
AI FOR HER Issue #29

Behind the Build: One founder's journey from basic prompts to AI that runs without me — Week 2 of 4
On May 28, I'm giving a live talk at Girl Foundry showing you how to step off the content treadmill with an AI agent that runs your content department for you.
This is a peek behind the build as I create it in real time.
Everyone wants AI agents to take content off their plate. But how do you actually do that?
You start with the foundation most people skip.
Before you automate anything, you need skills that actually work.
Skills are the building blocks of an agent. When your skills work the way you want, you get output you'd actually post instead of revising a caption six times or going back and forth with Claude for 45 minutes.
But if your skills are producing mediocre content and you build an agent on top of them? You're just scaling bad content faster.
Here’s how to make skills right the first time:
Stop telling Claude what to do. Start showing Claude what good looks like.
Most people build skills by giving instructions. "Write punchy hooks." "Sound conversational." "Be direct but warm."
Claude reads those and thinks "sure, I know what those words mean" and writes something that technically fits the description but has no soul.
Because instructions are conceptual, and that’s where nuance is lost.
The nuance lives in the examples.
When you show Claude what “good” looks like to you, and what “bad” looks like, it picks up on things you could never articulate: the rhythm, the specificity, the way you land on an unexpected word. The stuff that makes your writing yours.
Instructions tell Claude the rules. Examples show Claude your taste.
If you're getting mediocre output, don't add more instructions. Clean up your examples of “good” and “bad.”


Here are the two big mistakes I made with my first content skill:
Mistake 1: I made it too epic.
I wanted my skill to do everything. Write hooks, write captions, write video scripts, adapt for different platforms, match my voice, follow my frameworks. One skill to rule them all.
The result? Claude got confused. The output was inconsistent because I'd given it too many jobs. So I spent most of my timing saying, “No, not like that! Why can’t you get it right?!???”.
When I finally broke it into separate skills, each one focused on doing one thing well, the quality jumped immediately, and I spent way less time going back and forth.
Mistake 2: I gave Claude AI-generated examples.
I was trying to move fast, so I used captions Claude had already written as my example content. Which meant I was teaching Claude to sound like...Claude. The output was technically fine but it was missing me.
No wonder I hated all the content 🤪
When I swapped in examples of my actual writing, the skill finally started to sound like me.
So before you try to automate your entire content department, make sure you have strong skills that give consistent output you’re happy with.
Next week we get into the architecture that holds it all together.
If you want off the content treadmill too…
This is becoming a course. If you want priority access and waitlist-only perks before it opens, get on the list now.
Next week: the deeply boring decision that makes or breaks your entire agent.
xx
Emily
➡ Support AI For Her. Give this ad a click ⬇
Voice dictation that doesn't mangle your syntax.
Most dictation tools choke on technical language. Wispr Flow doesn't. It understands code syntax, framework names, and developer jargon — so you can dictate directly into your IDE and send without fixing.
Use it everywhere: Cursor, VS Code, Warp, Slack, Linear, Notion, your browser. Flow sits at the system level, so there's nothing to install per app. Tap and talk.
Developers use Flow to write documentation 4x faster, give coding agents richer context, and respond to Slack without breaking focus. 89% of messages go out with zero edits. Free on Mac, Windows, and iPhone.
HER POWER BRIEF
💄 Canva just rebuilt itself from the ground up with AI — describe what you need and it builds it, layered and editable, connecting directly to your Gmail, Notion, and Google Drive. Design menus: officially optional.
💄 Claude got memory — it now builds a rolling summary of your Projects and preferences, updating every 24 hours automatically, so you can stop re-briefing it from scratch every single session. Amen.
💄 ChatGPT dropped Images 2.0 — a thinking image model with 2K resolution, up to 8 coherent images per prompt, and shockingly good text rendering. DALL-E 3 is being retired May 12, so pour one out.
💄 Taylor Swift is trademarking her voice and image to fight AI deepfakes, and if it holds up in court, it could rewrite how trademark law protects — or weaponizes against — every creator online. The IP era has entered the Eras tour.
⬇️ Spread The Movement ⬇️
Know another incredible woman in AI who would love this newsletter? Use the button below to share it with her, your email list or community.



