If you only check one highlight this week:
B2B CMO: study the first two sections to seriously accelerate AI adoption on your team.
Everyone else: follow the Claude Cowork 101, whether you use Claude or you decide to use OpenAI's new Codex app.
And sorry for the final words quote this week, I couldn't help it.
-François
AI adoption acceleration: how Meta prepared for their upcoming layoffs. Or when the half-day hackathon won't cut it
You've probably heard the rumor: massive layoffs are coming to Meta. Maybe up to 20% of the workforce.
Here's what they did to ensure they accelerate AI usage and increase team productivity before that happens (note that this may be something they would have done anyway), but I highly suspect it’s related - and they used the data to inform who could be on the chopping block.
I'm sharing this not to suggest you follow the same path, but because I've noticed that executives are becoming increasingly impatient with the bottom 20% of employees who are reluctant or very slow to adopt AI.
Starting to be hard on them now could actually be a show of empathy (almost no matter where they go next, they'll need to use AI well).
What Meta did:
Zuck ordered a week-long AI sprint, where teams were instructed not to hold any of their regular meetings, but to learn how to use AI and set up their AI workflows and agents. Imperative, not a suggestion.
He agreed to postpone product releases and announcements by a week so teams could fully focus on that. I don't know if this was company-wide or only for the product and eng teams. I suspect that was difficult to do for customer-facing teams.
The internal platform teams set up a version of OpenClaw called myClaw, on their own infrastructure for safety and compliance. They also set up a system to gather and manage all context called "Second Brain.
They ensured employees had access to a lot of AI tools and told them that budget is not a concern this week. It seems like they could use whatever model they wanted, but they were mostly using Claude. and builders Claude Code, especially
I strongly suspect they were tracking AI tools adoption & usage to inform who will be part of the upcoming layoffs. (Note: Ramp also shared publicly having a four-level classification of all employees from L0 (reluctant) to L3 (advanced), and that if folks are still at L0 within a few months, they would lose their jobs.
They had to figure out a lot of stuff on their own.
Some folks I talked to reported making incredible progress during that week with their agents and workflows. A friend to whom I used to show stuff is now teaching me. Her agents save her a ton of time and help her be more strategic. It seems to have increased their job satisfaction.
Most CMOs I talk to are considering holding a half- or full-day AI hackathon once a quarter, but given the impressive results reported, it is well worth considering a multi-day company-wide mandate.
When everybody does that at the same time, and the platform and IT teams pave the way, things move faster.
Of course, if you want to pull off what they did, it should be a CEO mandate
Still hesitating?
As I finished writing the lines above, I just saw this on LinkedIn:
Another option to accelerate AI adoption: small pods with an AI trainer.
below are some best practices I collected from Guillaume, a growth marketer who became an AI builder and trainer.
I've referred him to a few clients who requested AI enablement for their teams. He’s been refining his training process. This is what he reported this week. The Chief of Staff of the CMO, who brought him in, is raving about the results so far.
Phase 1: all-hands briefing and some enablement
Start with some all-hands inspiration and training sessions.
There, the exec sponsor sets objectives and insists on how critical it is to do it right and fast.
IT or whoever is the expert present the key tools, processes and best practices
2 or 3 team members or small teams who've set up great workflows or agents give a demo with the build for inspiration
Phase 2: Create pods. Focus on each pod for one week to make them move fast.
Cap pods at ~12 people and split into small working groups of 3–4.
Design sessions as working time: most teams won’t do meaningful work on their agent between sessions
Run 3 x 90-minute sessions in one week, ideally to take teams from “zero to usable workflow.”
Space sessions every other day (e.g. Mon/Wed/Fri) to keep momentum
Session 1:
Start with workflow discovery, not generic training (already done). Best approach: use an organic kickoff discussion to surface real use cases, then turn those into live workflows.
Avoid over-specifying workflows upfront. Too much detail creates friction and reduces ownership.
Session 2:
This is hands-on build time with breakouts as a working group and facilitator coaching, moving across teams in this case, Zoom breakout rooms.
Session 3:
Time for demos: teams show the workflow working via screen share. They collect comments and questions, and from there can keep iterating.
Important before you start:
If you've been reading my newsletter, you know how critical central context is. So make sure you identify what's going to be your central team knowledge brain.
In the case of the latest trainings that Guillaume held, since the company was using Google Workspace/Gemini, they used NotebookLM as the initial team “brain” before connecting outputs into Gemini or other tools.
Some takeaways:
Prioritize small-group execution over large plenary teaching. Faster and drives better adoption.
Give each group enough expert support
My selection of tips, news and workflows
🤖 ChatGPT users rejoice!
If you're a ChatGPT/OpenAI fan, you must be getting tired of people raving about Claude all the time.
Well, this is your week!
OpenAI has released a new version of Codex, and the reviews are really good.
My current expectation is that the Cowork/Codex set of "professional agents" for non-technical users will be one of the most important and fastest growing product categories of all time.
Don't get fooled by the name: non-coders can use it. You can create and manage your agents there.
This may be the easiest and most secure tool (also called a harness) to use now, or OpenClaw-like features.
A lot of what you'll read in the section below about Claude Cowork is valid for Codex (skills, computer use, etc.), with one big difference: Codex should be easier to use, for now, to create agents.
