Source: https://learn.nodegx.io/build/one-job-per-conversation
Summary: Why to start a fresh conversation with Claude for each job, and how to pick up your project again.

# One job per conversation

Start a new conversation with Claude for each job. Claude does its best work in a short conversation, and NodeGX keeps what it needs to pick up where you left off.

## What counts as one job

One job is one thing you could check when it's done. "Build the booking page" is one job. "Fix the empty task list" is one job. "Build the whole app" is too big.

A quick follow-up can stay in the same conversation. If Claude has just built a page and you want the button bigger, just ask.

## Why long conversations go wrong

Everything in a conversation stays in front of Claude: every message, everything it read, every fix that didn't work. As that pile grows, the job you care about gets less of Claude's attention.

You'll notice it when Claude:

- goes back on something you both agreed earlier;
- asks a question you've already answered;
- tries an idea you already turned down;
- gives vaguer answers.

Long conversations also cost more. Every new reply carries the whole conversation with it, so a long one uses up your plan's limits faster. See [Plans and limits](https://learn.nodegx.io/setup/plans-and-limits).

When a conversation gets very long, Claude shortens the older part into a summary so it can carry on. A summary always loses detail, and you don't get to choose which. Keep conversations short and you rarely get there.

## Start a new one when

- **The job is done.** Start fresh for the next job, even if it looks related.
- **You spot something else that needs fixing.** Ask Claude to write it down, finish the job you're on, then do the new thing in its own conversation.
- **You're going round in circles.** See [When something goes wrong](https://learn.nodegx.io/build/when-it-breaks).
- **The answers are getting worse.** More messages won't fix a muddled conversation.

## How to start the next one

When you plan an app with Claude, NodeGX writes that plan into the project folder, along with a brief that says what the app is, who it's for and what it will not do. They stay there after the conversation ends. So a new conversation doesn't start from nothing: Claude opens the project and reads them.

Start a new conversation in Claude, then send:

```text
Please open my NodeGX project "[project name]" and read its brief and plan.
Today I want to [the one job for this conversation].
```

If you don't remember the project's name, ask Claude to list your NodeGX projects first. It sees the same list as **Recent projects** in NodeGX: the projects NodeGX has opened on this computer.

If you want to carry something over from the last conversation, ask for a note before you leave it:

```text
We're stopping here. Write me a short note: what's done, what's left,
and anything the next conversation should know. I'll paste it in there.
```

## One project per conversation

This is also how the NodeGX connection works. Once Claude has opened or created a project in a conversation, it stays on that project. To work on a different app, start a new conversation.

## Ask Claude to remember what you like

Claude Code can keep short notes for itself that carry over into later conversations. When you correct it about how you like things, ask it to note that down:

```text
When you learn something about how I like things done, remember it
for next time. Keep each note short.
```

For what you know and don't know, use the profile in NodeGX instead. See [Tell Claude who you are](https://learn.nodegx.io/setup/about-you).

## Quick reference

- One job, one conversation. Quick follow-ups can stay.
- Long conversations drift and use up your limits faster.
- To pick up again: ask Claude to open your NodeGX project and read its brief and plan.
- A different app means a new conversation.
- Ask Claude to remember how you like things done.