If you only check one highlight this week:
B2B CMO? Nico’s content creation system. Someone on your team needs to create one. And they don't have to do that with OpenClaw.
Everyone else: Why you should consider Claude Cowork as your work OS. Unless you're a fan of Gemini, in which case, see how it just launched Notebooks.
-François
ills
With agents, context and skills - i.e. file management - is the key
You’ve probably heard “With AI, context is everything”. I had not realized how critical that was for agents until I heard the Latent Space interview with Marc Andreessen, who deconstructs what agents are in the middle of the episode.
Of course, agents are more than just a set of files, but the key insight is that we can go VERY far if we figure out how to organize and automatically update all our context, skills, and instructions within those files. For everything else, LLMs and these new tools that improve at an insane speed will do much of the rest for us.
Andreessen describes agents as this stack:
A language model. Any LLM works (Claude, GPT, Gemini, open-source models). This is the "brain" that reasons and generates.
Direct computer access. So the agent can execute commands, run scripts, and use tools on the machine. This gives it hands, on top of the LLM’s brain.
A file system. The agent reads and writes plain files. They act as its memory, instructions (skills!), and knowledge, all of which live as files on disk or in the cloud.
Markdown as the memory format. All agent memory is stored as markdown files. Simple, human-readable, portable.
A heartbeat loop. A timer wakes the agent up regularly to check for work.
The LLM and our agent solution will bring 1, 2 and 5. We’re on the hook for 3 and 4 (with their help).
Because memory lives in files (not inside the model), you can swap models - or even switch tools - without losing anything, as long as you invested in your knowledge base in the form of that rich and auto-updating file and skills system. That’s why Kieran’s 6-min demo of his content creation system focused so much on these files.
Implications for CMOs and their Marketing Intelligence curators:
Your files are now your most valuable asset. Messaging foundations, style guides, process descriptions, competitive battlecards, campaign briefs, and post mortems.
Stop building for human onboarding & education only. Every document you produce should be readable by both your team and an AI agent.
Design your intelligence library like a file system. Use consistent folder structures. Name files descriptively. Write explicit metadata. This is the agent's memory architecture.
Plan for model swaps. Keep your knowledge in portable formats (markdown, plaintext, CSV). Your curated knowledge should survive changing models and agent platforms.
Start building "agent-ready" briefs now. For each product launch, or market shift you track, create a structured file that an agent could act on: what changed, what it means, what decisions you made, and why. You're building the memory layer for your agent.
» The marketing intelligence role is about to split in two: the curator (human) who decides what matters, and the agent (AI) that monitors, organizes, and distributes. The file system is the bridge between the two.
If you want to create your own set of personal markdown files with a lot of your context (identity, roles, projects, tools, communication style, domain knowledge, decision log), follow this video and use this tool to build your portfolio.
You can also read the following sections for more inspiration.
My selection of tips, news and workflows
🔍 This OpenClaw agent squad, built by a PMM, writes, critiques and improves technical blog posts
Want to be inspired and also a little intimidated? I love this workflow and agent squad that Nico Ehrman, Sr Director of Product Marketing at CAST AI, built and demoed at Product-Led Leaders a few weeks ago.
That has completely streamlined his content production. What I love about it is that, now that he's fully tested it, he can give it to any of his subject-matter experts (e.g. in engineering or sales engineering) as a super editor and writer. They can now super easily create their own technical blog posts based just on an initial idea - and some back and forths, of course.
This is both impressive and a little intimidating if you are not as technical as he is (I am not). Which is why I wrote the next section.
Key features:
Persona-first approach: content declared for a specific audience before writing
Templates for multiple content types (blog posts, datasheets, product pages)
Multiple reviewer agents, with various personalities (some representing CAST’s execs, others target personas), improve content quality and prevent AI-sounding output
Multiple LLM models in review process (Claude, ChatGPT, Gemini)
Persistent memory improves over time based on Nico’s feedback
$40 daily budget allocation with spend tracking
Integrates with Slack. Nico receives a message when the content is ready for his review
Creates blog covers and social cards using Gemini/Nano Banana Pro
Enables engineers to create technical content without PMM bottleneck
🦞 Do you also want OpenClaw-like agents, but safer and easier?
You've probably heard a lot about the OpenClaw hype but shied away from using it due to complexity and security issues (rightly so re: the latter) - and maybe a lack of a compelling use case. Well, you really only missed out on being ahead by a few weeks, given that most vibe coding and agent tools are adopting some of OpenClaw's best features, in easier and more secure ways.
Below is an example from Base44: Superagents. Perplexity introduced Perplexity Computer, and Manus keeps improving fast. Fortunately for me, Claude has also been moving fast (see the next section).
So, to all of you who have been asking me, "How do I set up my agent teams with an agent manager and sub-agents that do loops, etc., to get to great outcomes?" The simple answer is: “Sit tight. It's probably already in your favorite AI tool. If not, it is coming very soon.”
“Meet Superagents, a new way to build AI agents in Base44 without the usual setup. Just describe what you want in a conversation and your Superagent can connect to tools, run scheduled tasks, respond to events, and keep working for you around the clock. Your Superagent can live inside WhatsApp, Telegram, Slack, or your browser, so it's always ready when you need it.”
