Who owns what your agent factory learns about your firm

The most valuable part of an agent loop in a small firm is not the model. It is what the loop learns about the firm. If that knowledge sits with a vendor, in the vendor’s format and on the vendor’s terms, you do not have a system. You are renting one, along with your own experience.

In the first piece in this series we argued that agent loops will reach a service firm once its work is written down. This one is about what happens to that record afterwards.

What the loop actually learns

After a year of work, a loop built around jobs knows things no price list contains. How long a boiler swap really takes in a tenement and how long in a detached house. Which customers pay after the second reminder and which after the first. What tone the owner takes with a customer who has been waiting three weeks for a date. Which draft quotes he corrected, and what exactly he changed.

That is the most accurate description of how your firm works that has ever existed. More accurate than the technician’s notebook or the owner’s memory, because it does not forget and does not go on holiday.

A small firm rarely asks where that description lives. It buys a subscription because the invoicing program grew an assistant, and the scheduling program grew a different one. Each learns on its own, in its own cloud, and neither knows the other exists.

What you own and what you rent

An agent loop is not one program. It is a stack of layers, and each can be yours or the vendor’s.

Layer	What it is in a small firm	What you lose if the vendor owns it	The minimum to keep
The flow of work	the steps from enquiry to paid invoice, and who approves which	you cannot rebuild it in another tool	a flow written down so it can be read without the vendor’s dashboard
Rules and templates	how you price, how you write to customers, when you chase	experience gathered from hundreds of corrections	rules in a file or database you can export
Access	the inbox, the calendar, the accounting program	control over who reads your customers’ messages	one list of access that you manage
History	every draft, correction, approval and rejection	the firm’s memory of what worked	history in your own database
Evaluation	how many drafts went out unchanged, days to invoice	the ability to compare tools on your own jobs	your own numbers, from your own data
The model	which language model does the work	freedom to switch when price or quality changes	a model you can swap without rewriting the rules
Where it runs	whose server runs it	a say in where customer data goes	a choice: your machine, your own cloud account, or a provider you can leave

The model sits near the bottom on purpose. Models get cheaper and change every few months. Rules and history gain value every week the loop runs. If you could take only one layer with you, take the history and the rules.

Three roads and what each costs

A ready-made assistant on subscription. The fastest start. The price is that the rules, the history and the evaluation live with the vendor, and customisation ends where the settings page ends. When the vendor raises the price or removes a feature, a small firm accepts it by email, because it has no leverage. How that model prices a five-person firm is the subject of Per-seat pricing was designed for companies with seats.

Wiring it together yourself. An automation tool, a few connections, a model call here and there. The first result arrives after a weekend. The price is that six months later nobody except the person who built it knows what connects to what, and that person also has a firm to run.

Open components on your own infrastructure. The flow, the rules and the history live in your database, on your machine or your own cloud account. The model can be swapped. The price is that somebody has to be responsible for that infrastructure.

BiznesIQ took the third road: it is built on sovereign open source, and the data, rules and history sit in a database the firm controls. Not out of ideology. The point is that a vendor’s decision should not become your firm’s decision.

Where sovereignty becomes an expensive hobby

Sovereignty can be overdone, and in a small firm it is especially easy.

A five-person firm does not need a server in a cupboard for its data to be its own. A machine that nobody patches or backs up is a worse home for customer history than a decent provider. If nobody in the firm is comfortable running one, the honest answer is your own cloud account or a managed option, provided the export is complete and leaving stays possible. For BiznesIQ such an option is planned on exactly that condition.

The second trap is running everything on self-hosted models because it sounds independent. On some tasks they will be weaker. What makes sense is something else: being able to pick a model per task and change it without rewriting the rules.

How to measure whether the loop is yours

Sovereignty is not measured by a statement on the vendor’s website. It is measured by the cost of leaving.

Vanity metric	Outcome metric
“your data is safe” in the terms of service	how many days it would take to move jobs, rules and history to another tool
the number of integrations on the pricing page	whether the export holds drafts, corrections and reasons, or only rows
“everything is in the cloud”	whether you can read last quarter’s history without logging in to the vendor’s account

One test

Ask yourself, or your vendor: if we stopped paying tomorrow, what would we have to build again? If the answer is “the pricing rules, the message templates and the whole job history”, then your loop, however young, already belongs to somebody else.

How BiznesIQ keeps data in the firm’s own database, and what is still only a plan, is set out on the home page.

The choice is easy to name. You can build the loop quickly on someone else’s infrastructure and accept that what your firm knows about its own work will sit with the vendor. Or you can build it a little more slowly and keep that knowledge at home. Both have a price. The worse outcome is making neither choice on purpose.