This is one of the lowest effort, highest impact AI habits I have at work.
This is one of the lowest effort, highest impact AI habits I have at work. Most of my work at nettle has been geared towards sales, marketing, and a bit of ops. I use Claude as my workbench – it'd kill me if, every time I opened a new chat, I had to explain the nuances of our ICP, quirks of the sales funnel, or how our CRM deal board is structured. It has the Memory function, sure, but most of my experience with it (though that was with my now deceased ChatGPT) was me asking Chat to recall something and finding out it had deleted it. The joys!
So, in my first week at nettle, I spent roughly one hour building a "Sales Thinking Partner" as a Claude Project.
Essentially, it pre-loads all the context I need when starting any conversation from that Project – our ICP filtering criteria, messaging angles and tone, customer success numbers I can leverage, etc.
The steps were:
1. Take the context that was already pre-produced for me (sales onboarding doc + main sales deck).
2. Create new context, like why our sales funnel is structured in a certain way, what the attributes in our CRM are + their API slugs. This is what took the longest, despite our MCP connector to the CRM helping a lot to retrieve some of this info.
3. Feed all of that into Claude itself, explaining what I wanted, and having it produce structured Markdown files that were optimal for it to read.
4. Add everything as text content in the Project, and, once again, use Claude itself to write optimal Project instructions based on what I wanted to use it for.
In the seven weeks since joining, I have saved a couple of hours by not having to set the conversation context every time + having consistent output from my Claude. Every time I use that Project, it simply knows what I’m talking about.
In the one week our Thinking Partner was out of date because of a new addition to our product suite, I had to manually provide that context every time I interacted with it. That was so annoying that I spent half an hour last week adjusting all my Projects that needed it. Worth every second.
This is just one Project; I’ve others for creating marketing copy, building branded presentations (though I still find Claude 💩 for this), and more.
And for those using Cowork, you can essentially do the same context pre-loading by saving the Markdown files in the working folder/subfolder you point Cowork at, and then adding a line to your Global Instructions telling Claude to read those context files at the start of every session.
AI works best if it has structured context. Less than one hour of set-up creates huge leverage later.