🏃♂️ Claude - and especially Claude Cowork - is getting better at a breakneck pace
For all of you Claude lovers, and that's a lot of us around Silicon Valley, Claude has been getting better really fast. It shipped critical features recently, many inspired by the OpenClaw architecture.
Cowork was only okay when it initially shipped, and frankly, I wasn't sure why I should use it over chat or Claude Code.
But the improvements are impressive and frequent, so much so that it's becoming my work OS (until I need to revert back to Claude Code to build my own utilities or apps).
Here is an overview of what they've recently shipped. You can hear more in this podcast. I'll likely be sharing more of my journey with Cowork in upcoming newsletters
Better interface: still chat-based, but now you can see in the right rail (screenshot below): the plan, the progress vs the plan, the folders + context it's working with, and the connectors it's tapping
More connectors. You can see a screenshot below of the ones I selected
Scheduled tasks. For instance, every day, Cowork fetches the notes from my client meetings, automatically adds them to my Google Drive docs (my file system), and emails me a report with my to-do list.
Projects are now available in Cowork
Computer use: In Claude Cowork and Claude Code, Claude can use your computer to point, click, and complete tasks. When Claude doesn’t have access to the tools it needs (including the connectors above), it will point, click, and navigate your screen to perform the task itself. It can open files, use the browser, and run dev tools automatically — with no setup required.
Available in research preview for Claude Pro and Max subscribers, on macOS only.
Dispatch lets you have a single continuous conversation with Claude on your phone or desktop. You can assign Claude a task on your phone, turn your attention to something else, then open up the finished work on your computer. Now, Claude can use your computer on your behalf while you’re away.
Enterprise features to help teams deploy Claude Cowork company-wide: role-based access controls for Enterprise, group spend limits, expanded OpenTelemetry observability, and usage analytics for admins to see Claude Cowork adoption.
OpenAI user and fan? Something similar is coming your way: Fidji Simo, OpenAI's applications CEO, has announced that the company is working on a super app that will bring together chat and an easier-to-use version of Codex.
The connectors (i.e. MCPs) that I set up with Claude. Work across chat, Cowork, and code. I'm especially super happy with the connectors to Granola and Google Drive to collect and centralize all my context!
You can understand the connectors' capabilities and activate or deactivate them in Settings
The right rail of Claude Cowork in the desktop version
📚 Talking about centralizing context: Gemini launches notebooks. They sync between Gemini and NotebookLM
NotebookLM was one of Google's first successful AI apps post-ChatGPT. It has improved a lot recently.
For instance, I wanted to learn more about the AWS Coopetition playbook during my next run, so I conducted deep research on the topic, curated the sources I found into a NotebookLM, and told it to create two audio overviews for me. One that presents their approach, another that critiques it.
So I was quite excited to see that there is now an easy way to:
Tap my existing NotebookLMs in Gemini 👇
More importantly, create notebooks right in Gemini where you can add and create a lot of context
Notebooks let you use more sources depending on your subscription plan, helping you tackle bigger tasks and longer-running projects. And since notebooks sync across the Gemini app and NotebookLM, any source you add in one place automatically appears in the other. This continuity means you can use unique features of each app, like Video Overviews and Infographics in NotebookLM, even if you started a notebook in the Gemini app.
Notebooks show up in the left rail of Gemini
The different types of audio overviews you can create in NotebookLM
🛠️ Want to learn faster? Ask Claude to show you
Show me is a cool new feature by Claude.
Use a "Show me how..." prompt in Claude. It helps you not only visualize key concepts, but it may even vibecode an interactive answer for you right in the chat.
I tested with a simple prompt: "Show me how to set up a B2B marketing funnel model for ABM".
It came up with the 1st image attached.
I was initially disappointed: I thought it wasn't interactive...
Until I read the fine print: "Click any stage to go deeper". I then did that and immediately generated a second prompt: "What are the best tactics to engage target accounts in ABM? How do I coordinate marketing and sales?"
It then generated the second image plus wrote relevant tips underneath.
Want to try it?
Simply drop this prompt into Claude: “Show me how compound interest works.” Pick Sonnet 4.6 Extended as the model.
💡Tips for using Claude Code better
A lot of great tips for working with Claude Code in Monica Lewis's post and in the comments.
The tl;dr:
“1️⃣ Create a stellar Claude.MD file
2️⃣ Trust but verify
3️⃣ Setup a good development environment
4️⃣ Use voice-to-text:
5️⃣ Work in parallel:
6️⃣ Use ChatGPT or another LLM and Claude in tandem
7️⃣ Ask it to clean up your code
8️⃣ The last mile takes half the effort”
💡Stitch = a free vibedesign tool by Google
ChatGPT 5.4 already has a reputation for doing terrible design. I find Claude Code impressive. This week, Stitch by Google has gotten great reviews. If you need to design a website or an app, then turn it into code and/or port it into Figma, Stitch looks like a great tool. especially because it seems to be really great at consistently following your brand style guide. See a detailed overview in this Twitter thread.
Final Words
It's basically LLM plus shell (i.e. computer access), plus file system, plus markdown, plus cron (i.e. heartbeat). And it turns out that's an agent.
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.
1
