Tune Claude for one person. Or optimize it for the team.

You are thirty messages deep into a Claude chat. You have been building context for an hour. The material you are creating is mind-blowing. You send your next message for finetuning to Claude. The response is slow. Then it gets truncated and the chat stalls.

You start a new chat. You re-upload the files. You re-paste the tone of voice. You re-explain the project. The document survived. The thinking did not.

If you use Claude a lot, you have lived this.

There is a whole genre of advice going around online on how to stop this from happening.

It is good advice. Most of it works. I use it myself.

The short version of how to prevent this:
- Convert your PDFs to markdown before you upload them to Claude
- Do not dump big docs and input material into a chat.
- Set up a Claude Projects and reuse them - to store material in and to be able to go to the next chat and just say ‘pick up from where you left off just now”
- Edit your last message instead of replying to it.
- When a chat starts sagging, restart. Do not drag a dead thread through a new topic.

Each of these saves tokens. Each one stretches your chat a little further before the wall.

Go and use them. If you are one person running Claude to make your own work faster, they will change your week. 

That is a cluster of tactics to cater for a few things

‍

That is a cluster of tactics to cater for a few things

‍
1. Making sure Claude works from the right context, that is pretty much fixed 

2. Making sure Claude works from the right instructions, that are the same pretty much always

3. Making sure Claude is not pulling the understanding of 1 and 2  into the chat, and rereading that with every interaction, burning your tokens and your chat space

‍

Who is doing that work, for whom?

If the answer is "me, for me," go with the above list. The advice is written for you. If the answer is "me, for a team of five or twenty, all of whom also want to use Claude well, all of whom are living their own version of the Claude crunch," the above list is not solving your joint problem.

These and similar ‘Claude tips’ you find online are written for individuals, not companies. 

A company running AI well does not look like twenty well-tuned Claude accounts. Each re-explains the company to Claude a little differently.

If twenty people would produce work that way with Claude, knowledge that makes your company your company (your tone, your positioning, your research, your positions, what you want prospects to take away from a conversation) 1) lives in twenty private Claude instances, or 2) lives in Claude projects.

The problem with 1) is it is unmanageable, uncontrollable and will create different outputs across the company and with 2) it is locked into Claude, and can't use it elsewhere in tools or automations. 

‍

The second problem 

Projects, Preferences, Styles, Memory, all of these described in tip lists like above, all of it is Claude-proprietary. Every hour you spend tuning those is an hour bolted to one platform.

The day you want to try a different model, the day your team rolls out Gemini for something specific, the day the next generation of models lands, you are building your own AI sales dashboard, or using a new tool, you start over. Individual productivity has a short shelf life this way.

The thing that actually defines how your company works does not change with the model. Your ICP is the same when Anthropic ships a new context window. Your tone of voice is the same when Google launches something new. Your positioning is not waiting on Sonnet 5.

That context is yours. It should live in a place that is yours.

‍

Here is what a prompt looks like when it does

Using the {Company Name} Blog Structure Guide in the {Company Name} Brain, write a blog post about [topic].

Apply the {Company Name} Tone of Voice Instructions throughout.

Source material: [conference quotes/data/etc > describe,attach,in library, or in the Brain].

Content direction: [draft headline/description].

‍

Four lines. No attached tone of voice PDF. No re-pasted positioning. No thirty-page research report dragged along. The company's thinking is already in a place Claude can reach. The prompt tells it which parts to use.

‍

The same request without a Brain is a different kind of message. Several pages of tone instructions attached, because Claude has no standing access to them. The output structure explained, to make sure it looks like what we have in mind.
Last year's research is attached as a thirty-page PDF, because Claude needs it. The company's positioning re-stated for the hundredth time, because the context does not persist.

That message is two thousand tokens before the actual ask. It eats your chat capacity on the way in.

Same output request. Wildly different input. The difference is not prompt craft, but where your company's thinking is stored.

Three layers, roughly

What lives once. The things that define your company and do not change much. Tone, positioning, research you will use for a year, takeaways, positions, offering, ICP. This lives in the Brain or a database you have set up. 

What is alive this quarter. The campaign you are running, the pitch you are writing, the report in progress. This lives in a project.

What is a one-off. A specific document Claude needs to see once. This is an attachment - ideally in MD. Keep it minimal.

Most teams have no clean version of the first two layers. They push everything into the third, and then see their chats melt away.

The part that matters most in a long-life business: The Brain is yours. Not Claude's.

You own the layer. You decide what lives in it. You can plug it into Claude today, into a different model tomorrow, and a new tool next year, into whatever is sensible the year after. Your company's thinking does not have to move when the model does.

Every hour you spend on Projects, Preferences, and CLAUDE.md, you are investing in Claude. Every hour you spend on your Brain, your knowledge, and SKILLS, you invest in your company’s ability to use the compute it needs without locking-in or daily frustration. 

POV

The Claude productivity tips going around right now solve a real problem. They are good advice. 

They are just not solving your problem, if you are running a company that uses AI seriously.

Individual tips optimize one person.

Company thinking optimizes everyone. And it is the only version that compounds over time.

‍

‍