This week we're sharing a couple practical things you can try that make your AI tools work way better than they already do. 

Let’s get into it.

Russell:

At the end of every day, I often find myself hesitant to close out of my sessions with Claude Code because I want to make sure that I can pick up tomorrow where I left off. Sometimes I’ll even ask Claude to write summary documents about things that I’ve been working on, and then I tell Claude to read back that document the next day so my work isn’t lost. 

One of the best things about AI tooling is that it’s now super easy to build customized workflows to solve small pain points like this, so this week I decided to formalize this process.

I built a new skill called “wrap.” It’s a skill that I use in all my projects at the end of each day, where I close out by telling Claude “let’s /wrap.” And it creates a summary document about everything that I did, so that in the morning, Claude can just pick up where I left off.

The other great thing is that it creates a continuous log of the work that I’ve done, so I can now also easily reference work that I was doing a few days ago and Claude can go and find it. The wrap skill combined with a project management tool like Linear (which I wrote about last week!) allows me to have really solid documentation around my work so that nothing gets lost and it’s super easy to stop or resume at any time.

Julia:

This week I'm going to share a simple unlock. I was deep in Claude Code working on a project that required a ton of reference documents, and after feeding in a handful of PDFs, Claude started losing the plot. Forgetting things, weaker answers, just getting dumber the more I gave it. In talking to friends, family, and clients, a lot of people are using AI to process documents faster and running into this same issue without knowing why.

Here's what's happening: when you feed a PDF into ChatGPT or Claude, every page gets rendered as an image alongside the text. That visual processing eats way more memory than the text alone, and it adds up fast. Every conversation has a memory ceiling and you're blowing through it with images.

Hitting the (context) wall

The fix is simple. If the PDF has selectable text (which most do) you can convert it to plain text first and the AI reads it just as accurately for a fraction of the cost. In Claude, just say "convert this PDF to text before analyzing it" and it handles it.

Once your documents are cheap to load in, you unlock the thing AI is actually best at:  cross-document thinking. "Compare these three contracts and tell me which one has the most favorable terms." "Across all these meeting notes, find every deadline I committed to." That kind of work is tedious for us but trivial for AI when you've given it a clean format to work with.

Next time, before you hand AI a pile of documents, strip them down first. There will be more memory left for the actual thinking.

Stay curious,

Julia & Russell

Reply

Avatar

or to participate

Keep Reading