|
The AI Solopreneur
Newsletter | AI Context
|
Stop feeding your AI the same story every day
You ask a question. You get a flat answer back. Most people blame the model at that point. I blame the setup.
A model with nothing to work from produces the most average answer it can find. Give it your background, your standards, your history with a client, and the same tool writes like it has been sitting in your business for a year. The intelligence did not change. What you handed it did.
Two people can type the exact same prompt into the exact same tool and walk away with two different outcomes. One of them loaded context first. The other one asked cold and hoped.
|
200K
Tokens of standing context Claude can hold inside one project. A style guide, a full client history, a year of notes, all read before you type a single question.
|
Why re-typing the same background kills momentum
A normal chat remembers itself just fine. What it does not have is any memory from the chat before it. So every new conversation starts blank, and you paste the same background in again. Then again the next day. Somewhere around the fifth time, most people stop bothering and decide the tool was not that useful in the first place. The tool was fine. Nothing was carrying the context between chats.
Claude Projects is a workspace built to hold that context instead of losing it. Upload your files once. Every conversation that follows inside that workspace already knows them. No re-explaining, no generic first draft you then have to fix by hand.
How I actually run mine
Four projects cover almost everything I touch in a week. Each one holds only what belongs to it, so the answers inside stay specific instead of average.
1 / A workspace for every active client
Each engagement gets its own project, loaded with their systems notes, past proposals, and where the work actually stands. Connect Fireflies or Granola, then ask Claude to pull in the meetings with that client, and every call we have had sits right next to the rest of the record. A question mid week gets answered from the real record, not from a memory I am reconstructing on the spot.
2 / One workspace holding the entire brand voice
Style rules, banned words, tone, past posts that worked. A script or a newsletter draft starts already sounding right, instead of me correcting a generic first pass every single time.
3 / A community project that remembers who is who
Event notes, live conversations, the shape of what the group is building toward. Useful the moment I need to draft something or brief someone without re-telling the backstory from the start.
4 / A long personal writing project
A memoir in progress lives in its own workspace, chapters, timeline, and character notes included. The continuity holds across sessions. I do not spend the first ten minutes reminding it where the story left off.
|
Why this matters to you
Check what your AI had to work with before you judge the answer it gave you. Give it something real, once, and stop repeating yourself for good.
|
|
Tool of the week
Meetgeek doesn't stop at taking notes. Connect it to Claude directly, and Claude can pull your meeting transcripts, summaries, and action items straight into a conversation. Your meeting history becomes context Claude reasons over, not a file you dig through by hand.
get.meetgeek.ai/jt8ry3p73t01
|
Pick the one thing you ask about most. Build one project for it this week. Write down what you would want a new hire to know before their first day, and load that in. Then judge the second conversation, not the first.
Talk soon,
Mikkel
The AI Solopreneur • AI Strategy for European SMBs
|