This week
I was workshopping website copy when Claude pushed back on an exclamation point. Said it didn't match my voice. I'd never written "don't use exclamation points" anywhere in its instructions. It inferred that from the style guide I'd referenced.
That's not how Claude usually behaves. I've had a style guide for a while but it always felt like the AI-isms won out over my style rules. The difference was a rewrite of my CLAUDE.md file: the persistent instructions Claude Code reads at the start of every conversation.
I'd heard someone describe writing this file as a job description for your AI agent and decided to test it. Four sections: who I am, how we work together, how the workspace is structured, what tools are available.
After the rewrite, working with Claude felt much more like working with a colleague.
I made a tutorial on how to give your Claude its own job description... and why less context often outperforms more.
What I'm reading
Andrew Ng on AI job anxiety — Andrew Ng / The Batch, deeplearning.ai. Andrew Ng writes about something I hear more and more: job insecurity at every level. He acknowledges the anxiety is real: even AI lab insiders privately admit uncertainty, despite public confidence and hype. Everyone seems to have a voice inside them asking: What is there to hold onto now? His answer: community and skills. Both stay portable when everything else shifts.
This is how the Every editorial team uses AI — Kate Lee / Every. Kate Lee runs the Every editorial team and writes a very honest piece about what AI-native writing actually feels like in practice. She compares AI-native writing to sculpture: "You get a lump of material and shape it into what you want it to become." I can relate to this workflow. I got Claude to take a first stab at this newsletter draft, and I've spent the last hour or so chiseling away at the finer details. I'm not sure that it saves TIME, but it's fun and it does shift my focus from broad strokes to details.
The SUCCES framework — Chip and Dan Heath / Made to Stick (book). The universe conspired to get me to finally pick up this book. It's been patiently waiting on my Kindle ever since I agreed with my former boss 4 years ago to set a quarterly goal to improve my async communication. In it, they outline the SUCCES method, a framework for creating memorable ideas: Simple, Unexpected, Concrete, Credible, Emotional, Stories. The obvious application is to ask AI to review a text with these principles in mind. But I'm also curious about applying the framework to your instructions to the AI. Will it improve its performance... and your working relationship? Does AI fall victim to the same sticky techniques as humans? I intend to find out.
Slow note
I drank my coffee outside this morning. It's still a bit cold here in the north. We've had a few warm days, and this one hadn't caught up yet. Nice, actually, to feel it for a few minutes before the day started. Oh and by the way I left my phone inside.
Although I thoroughly enjoyed my screen-free time outside, I've never resonated with the "don't look at your screen until 10am" advice. I do yoga in the mornings using a YouTube video. My to-do list is on my phone. Almost every morning I have to check the list of extra spare clothes or empty jam jars that I have to bring that day to my daughter's daycare. The rule, as stated, doesn't fit my life.
But for the past few months I've had a different one: no social media or AI before 10am. Instagram, LinkedIn are both off until then. Claude can wait. A more intentional morning than any blanket rule ever gave me.
Until next week,
Louise