The reminder sequence has run for four months. The owner stopped thinking about it in the second week, which was the point. Then the vendor changes the plan. The tier the firm signed up for is gone. The credit meter resets lower. The template the firm customized now sits behind an upgrade, and the setting that kept client names out of the vendor's training data has moved to a different page.
Nothing is broken, exactly. But the job the firm depends on now runs on someone else's decisions. That is the question this article is about: who controls the job once it is automated. Our answer, and how we build to it, is on the First Agent Sprint page.
Platform risk is not a website problem
I made the plainest version of this argument for websites a while back, in Understanding and Mitigating Website Platform Risk:
"Platform risk occurs when your business relies on a third-party service provider… these platforms aren't the property of your business, and you could have your branded profile and audience ripped away without warning or recourse."
My point then was that the risk does not only come from the usual suspects. A business can do everything right and still find that something it depends on can be changed, priced, or switched off by someone it does not control. As I put it: "Using software that can have critical functionality disabled by one person without warning is as significant a platform risk as using a Facebook account as your only online presence."
Replace "website" with "the job that answers your after-hours inquiries" and the argument does not change. It gets sharper. A website that goes stale costs you visitors. An agent that stops sending the reminder, or starts sending a different one, costs you cash and trust before you notice.
What renting actually means
Most "AI agents for small business" products rent you a seat and leave the setup to you. That model is honest about what it is. It is also built around the vendor's interests, which are not the same as yours.
Renting means the vendor decides the price next year. It means the logic of your job, the rules about who gets a reminder and who gets a person, lives inside a product you do not control. It means the log of what was sent sits in their system, on their retention terms. And it means the people who know how your agent works are a support queue, not two people you can call.
None of that is sinister. It is the shape of a rental. For a task you could drop tomorrow, it is often the right shape. For the job that lands on the owner every night, it is a dependency you did not mean to take on.
The terms are part of the build
There is a second reason ownership matters, and it is the one a CPA or a lawyer will raise first.
Generative tools differ in what they keep. Some are set so that prompts are not used to train a model. Some are not. Some change their terms after you sign up. A firm that has written down exactly what an agent may read can still undo that work by running the agent inside a tool whose terms nobody at the firm has read. The access list and the vendor terms are the same decision, a point we make at length in What the Agent Is Allowed to Touch.
When the agent is yours, that decision stays with you. You know which tools it runs on, because they are the tools you already pay for. You know what it can read, because you approved the list.
What owning the agent means
Every Berry Buckley agent belongs to the client. In practice, we build toward that in a few ways.
It runs on your tools. QuickBooks, HubSpot, Google Workspace, Square, Squarespace: whatever you already pay for. We do not move the job into a new system you have to live in, and we do not rip anything out.
The rules are written down, and they are yours. Every agent ships with an owner playbook: what it can read, what it can send, and what it must never touch. You approve it before the agent runs on its own. It is written in plain language, so a partner, a client, or a successor can read it.
You can see what it did. The log shows what went out, when, and why. You can turn the sending off without turning the reading off.
The price is not a meter. A First Agent Sprint starts at $7,500, fixed fee. Care plans start at $750/month for firms that want us to keep the agent current. The audit is free, and every engagement is a project fee. The fee pays for the build. It does not rent you access to your own work.
Here is what that means in practice. You own the agent's configuration, its prompts, the written playbook, and the data it works with, and everything runs on accounts in your firm's name, not ours. If you end the care plan, the agent stays yours and keeps running. You just stop getting monitoring, tuning, and fixes from us. The care plan itself covers monitoring, updates when your tools change, small adjustments, and a periodic review.
What ownership does not mean
Owning the agent does not mean you have to run it alone. It does not mean a code repository dropped in your inbox with a goodbye note. And it does not mean we promise the tools underneath will never change. QuickBooks will update. Google will move a setting. That is what the care plan is for.
It also does not make a bad job a good one. If the first job only works with access to the client file, owning it does not fix the risk. It moves the risk to you. The first job should be narrow enough to explain on one page.
The question to ask any AI vendor
Before you sign anything, ask one question. If we stop paying you, what do we still have?
If the answer is "an export," ask what is in it. If the answer is "nothing, but you would not want to leave," treat that as the answer. A firm's dependence on a vendor should be a choice the firm made, not a surprise it finds out about on renewal day.
The closing question I asked about websites works here too: how do you plan to avoid platform risk? For a first agent, our answer is simple. Build it on the tools you already own, write the rules down, and hand the firm an agent that belongs to it.
Questions owners ask
What does it mean that the client owns the agent? The agent is built for your firm, on your tools, under rules you approved, and it belongs to you. You own its configuration, prompts, written playbook, and data, on accounts in your firm's name, and if you end the care plan it keeps running without our monitoring, tuning, and fixes.
Is a rented AI seat ever the right choice? Yes, for tasks you could drop without consequence. For the job the firm depends on every night, the risk is higher than the price suggests.
What happens if the tools underneath change? They will. A care plan, starting at $750/month, keeps the agent current as your tools and workload change.
Do we need a new platform for the agent to run on? No. It runs on the tools you already pay for.
Why do vendor terms matter if the agent only reads the invoice list? Because what the tool keeps is part of what the agent can expose. The access list and the vendor terms are the same decision.
If this is the job still landing on you, start with a free 45-minute First Agent Audit with my partner Paul Buckley. You will get a written brief within one business day: the workflow, your tools, the risks, and a recommended sprint, or a clear no.
Book a First Agent Audit · How the sprint works: /first-agent · Where firms usually start: /agents
Sources
Berry, Drate. "Understanding and Mitigating Website Platform Risk." Berry Buckley Insights.
Buckley, Paul. "What the Agent Is Allowed to Touch, and What It Must Never Do." Berry Buckley Insights, October 2026.