David Arnoux

I help B2B founders and marketing leaders grow with AI, GTM strategy and LinkedIn.
20 years scaling companies: Growth Tribe (40,000+ learners), e-commerce brand (€30M+), SaaS (15m+ ARR).

Now:
→ Fractional CxO for AI adoption & GTM @ heyarnoux.com
→ Building AI tools for founders @ humanoidz.ai (ViralBrain, NuggetFinder, BuilderBox)
→ Running an invite-only AI community @ thegenaicircle.com

I write about AI, growth and building in public.

Co-CEO

Growth Tribe is here to empower adults to always be learning. Our solutions ensure people & organisations can acquire the digital skills they need to adapt, stay up-to-date and in the lead!

We’ve built a reputation for fast-paced, impactful & enjoyable learning, offering training & coaching in Data Science, Analytics, Artificial Intelligence, Growth & Marketing, Innovation, UX Design & Product Management. 

With campuses in 6 European countries, we have supported the learning journeys of over 16,000 adults in over 900 organisations. Our 80+ staff are dedicated to creating Europe’s best learning experiences, to empower humans to always be learning.

most people using Claude Code are at level 2. they don't know it.

we mapped out 10 levels of Claude Code mastery based on hundreds of real setups inside our community. from "terminal tourist" (just installed, treating it like ChatGPT) to "swarm architect" (agents spawning agents, autonomous loops running 24/7).

get it here: https://lnkd.in/ez_6ujUk

here's what we found:

level 0: you type prompts and hope for the best
level 1: you have a CLAUDE . md file. you know /compact and /cost
level 2: you connected MCP servers. Slack, Notion, Drive pulling live data
level 3: you built custom skills. repeatable workflows, one command
level 4: you have memory files, patterns, examples. context engineering
level 5: multi-phase skills. subagents. systems that chain together
level 6: headless mode. scripts calling Claude Code. JSON piping
level 7: Playwright browser control. screenshots, scraping, PDF generation
level 8: parallel sessions. orchestrator + specialist agents
level 9: cron jobs. background agents running 24/7. Claude Code as infrastructure
level 10: autonomous loops. agents that build agents. a handful of people on the planet

the gap between each level isn't knowledge.....it's mostly just setup.

so we built /level-up.

it scans your actual environment: your CLAUDE . md, your skills, your MCP servers, your memory files. tells you exactly where you are. then gives you a personalized roadmap to the next level only. not all 10, just the next one.

best part: it offers to build the first step for you right there in the session.

want it? please go ahead and follow all the simple instructions here:

👉 https://lnkd.in/ez_6ujUk

built with the GenAI Circle community. 350+ builders testing this stuff every week.

Helping GTM Leaders & Founders Grow With GTM x AI | Fractional CxO | Building Linkedin Tools @ humanoidz.ai

skills are the new "apps"

if you're still prompting...there's a better way....

most people using Claude Code are still typing the same instructions every single session. "write in this tone." "follow this format." "check these files first." every. single. time.

meanwhile the smartest users I work with built something different. they turned their best prompts into reusable skills. and skills are being shared everywhere. one command, entire workflows. not copy pasting from a notes app. actual saved playbooks that Claude runs on command.

You can build one that writes ad copy by pulling real Meta performance data. Some have replaced Canva entirely, canceled all design subscriptions. Someone in our community made a procrastination coach that pushes back when he goes off task. Another member is running multiple multi million dollar GTM builds simultaneously.

(the secret is the compounding effect. skills don't just save time once. they stack. single skills become skills with context, then skills with hooks, then skills that chain into full workflows. eventually the system builds its own tools)

and there are already 7,000+ pre built skills you can install today. most people don't even know they exist.

I put together a full playbook covering:
→ what skills actually are (and why they're not just fancy prompts)
→ the 5 levels of skill progression from basic to self building
→ CLAUDE [.] md as your AI's long term memory
→ skills vs hooks and when to use which
→ the "builder's trap" and how to avoid it
→ a 4 week plan to go from zero to compounding system
→ where to find 7,000+ ready made skills

want it? you know the drill:

1. like
2. comment "skills skills skills"
3. repost just to be sure

(make 100% sure we're connected) will send it over.

thanks to the GenAI ⚫️ Circle team for road testing every single one of these patterns.

Skills Are The New Apps - Claude Code

I replaced canva with a command line.

