If you only check one highlight this week:
B2B CMO? How to build your knowledge system for your AI agents. Someone on your team needs to take the lead on this. Some CMOs do it themselves
Everyone else: How to vibecode your agents or apps by starting with detailed specs.
-François
How should you approach building with Claude Code or Codex if you're not technical?
Last week, on our PLL community call, Nico Ehrmann demoed a phenomenal conference scouting and planning app he built (watch here). It would make any CMO or Head of Field/Events Marketing drool.
He built it all with Claude Code.
At the end of his demo, someone from the group asked how someone less technical than Nico should approach building something like this.
We then had a super insightful conversation. We had developers and less technical folks in the virtual room.
Here are the key takeaways:
Themes & Best Practices for Building with AI
1. Spec Before You Build
A well-formulated brief is the single highest-leverage investment you can make.
Work with AI to write, in plain English: problem, current state, goals, constraints, why it could fail, ideal solution, etc.
Ask the model “what am I missing?” and iterate 2–3 rounds before letting it build code
Break the spec into small stories each model can one-shot
The advanced folks use multi-model critique: e.g. plan with Claude, critique in Codex/ChatGPT, implement in Claude. Models are notoriously bad at critiquing their own stuff
Here's a presentation I gave showing how I work on defining the plan and understanding it with Claude (focus especially on what's in orange).
2. Start Small. Then Test and Iterate
As good as models are right now, initial complexity kills projects: ship a working kernel first and grow from there.
Get one thing working end-to-end before adding extra features
Test each addition, roll back what breaks
Nico’s conference scout started small. He added features over weeks
3. Build and Maintain a Shared Context Layer
Agents are only as good as the context they can access.
Store knowledge in markdown files (brand, ICP, competition, messaging).
Save locally on your machine and sync to GitHub so you can share easily
Use a “marketing brain” with full company context as your starting point for any new project
4. Solve a Real, Recurring Problem
Pick a task that costs you hours every week and automate it
Nico’s trigger: the same conference questions kept coming up every year, so he built the tool instead of answering them again
5. Mind the Security Boundaries
If your app’s code touches real data, scope access deliberately:
Limit AI to the subset of data it actually needs, not the full database
Use authentication layers, even for internal tools
Loop in your security/IT team early—they’re often eager to help
My selection of tips, news and workflows
🔍 How to build your knowledge system for your AI agents? How to keep feeding and updating them?
At the risk of sounding like a broken record, your AI or agents are only as good as the context and skills you bring them. Some started, in a confusing fashion, calling that our “agent operating system”. I prefer calling it “our second brain”.
As mentioned in the first section, the consensus today is that the best approach is to: build and maintain a set of Markdown files (context, skills, etc), save them on your machine, and sync them on GitHub.
You can build all that with Claude Code or OpenAI Codex. Once on GitHub, you can share the key repos (folders and files) with:
other team members
your favorite AI tools
You can even ask Claude/Codex to build your own connector (MCP).
The post below by Joni, CMO at Barona, confirms this approach. If you want a good step-by-step on how to do this, listen to this (Spotify, YouTube), or take this self-paced course.
🤖+🧑 How to become an AI-augmented CMO
A few weeks ago, I joined the CMO Huddles community call and discussed my observations and tips for becoming an AI-augmented CMO. You can watch here.
I made these points, among others:
Three mistakes CMOs make with AI
Delegating AI work instead of staying hands-on.
Being a judge during hackathons or build sprints - be a player instead.
Going easy on people who refuse to adopt (provide training, encouragement, and tools first).
Block the time to learn and build together. Often
Quarterly hackathons = necessary but not sufficient.
Maintain a shared list of your team’s best skills, agents & workflows + how they work
Every all-hands, ask two or three people to showcase their workflows.
🖥️ Should you use Claude Code in the terminal or the Claude desktop app?
Non-technical friends often ask me this question.
Earlier this month, Anthropic rebuilt the Claude desktop into a great home for Claude Code. Embedded terminal. Computer use. Browser preview. Visual diffs. Click-to-connect for GMail, GDrive, Granola, Slack, Figma, GitHub, Notion and many more (all in settings, under “connectors”). Even AllTrails! No command-line knowledge required.
The desktop app may even be more powerful than the terminal for non-coders.
I had Claude create this comparison page so you get a sense for what you can now do with the desktop app - which is also the home of Claude Cowork.
🤩 Skills for the strategic win!
When a strategic GTM, pricing, and PLG mind shares his skills, you don't hesitate: you say thank you (I did!), and you add it to your Claude or Codex account (I did!).
Kyle Poyar has extensively studied PLG, SaaS and AI pricing and strategic marketing in general. He just shared four of his skills. I have not tried them yet, but knowing the guy, I don't see how they could be bad.
If you download them, please give him a like or leave a comment on his post to encourage this kind of sharing.
The four skills are:
/deep-gtm-research skill for GTM research
Runs a structured workflow: gathers context first, proposes a research plan for your sign-off, prioritizes primary sources, and delivers a pyramid-structured report with in-text citations and a source table. It essentially follows the same workflow of a strategy consultant or BizOps manager, as outlined in Torsten Wallbaum’s fantastic guide.
/pricing-teardown skill for pricing analysis
Evaluates your pricing page across 10 dimensions — 7 for the human buyer experience (value prop, plan clarity, cognitive load, trust, behavioral psychology, transparency) and 3 for AI agent readiness (machine-readable pricing, FAQ coverage, per-tier depth). Each dimension gets a 1-4 score with a specific finding from your actual page, an overall letter grade, and a prioritized set of quick wins vs. strategic improvements.
/icp-sharpener skill for pressure testing your ICP
It asks six questions to understand your best customers, what they had in common before they bought, and who churned and why. Then it synthesizes the answers into a few distinct segments along with a firmographic profile.
/gtm-plays-brainstorm skill for identifying outbound plays
Collects your product, ICP, GTM motion, and company stage — then selects 5-8 plays from a library of 25+, ranks them, and explains exactly why each one fits your company specifically. Includes the trigger signal, how the automation works, what tools enable it, and a "where to start" recommendation for the play to launch first.
Final Words
Automate to 100%. An automation that is only 95% accurate isn't actually an automation. It's a liability. Put in the "elbow grease" to get it to 100%.
Thanks for sharing these highlights with busy marketing execs around you.🙏
Someone forwarded you this email? You can subscribe here.
François | LinkedIn
I'm a CMO, advisor, and "CMO Wingman". Yes, that's a thing :-). Ask my clients: in this AI era, CMOs need a strategic proactive advisor more than ever. I’m former CMO at Twilio, Augment Code, Apollo GraphQL, Decibel, Udacity and Head of Marketing for LinkedIn Talent Solutions.
