What to stop pasting into every AI agent session

For many people who use AI agents for client work, every session starts the same way. Open the notes. Find the client. Copy the business name, the tone, the decisions, the things not to do. Paste. Then type the actual task.

It feels like care. It's really a tax, and you pay it every time.

This is a short note on what to stop pasting, where it should live instead, and the one thing to keep for yourself.

Why pasting feels like the right thing

An agent starts each session knowing nothing about the last one. Claude Code's documentation says it plainly: each session begins with a fresh context. So if you want the agent to know that a client spells their name with a hyphen, or that prices never go on the site, you have to tell it again.

Pasting is the obvious way. It works, mostly. The trouble is the "mostly".

A pasted brief is only as current as the note it came from. Here's how it usually goes: something gets agreed on a call, the note doesn't get updated, and the agent works from the old version. Or you paste in a hurry and leave a section out. You find out when a draft comes back with a word the client asked you never to use.

None of that is the agent's fault. It does exactly what it's told, with what it's given. The gap is in the setup: the truth about each client lives in your head and in several notes, and you hand-carry pieces of it into each session.

We covered a similar slip in the week a promise got lost. Different job, same cause: something agreed out loud and never written in the one place you'd look.

What to stop pasting

Everything that doesn't change from one session to the next. That tends to be most of it.

• Who the client is. The business, its customers, where it trades.
• The exact facts. Name, address, phone, hours, service area, each with where it came from.
• How it sounds. A few plain rules, and a sentence the owner liked.
• Decisions. Each one dated, so you know which is current.
• What's out of scope. So the agent doesn't helpfully build an online shop nobody asked for.
• How you work. Your spelling, your review steps, what always needs your approval.

The client parts belong in a file in each project folder that Claude Code reads at the start of every session. The parts about how you work belong in one file in your home folder, so you write them once. We set out the full shape of that file in how to give your AI agent a client brief it won't forget.

Keep those files short. Claude Code's documentation says specific, concise instructions are followed more consistently than long ones. A short brief with exact facts and clear decisions is also easier to keep true.

What to still type

Very little: the task for today, in a sentence or two, and anything that changed since the files were last touched.

That second part is the discipline. When a client changes their mind on a call, update the file before you open a session. If you catch yourself typing a correction into the chat that you typed last week too, that's a line missing from the file. Add it there.

Sessions get shorter. More to the point, they stop depending on how carefully you copied things that morning.

The part files don't solve

A file in a project folder is good for the site. It's less good for everything around the site: who you're pitching, what each client bought, what they've asked for this week, whether they've paid. That information changes daily and belongs to the agency, not to any one project.

For that, the agent should read the current record directly, not a copy. That's why Volant is built the way it is. Your agent connects to it directly, so it reads the client's details, notes and decisions as they are now, with nothing pasted in. It works with Claude Code.

We've written before about why we'd build an agency around your agent rather than inside a platform. This is the same idea at a smaller scale. The agent does the work; the facts live in one place that you own; and you stop being the courier between them.

What to keep for yourself

Approval.

Taking the pasting out of your day makes it tempting to take yourself out of more of it. An agent that knows the client well can draft the email, the page, the invoice. It still shouldn't send any of them without you.

So the one line at the top of every brief stays the same: nothing goes to a client, their customers or a live site until you've approved it. Keep that as a habit, and approving what your agent sends explains how Volant handles it. Less copying frees up attention, and that attention is best spent reading what goes out.

If you're starting out

You don't need any of this on day one. With one or two clients, a note and a careful paste is fine.

The moment to change is the second time the agent makes a mistake you've already corrected. That's the sign the truth is living in the chat instead of in a file. Move it. If you're just getting going as a one-person agency, the page on starting a one-person AI agency covers the rest of the early setup, and why Volant exists says more about the questions it's meant to answer.

Nothing about this is clever. It's writing things down once, in the place the work happens, and keeping the last word for yourself.