2 seconds per design. $0/month. (true story)

open canva. find design. click text. change font. download. check result. 10+ minutes per change. need 50 social cards? do it 50 times. manually.

brand consistency? hope your team remembers the hex codes. version control? doesn't exist. someone overwrites your template and it's gone. then one of our community members showed us a different approach.... a script that generates the same design in 2 seconds flat.

the setup? three free tools.

→ Claude Code writes the HTML and CSS. you describe what you want, it builds it
→ Puppeteer renders to pixel perfect PNGs and PDFs. open source
→ TypeScript handles logic and loops. 500 variations from one template

The real power is the brand guide. one markdown file with your colors, fonts and spacing. change a color in the guide, every asset updates automatically.

try doing that in canva.

what you can generate:
→ social media graphics (instagram, linkedin, stories)
→ PDF reports, proposals and one pagers
→ full presentation decks
→ linkedin carousels
→ email headers
→ batch generation (50 cards = ~2 minutes instead of ~4 hours)

canva pro: $120+/year per person
code approach: $0

we put together a detailed 7 step playbook:
→ environment setup (project folders, install puppeteer and typescript)
→ brand guide system (one file rules everything)
→ your first design script (working in ~30 minutes)
→ reusable component library (like lego blocks for layouts)
→ PDF report generator (multi page client docs)
→ presentation deck generator (full slides, multiple layouts)
→ quality control scripts (automated brand compliance checks)

first timer? just do steps 1 to 3. your first design in ~30 minutes.

want it? you know the drill:

1. like
2. comment "Design"
3. repost to help others

(make 100% sure we're connected)

first come first serve!

thanks to the GenAI Circle team for building this one.

auto-generate design assets with Claude Code

still feeling left out with Claude Code?

here's how to go from zero to intermediate...

some people haven't even started with Claude Code (usually a bit lost or haven't dedicated the time to get set up.) Most people use maybe 20% of what it can actually do. they run one task at a time. research? wait 5 minutes. audit a page? wait again. draft emails? more waiting.

one thread. one bottleneck. painfully slow. what you need is multi-agent orchestration. it's basically running claude code like a small team where everyone works in parallel.

there are 3 layers to this:

→ sub-agents: fire off 3 tasks simultaneously. each gets its own context window. results come back while you keep working. like hiring freelancers for specific jobs.
→ skills: reusable workflows you write once in .claude/commands/. think /commit, /review, /audit. write once, invoke forever. consistent quality every time.
→ agent teams: multiple claude instances collaborating. a lead agent breaks down the work, then frontend, backend and tests all run at the same time. like having a PM + 3 developers.

(the trick is starting simple. sub-agents in week 1. your first skill in week 2. combine them in week 3. agent teams only when you actually need true parallelism.)

By the way, one thing people often miss : token usage. sub-agents use 300-500% more tokens. agent teams can be 7x. so use haiku for simple tasks, set max_turns limits, restrict tools per agent.

the carousel above gives you the overview. but i also wrote a really detailed playbook that takes you from literally nothing to running multi-agent setups:
→ how to install claude code and authenticate (5 min setup)
→ picking your interface: terminal vs IDE extension
→ the best adoption strategy (pick one use case a day, not a grand plan)
→ all 3 orchestration layers broken down with real examples
→ building custom agents in .claude/agents/ with restricted tools
→ writing your first skill in .claude/commands/
→ chaining skills together for compound workflows
→ hooks: auto-linting, guardrails, automation glue
→ cost management (because multi-agent burns tokens fast)
→ real community setups from members running this daily
→ a 4 week roadmap from zero to intermediate

basically the guide i wish i had when i started.

want it? you know the drill:

1. 👍 like
2. 💬 comment " /claude "
3. 🔄 repost just to be sure

(make 100% sure we're connected)

will send it super fast (first come first serve)

thanks to the GenAI ⚫️ Circle team for putting this one together.

Beginners Guide run Claude Code like a five-person team

Someone built a site to check if AI will kill your SaaS.
not just any AI...
It's called Death by Clawd. Tells you how screwed you are. You type in your product and it runs a "SaaSpocalypse Survival Scan."  Funny. Satirical. Just a wrapper...But also.. not totally wrong.

The same week, NVIDIA announced NemoClaw at GTC. For anyone who missed it: they're making it easy to run agentic AI safely on local hardware. OpenClaw + sandboxing + NVIDIA workstation GPUs. These two things are more connected than they look...

Connect an AI to your actual systems. