping, matching and chasing.

From turning emails into work orders to matching invoices and following up on missing information, explore examples of agents working alongside your team.

These are examples, not a menu. If a person does the work on a computer, whether in email, a website, a spreadsheet or by moving information between your systems, an agent could take it on. Each example is based on a public business request. They are proposed solutions, not completed client case studies.

One job at a time Your rules from the first day Your team decides what it learns Built in ten business days

Where we start

Some jobs are a good fit for an AI agent. Some are not.

An AI agent is software that does a routine job the way your team does it, and asks a person when it is unsure. We start with one job your team can walk us through from start to finish. A good fit has three signs. The fit test checks all three before we build anything.

01 · It repeats

The same job keeps coming back.

Your team does it every day or every week. Each time, they read similar information and make one of a few routine decisions. For example, matching an invoice to an order, or entering a work order from an email.

02 · The rules exist

Your rules are written down, or can be.

Price lists, contracts, policies and written rules tell the agent what to do. Past orders show it what good work looks like. Some rules may live only in your people's heads. During the build, we write those down and confirm them with your team.

03 · It can be reviewed

A person can judge the result.

Someone on your team can tell a good result from a bad one, so they can review the agent's work. You decide what the agent may do alone and what it must never change. Unusual and high-stakes cases go to someone on your team for approval.

Why not just connect the systems?

Connecting two systems moves data from one to the other. It doesn't read the attachment, catch the mismatch, or chase the missing form, and that is where the hours go. Buyers who tried an integration still had the problem.

What the agent does on every case

  1. 01Reads everything about the caseEmails, attachments, spreadsheets and records in your systems
  2. 02Applies your rulesYour price lists, policies, limits and past cases
  3. 03Shows its workIts proposed result, and where each fact came from
  4. 04Acts or asksDoes only what it is allowed to do, and asks a person when it is unsure

Our method

The same four steps for every job.

Every example on this page follows these four steps. Each example shows what the agent starts out knowing, who reviews its work, and what stays with a person.

  1. 01

    Before the build

    Fit test

    We study one job and measure how it runs today: how often it comes up, how long it takes, and where it goes wrong. That gives us a "before" to compare against. We check the three signs above, and we find out which systems, inboxes and files hold the information and who must approve access. If an agent is not the right answer, or a ready-made tool would do, we tell you. Whatever we find, you keep the written map of how the job works today.

  2. 02

    Ten business days

    Build it

    Senior engineers gather what the agent needs from your systems, files, inboxes and meeting recordings, and from your people. They review the material with you and remove what is out of date or wrong. Then they build the agent and try it on your real past work, like invoices or orders, until it passes tests you agreed to in advance. They also set what it does alone, what needs approval, what it records, and how to switch it off.

  3. 03

    Ninety days, included

    Put it to work

    Your team starts using it. At first, someone on your team reviews the agent's work before anything is sent or saved. That person decides what it learns: each change they make to its work is saved as a rule, an example or an exception. It learns how your company does this job. It does not watch, rate or profile your employees. You can read everything it has learned and remove anything. It does useful work from the first day and improves fast. Small adjustments never fully stop, and that is normal. We fix problems, adjust it as unusual cases come up, and compare the results with the "before" numbers.

  4. 04

    After ninety days

    Keep improving it

    The person who works with it keeps adjusting its work, answering its questions and handing it new kinds of cases. It handles more over time. You decide when your team can stop reviewing a type of task. You can run it yourselves with the source code, the tests and a guide. Or our engineers run it for you: we watch it, update it when your systems or the AI models change, make sure what it learned stays inside the limits you set, and send a short monthly report. You also choose where it runs: on your own servers, in your own cloud account, or in our data center.

What you own

You own the agent, the information we gathered for it, everything it has learned, and its tests. What it learns is saved as plain-English notes. You can read all of them and delete any of them.

Use-case library

Fourteen use cases. Four kinds of work.

Each one comes from a request a business posted, paraphrased. Start with the kind of work that sounds most familiar, or with your industry. Each write-up shows the work, who does what, the rules the agent follows, what it learns, and when a ready-made tool would be enough.

Kind of work 1

Work that arrives by email and phone

Work orders, bills, voicemails, texts and maintenance requests reach the business as messages, and each one has to be read, sorted and entered into the system that runs the work. The software holds the record once it exists, but someone still has to decide what each message is, where it belongs and whether it fits the rules, so people do it by hand.

The first principles →

Kind of work 2

Systems that don't talk to each other

This is the work of carrying the same facts from one system into another: a client's details into each carrier site, a client's figures into a review sheet, a booking into a door code. Connecting the systems moves the data, but it does not read the email, check the figures against the rules or chase the missing piece, so people still do that part by hand.

The first principles →

Kind of work 3

Checking and matching

Checking and matching is the work of holding one record against another: a bank line against the books, a delivery ticket against an invoice, a payout against the sales behind it, a customer's request against the price list. People still do it by hand because the records never line up exactly, and someone has to decide what each mismatch means.

The first principles →

Kind of work 4

Follow-ups and deadlines

This is the work of noticing that a date is coming or something has changed, then getting the right person to act before it is too late. EZLynx, AppFolio and the city's DOB and HPD records already hold the dates, but people still read them line by line, re-enter them into spreadsheets and decide by hand who needs a follow-up.

The first principles →

Your use case

The right starting point is work you can show us.

Show us one job your team wants help with. The first call costs nothing. We will tell you whether it fits a ten-day build and why.

Show us the work