Built to order

Custom AI Agents

One agent, one repeatable process, measured against criteria agreed in writing.

Book a call

What it produces

From: Dev Patel, Bramble & Co <sales@example.com>
To: lena@example.com
Subject: Your quote: 40 cartons of 12 oz cups

Hi Lena, your quote for 40 cartons is attached: total 1,585.00 ex tax, delivered next week.

Illustrative sample data

Today
Senior people reading, researching and drafting by hand.
With it
The agent drafts. Your team approves what matters.

Watch it run

A quote request at 9:14 am

Step 1 of 7

From: Lena Ortiz, Cedar Street Café <lena@example.com>
To: sales@example.com
Subject: Quote please: 40 cartons of 12 oz cups

Hi team, could you quote 40 cartons of the 12 oz compostable cups, delivered next week?

Thanks, Lena

A new email in the sales inbox starts the agent.

CRM and ordersRead-only

Context: Cedar Street Café

Order	Item	Cartons
SO-8812, 14 Aug	12 oz compostable cups	30
SO-8740, 17 Jul	12 oz compostable cups	30
SO-8655, 12 Jun	8 oz compostable cups	20

Noted

Trade tier B
64 cartons in stock

It reads the account, recent orders and the current price list, and nothing more.

QuoteDraft

Draft quote Q-2291

Item	Qty	Unit	Line
12 oz compostable cups	40	38.50	1,540.00
Metro delivery	1	45.00	45.00
Total ex tax			1,585.00

Prices come from the client's own list for the customer's tier. The agent never invents a price.

LogComplete

Run log: quote-agent

09:14:02 trigger email.received sales@
09:14:09 context crm.account, orders x3, price_list
09:14:21 draft quote Q-2291 total 1,585.00
09:14:23 checks 4 run, 0 failed
09:26:40 approval dev.patel approved
09:26:42 delivered email.sent lena@example.com

Run complete. Inputs and outputs kept for review.

Every run keeps its inputs, outputs and decisions, for review and the monthly report.

Bramble & Co and its customer Lena Ortiz are fictional.Illustrative sample data

Why it matters

40%+

Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls. Source

Control and data

A person decides

Anything that sends, pays or deletes waits for a person.

Your data stays yours

Models see only what each task needs, and never train on it.

Connects to

In your accounts

Workspace
- Microsoft 365
- Google Workspace
- Notion
Records
- HubSpot
- Salesforce
- Pipedrive
Data
- Airtable
- PostgreSQL
Approvals
- Slack
- Microsoft Teams

Named for reference. No endorsement implied.

In full

Every agent has named approval steps before it sends, pays, deletes or changes anything that matters. Before it goes live, it runs alongside your existing process for a week so your team can compare its work with their own.

By default it runs in accounts and infrastructure in your name. If you would rather we run it, we host it in accounts set up only for you, kept separate from every other client. The agent sends an AI model only what each task needs, through business or API accounts that do not train on your data, and the model provider never gets standing access to your systems. How we handle data

Questions

Open one

Which AI models do you use?

Usually Claude, from Anthropic directly or through Amazon Bedrock in Australia, and sometimes OpenAI models, depending on what fits the task. The agents we build use business or API accounts only.

How do we know it works before we rely on it?

We agree success criteria in writing before the build, test the agent against real examples from your business, then run it in parallel with your team for a week before it takes over.

What happens when the agent is unsure?

It stops and asks. Cases below the confidence threshold we agree, or outside the process it was built for, go to a person with the agent's notes attached.

In detail

Open one

What it does

Takes one defined process from trigger to finished output, inside boundaries you set.
Works in your own tools through their APIs: CRM, inbox, shared drives and databases.
Logs every run with its inputs, outputs and decisions, so you can see what it did and why.
Sends low-confidence or high-stakes cases to a person instead of guessing.
Is measured against written success criteria, tested on real examples from your business.

How each run works

Every run follows these 5 stages, inside your own tools.

01Something triggers the agent. A new email, form submission, CRM record or scheduled run starts the process.
02It gathers the context. The agent reads the record, the related documents and the history it needs from your systems, and nothing more.
03It does the work. It drafts, classifies, scores or extracts, following the instructions and examples agreed in scoping.
04A person approves what matters. Anything that sends, pays, deletes or changes important records waits in an approval queue for your team.
05The result is delivered and logged. Approved output lands in the right system, and the run is recorded for review and monthly reporting.

Spec sheet

Function: Operations
Status: Scoped with you and built for your business, using the same delivery process as every engagement.
Connects to: Microsoft 365, Google Workspace, HubSpot, Salesforce, Pipedrive, Notion, Slack, Microsoft Teams, Airtable, PostgreSQL
Built for: Recruitment
Professional Services
Manufacturing and Distribution
Healthcare
Legal
Accounting and Finance
Research: Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls. Gartner press release, June 2025 (opens in a new tab)

Name the process that eats your week.

We will scope one agent around it, with success criteria in writing.