Tip: Start by setting up your connectors (MCPs).
📺 A Claude Cowork 101
Claude Cowork has evolved a lot since its launch.
JJ Englert, AI trainer, joins Claire Vo’s How I AI podcast and shares how he uses Claude Cowork as his daily business partner, from:
organizing his workspace to
creating skills, to
preparing a daily morning brief, and
running multi-agent review systems
Full episode here, plus a detailed summary (since Cowork is becoming so powerful, I thought it was worth giving it so much real estate) of key capabilities and use cases below.
When to Use Cowork vs Claude Code
Cowork for business mindset and personal productivity, for knowledge workers who aren’t highly technical
Claude Code to build apps and tools
in a desktop app or terminal-based development work
More technical implementations
Cowork can serve as an on-ramp to Claude Code
Available in the Claude Desktop app
Desktop has three modes: Chat, Cowork, Code
Cowork’s Core Capabilities
Performs actions on your behalf, including through connectors (i.e. connected tools via MCP)
Desktop and browser (via the Chrome extension) access: Can organize files, book reservations, and find restaurants
Task automation
Projects vs Tasks Architecture
Tasks: Individual chat sessions that can work independently. Work better within a project
Projects: Container for multiple tasks with shared memory
All chats/tasks within project share same context
Persistent memory across conversations
Orchestrator view of active agents/tasks - you can think of them as subagents
Sub-agents operate with fresh context windows for diverse perspectives
Project instructions: High-level guidance specific to each project
Right Context panel shows folder contents and project memory
Connectors
One-click authentication to business tools via connectors (MCP)
Available connectors: Google Calendar, Drive, Notion, Slack, Gmail, Bitly, Granola and MANY more - go here and click Browse Connectors in the top right
Granular permission controls per connector
Can disable specific actions (e.g., create email drafts)
Set to “always allow” or “ask permission”
Dual function: Task automation + data ingestion for better AI performance
You can use a progressive trust approach in permissions you give to each tool: Start small, gradually increase AI access. Balances functionality with security
Skills
Skills = instruction sets in markdown files for repeated tasks
Skills are invoked automatically when relevant tasks arise
Skills can chain together for complex multi-step workflows
Tip: Include good/bad examples to train AI on success criteria
JJ’s skill examples:
Email writing (analyzes last 30 days of sent emails for tone and writes his email style guide)
Social media platform-specific formatting and voice matching
Getting Started: Projects and Folder Structure
Good folder structure = better, faster results. Claude navigates files quicker than you do, but only if things are organized.
3 setup rules:
One project = one folder. Everything for a given area of work lives in one place.
Add a "brain" file. A markdown file with your preferences, voice, and team info. Claude reads it before every task.
Generate a workspace map. Ask Claude to map your folder structure so it navigates efficiently without ingesting everything.
JJ's example: /jj/projects/ → JJ's Cowork Workspace/ → JJ's Brain/ (about-me, brand-voice, email-style-guide, team-members…) + Content Creation/ by platform + Current Projects/ + workspace-map.md at root.
Workspace Map Function
Claude-generated navigation system for folder structure
Maps out all folders/files to help Claude navigate efficiently
Reduces token usage by directing Claude to specific relevant folders
Example: Twitter post request → automatically uses Twitter folder skills
Created by asking Claude to “create workspace map” for any folder (F's note: you can schedule a task asking to update it daily. )
Scheduled Tasks and example: Morning Debrief System
Daily automated review at 7:30 AM
Analyzes email, Slack, calendar for day planning
Identifies messages needing attention
Prepares meeting prep materials
Outputs a structured daily action plan
Runs within project context (access to all skills/connectors)
The Anti-To-Do List Concept
Identify and focus on tasks you never want to do manually again
Build skills to eliminate repetitive work
Burn down list of annoying recurring tasks
Examples: First-draft emails, calendar conflict resolution, meeting prep
For me, it was manually editing Zoom video calls to extract each lightning talk, summarize, and post to YouTube (note: I used Claude code to create this internal tool)
Use Case: Multi-Perspective Review System
Sub-agent framework with different personas
Example personas for JJ who trains people on using AI: AI builders, AI executives, security-worried users
Each agent reviews work from their own fresh context window
You can do the same and simulate colleagues' feedback (boss, engineering partners, etc.) before expensive synchronous meetings
Particularly valuable for solo founders/remote workers
Use Case: Newsletter Workflow Example
10-step automated skill sequence, including:
Interview him for content topics
Sub-agent internet research on AI trends
Subject line optimization
Body segment creation (each with specific skills)
Sub-advisory board evaluation
Tip: Feed it past performance data (good/bad examples)
💡Want the easiest way to get started with Claude Cowork? Test this
If you use Google Mail, Calendar, and Workspace as well as Granola:
Enable the connectors in the settings of Claude, and then
Click on that "Create a daily briefing" section in the Ideas of Claude Desktop (see the top left tile in the screenshot below).
Watch the magic happen.
I was floored by the quality of the daily agent it created in five minutes for me: it would flag the important emails of the day, give me a good briefing for my meetings, and more (see last two screenshots).
Final Words
Meta is building an AI clone of Mark Zuckerberg.
- Is this is the only bot in history that's ever going to be criticized for being too lifelike?
- It'll be the first time that both a person and their AI avatar failed the Turing test.
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.
