Why we think an agency should be built around its agent, not inside a platform

This is a note from the Volant team: our view and our opinions, written at the end of December 2025.

The end of the year is a good time to write a decision down, so it can be checked later. Ours is this: a one-person agency works best when it is built around its AI agent, not inside a platform. That is also why Volant is built the way it is. Here's the reasoning, as plainly as we can put it.

What problem are we actually trying to fix?

Not the websites. Getting an AI agent to build a good website is no longer the hard part. Give Claude Code a clear brief and a folder, and it will write the pages, fix what you point out, and check its own work when you ask.

The hard part is everything around the website. Picture a typical week for someone working alone. Which businesses are worth approaching? What was promised to the florist about her Mother's Day page? Has the dentist paid the deposit? Is the booking form on the physio's site still sending? Those are the questions a one-person business answers every day, usually by searching an inbox, a few chats, a spreadsheet and memory.

If you've read our earlier piece on what a one-person AI agency is, this is the same point from the inside. The agent takes on much of the making. The running of the business stays with the person, and it becomes most of the job.

Why not do it all inside a platform?

Agency platforms are easy to understand, and it's clear why people use them. They put a lot in one login: websites, calendars, texts, email campaigns, reviews. If your business is selling those marketing services every month to many clients, that bundle makes sense.

For a lot of people starting out on their own, the situation is different in three ways.

First, the making already happens somewhere else. An agent like Claude Code builds sites in ordinary files on your own computer. A platform with its own page builder asks you to move that work inside it, and then the site lives on the platform's terms.

Second, the client's things should stay the client's. Their domain, their email, their listing, their payment account: in their name, from day one. If you stop working with a client, or a client stops working with you, nothing should be trapped.

Third, your own notes on the business should stay yours. Who you pitched, what you agreed, what each client has asked for. That's the memory of the business. It shouldn't depend on a subscription you might one day need to cancel.

None of that makes platforms wrong. It makes them the wrong shape for this kind of business, and it's why Volant keeps those records with the owner instead of on someone else's service.

What does "around the agent" mean in practice?

It means the agent is the thing that does the work, and everything else is arranged so the agent can work well and you stay in charge.

Two developments made this feel possible rather than hopeful. In November 2024 Anthropic released MCP, a shared way to connect an AI program to other programs and files. That meant an agent could read your notes on a client instead of whatever you last pasted into a chat. Then in October 2025 Claude Code added plugins, a way to package and share the commands, helpers and connections you've set up. An agent you can connect and extend is an agent you can build a working routine around.

So the shape we recommend is simple to describe:

• The agent does the making. Research, drafts, builds, checks.
• The records live in one place the owner controls. One home per client for the facts, the agreement, the accounts, the money and the history. We wrote about the client records for freelancers that we think matter most.
• Nothing goes out without you. Every message, quote and published page waits until you've read that exact version. Approving what your agent sends isn't a feature added at the end. It's the rule the rest hangs from.

This is the whole idea behind Volant in three lines. Your agent keeps doing the work it's good at. Volant keeps the clients, promises and money in order around it, and asks you before anything leaves.

What worries us about this approach?

Plenty, and we'd rather say so.

An agent is only as good as what it's told and what it can see. If the notes on a client are wrong, its drafts are confidently wrong. That's why the facts need checking by a person before anything uses them.

An agent that can use tools can use them badly. Connecting it to real accounts and real clients means being careful about what it can reach. It also means being honest: the tools an agent uses on its own are its responsibility and yours, not something a separate app can fully watch over.

And the business side is still work. An agent doesn't find clients for you or make them pay. Anyone who tells you otherwise is selling something.

Who is this for?

For one person doing client work, who can already get good results from an agent and wants the rest of the business to hold together. Designers, developers, marketers, and people starting a one-person AI agency, sometimes next to a day job.

If that sounds like you, the useful question isn't which tool builds the prettiest page. It's where the clients, the promises and the money will live, and who controls them.

That's the bet we're writing down at the end of 2025. The agent does the making. The owner keeps the decisions, the promises and the money, in a place they control. If you want the longer version of why, it's on the page about why Volant exists. We'll read this note again next December and see how it held up.