How it works
How We Build an AI Agent for Your Business, Step by Step
From a free 30-minute audit to a working prototype in 5 business days, then a production agent typically live in about 2 weeks. Here is every step, what we need from you at each one, and how approvals and logs work once the agent is running.
What are the steps?
Six steps. The prototype is free, and the last step is optional.
1. Free 30-minute audit
A call by phone or video. You tell us how the work flows today: what arrives, from where, who touches it and which systems it ends up in. We ask about volume, mistakes and the jobs your team puts off.
By the end of the call we name the 2 or 3 jobs worth handing to an agent, with a rough cost and timeline for each, and agree on the one to prototype first. You keep that list whether you hire us or not.
2. A free working prototype in 5 business days
"I'll build you a working prototype in 5 business days, free, no obligation." That's Scott's promise, and here is what it means in practice:
- One workflow or one agent, built against your real tools in a test setup or on sample data you provide. It doesn't touch live customers or production data.
- It's working within 5 business days of the day we have the sample data and test access.
- You see it running on a video call.
- If you don't go ahead, you owe nothing and you keep a written summary of what it does and what the full build would cost. The working prototype itself stays with us.
- We take on 4 free prototypes a month, first come, first served. If the month is full, we give you the next start date.
The full terms are on the free prototype page.
3. Fixed price in writing
If you want the agent in production, we write a short scope. It covers:
- The job, in plain language, and what "done" looks like.
- The inputs the agent reads and the systems it writes to.
- The permissions it gets, and nothing beyond them.
- Who approves which actions, and who covers when they're away.
- What the agent won't do.
- An estimate of monthly running costs, such as model usage, based on your volume.
- One fixed price and a target go-live date.
A first agent is from $995, and the prototype work isn't charged on top. Prices for other builds are on our pricing page.
4. Production build, typically live in about 2 weeks
We create the agent's accounts in your live systems, connect it with the least access it needs, and turn the prototype into a production agent. The approval steps and the activity log go in at the same time, not as an afterthought. Then we test the awkward cases: the order with a missing PO number, the invoice in a strange layout, the caller who asks three questions at once.
Before go-live you get a staging version, a copy of the agent connected to test data or a sandbox of your systems. Your team runs real-looking cases through it and tells us what's wrong. We fix it and you test again. Nothing reaches a customer, your books or your CRM until you sign off. The production agent is typically live in about 2 weeks after you approve the price. Agents that need new hardware or many connected systems take longer, and the scope says so up front.
5. Go-live, with 2 weeks of support
The agent starts on real work with approvals switched on for anything that matters. For the first 2 weeks we watch the log with you, fix edge cases and tune the instructions. You also get documentation written for your team, not for developers, and a walkthrough of how to approve, pause and check the agent. You own what we build: the workflows, prompts and documentation.
6. Month-to-month support, if you want it
After that, monthly support is optional from $99/month. It's month to month, and you can cancel anytime. It covers monitoring and log review, fixes when a connected app changes, prompt tuning and small changes. If you'd rather run the agent yourselves, the documentation tells your team how.
What do we need from you?
Less than most owners expect. Nobody on your team needs to be technical.
- One owner: a person who knows the job and can answer questions within a day or two.
- Real examples: a few dozen recent samples of the work, such as emails, PDFs, quotes or call transcripts. The free prototype is built on these, and we can work with redacted copies.
- Your rules: the decisions your team makes by habit, like which customers get priority or when a quote needs a manager.
- Access for the agent: its own account in each system, set up by your IT provider or with us on a call. We never ask for a staff member's password.
- Approvers: who approves what, plus a backup for holidays.
- A couple of hours for staging: time for your team to test before go-live.
How do approvals work?
The agent does the preparation. A person makes the call on anything that matters.
When the agent wants to take an action that needs sign-off, it prepares it as a draft and sends an approval request to the right person, by email or as a Teams message. The request shows what came in, what the agent proposes to do, and why. The approver can approve it, edit it first, or reject it with a note. Every choice goes into the log, and rejected cases help us improve the agent's instructions.
By default, anything that sends money, signs something or emails a customer needs approval. You can tighten or loosen that per action. Here is a typical starting point:
| Action | Typical default |
|---|---|
| Read, sort and file an incoming email | Automatic, logged |
| Create a draft order, bill or CRM record | Automatic, as a draft only |
| Update a CRM field or add a note | Automatic, logged |
| Post an order or bill to your books | Needs approval |
| Email or text a customer | Needs approval. Routine messages like booking confirmations can be relaxed once proven |
| Send or schedule a payment | Needs approval. We recommend it stays that way |
| Sign or accept an agreement | Needs approval |
| Delete records | Not allowed |
What goes in the activity log?
Enough to answer "what did the agent do, and why?" for any run, months later.
Each entry records
- The time, and what started the run
- What the agent read
- What it decided, and its stated reason
- Each tool it used, and what changed
- Who approved, edited or rejected it
- Any error, and where the case went next
Where it lives
In a place you can open without us, such as a SharePoint list, a Google Sheet or a database table, alongside the run history in n8n or Make.com.
You decide how long entries are kept, so retention matches your privacy policy. You can export the log at any time.
Questions about the process
How long does it take to get an AI agent live?
You see a working prototype within 5 business days of us having your sample data and test access. Once you approve the fixed price, the production agent is typically live in about 2 weeks. Agents that need new hardware, many connected systems or complex approval rules take longer, and the scope says so up front.
Can we start with a person approving everything?
Yes, and we recommend it. Once the log shows the agent getting a type of action right consistently, you can let that action run on its own and keep approvals on the rest.
Does anyone at our company need to be technical?
No. You need someone who knows the job well. If you have an IT person or provider, we work with them on accounts and access. If you don't, we walk you through it on a call.
What if we want changes after go-live?
During the 2 weeks of included support, we fix anything that doesn't work as scoped. New features after that are covered by monthly support if they're small, or given their own fixed price if they aren't.
AI made building about 5x faster. Get the savings.
Book a free 30-minute call. If there is a good first build, I will make you a working prototype in 5 business days, free, with no obligation.
Get a Free Working Prototype