AI in practice
Feel locked-in by your AI tool? A context library sets you free!
There are so many AI solutions available and they develop so rapidly that you might feel overwhelmed in knowing which one to choose. On top of that, I notice many people feeling hesitant to switch to a new tool, because they feel that ChatGPT knows them so well. I have a super simple, smart tip for you today if you also feel locked-in to Gemini, ChatGPT, Copilot or Claude because you worked with it already a lot.
As you might have read in the previous Grok Bot and Buzz blogs, I test a lot of AI tools. Literally during one week, I went from Claude Code to setting up Buzz to setting up Grok Bot. And back to Claude Code again ;-)
What a context library actually is
The reason that I was able to switch and test so quickly without losing context is because I set up a so-called context library. That sounds fancy, but it is nothing more than a folder on your computer that contains all the relevant context that AI should have about you and your work so it can add value to you. Because the quality of the context you provide is BY FAR more important for the quality of the output than the models or tools you use.
In this folder you add things that are relevant to know. In my case: my book is in it, all the website code for Be Your Own CEO is in it via GitHub (which contains all the content, PDFs and transcripts of the videos, so super rich context), my zannavanderaa.com website, my vision, the memory of all the decisions we took while building, what I want to achieve with each project, the anonymous stories of the pioneers so it understands the emotional impact of the platform, any research I did for example on how kids learn about financial education, et cetera.
Treat it as onboarding a new colleague
You drop all relevant context into that folder and you point any new AI tool first to that folder and tell it to deeply understand all the info that is in that folder. It will not digest all that info immediately. So what I tend to do, instead of being annoyed when it doesn’t get things immediately, is treat it as if I’m onboarding a new colleague. Ask things like: what questions do you have about my driver analysis technique? Can you summarise my vision in 3 sentences? But also: what questions do you still have after reading all the context? Or what do you need from me to be the best [fill in role you want it to be] to me?
One of the best tips is: the more you treat AI as you would a human, a new colleague, a junior assistant, the more value you will get out of it.
Talk to your AI instead of typing at it
The second best tip when it comes to context: use dictation, talk to your AI. I have used Wispr Flow (wisprflow.ai) for almost a year now and it’s amazing. I talk to my AI and as such, you naturally give it way more context, and it saves you a huge amount of time to type all instructions or the dump of a meeting you had. Of course you can use the audio button, but I still feel it doesn’t work nearly as good as Wispr Flow.
Ask AI to keep the library up to date
The third tip for context is to ask AI to keep your Context Library up to date! That way, it adds any relevant output, memory, decisions, progress et cetera to your folder and your context never goes stale again. Now you are completely flexible to switch to other tools without losing the valuable things you worked on with your existing tools.
And why I returned to Claude Code
The final tip, and why I returned back to Claude Code again? There are so many options out there, the key is to focus on what you want to achieve, what do you want AI to do for you. In my case, as a solopreneur, I don’t need to work with others while I’m building. On top of that, I noticed that even though OpenClaw and Buzz were using my Anthropic account and thus their models, it was nowhere near the same quality as I got from Claude Code directly. Now I’m back in my happy, flexible, build and execute rhythm that energises me most.
