Sunday, September 13 · Issue #1
The Weekly Signal is a weekly letter about doing good marketing work as AI changes how we do it. Expect useful ideas, people worth knowing, and things I’m figuring out in my own work.
What’s on my mind.
Learning AI shouldn’t be a second full-time job
I love working with AI. I will happily disappear into a problem for hours. But we can't make that the entry requirement for keeping up. In the first Weekly Signal, I talk through the decisions tutorials leave out, finding people who understand your situation, and choosing what deserves your limited learning time.
14 min with Hanna · Watch or read on the article page
Good thinking deserves a bigger audience.
The reads, watches, and listens that made the shortlist. These people get it.
Good AI work should be something your team can join
Sergey Karayev · The Multiplayer AI Manifesto · Article
We moved from shared documents into private AI chats, then started copying answers between them. The manifesto makes a useful case for shared agents, shared sessions, and learning that stays with the team. If a colleague has to reconstruct your entire conversation to continue the work, there’s still a handoff problem to solve.
Give AI a job before you give it a prompt
Kyle Kober · ChatGPT Work onboarding guide
Kyle’s guide starts with the work you actually do. That sounds obvious, but it’s easy to spend an afternoon configuring an agent without deciding what a useful result looks like. Pick a recurring responsibility, give it the relevant context, and define the output you need. A much better starting point than collecting another 50 prompts.
Your own work is the benchmark
Alex Finn · Post
Alex recommends building a small testing harness around the tasks you do most often. Keep a few real briefs, reports, or research questions and use them to compare models. Judge the result by accuracy, usefulness, and how much editing remains. A model winning somebody else’s benchmark tells you very little about whether it will make your Monday easier.
This part stayed with me.
“Focus on the context layer before you build a pile of skills.”
Emily’s latest newsletter goes inside the AI setups at Buffer, LangChain, and Mintlify. The teams organize the work differently, but their skills all depend on having useful company context. This part is easy to skip when building the next thing feels more fun.
For your “wait, we could do that?” list.
Creative ways to put AI to work on growth. A few possibilities for your next experiment.
Build a short-form format scout
Eric Siu shared a skill that uses Astra to scan high-performing short-form content, recommend formats around his topics, and refresh the research each week. It gives you a way to study what’s working without spending the whole morning scrolling. This is his documented workflow; the test below is a starting point for yours.
Start here: Choose one topic your audience cares about. Review the formats the scout finds, pick one, and make one piece. Track qualified replies or inquiries alongside views before asking it to scale.
An account-specific demo beats a first-name mail merge
Lorcan O’Rourke · Solutions Engineer, Clay · Demo
Lorcan connected Claude Code, Clay, HeyGen, and ElevenLabs to research target companies and create personalized demo videos. He shows the system working across 20 accounts. The interesting part is the research shaping what each company sees. For a small group of valuable accounts, that’s a more useful experiment than making the same generic pitch cheaper to send.
A clip machine is only as good as its feedback loop
Cody Schneider · Short-form production breakdown
Cody breaks down a system that pulls from a large video library, finds clips using word-level timestamps, edits them, and feeds performance back into the next round. Production is only half the job. If the system learns from views alone, it can get very good at entertaining people who will never buy. Decide what a valuable response looks like before increasing the volume.
A few things to know before Monday.
What changed, what’s useful, and what it means for your work. Occasional meeting small talk included.
Check a skill before you hand it the keys
NVIDIA’s open-source SkillSpector scans agent skills for risks such as prompt injection, data theft, and unsafe code. It can inspect a repository, directory, ZIP, or individual skill file before installation.
What it means for you: Add a scan to your process for evaluating a new skill. Read the findings and permissions too; a clean report is useful evidence, not a guarantee.
An email can be the start button for an agent
Riley Brown shows how AgentMail and a webhook can trigger a GrokBot routine when an email arrives. Forward a message, and the workflow starts without opening another chat. His walkthrough covers the email setup before moving into payments.
What it means for you: Think about the routine messages that already kick off work: a brief, an inquiry, or a report. One could trigger a draft or a review checklist for a person to approve.
An entire brand relaunch in a weekend
We used GPT-6 Astra to do an entire brand design, standards, and collateral relaunch in a weekend, including logos and realistic merch mockups. It looked better than what most designers have given me in the past. I’m calling it a legitimate game-changer for creative development, in a sea of course-sellers saying “game changer” every other day. (Yes, I hear the irony.) But try it yourself. You’ll see.
What it means for you: Give it a real brief and your actual brand references. Ask for an editable diagram, presentation, or visual concept you need, then judge how well it follows the brief and handles revisions.
A reminder of what great marketing can feel like.
Campaigns, communities, experiences, and other good work that makes people stop and care.
The morning briefing you can actually put down
Karen X. Cheng · The Morning Newspaper
Video still: Karen X. Cheng, The Morning Newspaper.
Karen made an agent that turns her calendar and email into a personalized newspaper and prints it while she sleeps. I love the physical part of this. You can read what matters, finish the page, and get on with your morning. There’s a stopping point built into the experience.
Something to borrow: Give useful information a shape that fits someone’s day. A page they can finish may do more for their attention than another feed they can keep opening.
Honorable mention: go make something weird
Playgrnd · Brett
Playgrnd offers 32 free little tools for animation, dithering, and playful SVG experiments. No account required. If you’ve spent the week making useful things, this is a nice excuse to make something simply because you like how it looks.
You’re finding good stuff, too.
This week’s question:
What is something that you've tried to do with AI at work that sounded straightforward until you actually tried it?
Tell us where you got stuck. Leave a comment on this episode's Beehiiv article so we can compare notes.
Know someone doing great marketing with too much on their plate? Send them The Weekly Signal. One less thing for them to dig through.
Referrals get early merch.
Share The Weekly Signal with a friend for early access to merch like this tote.
Someone sent you this? Good friend.
Subscribers get sticker sheets.
Get The Weekly Signal in your inbox every Sunday.
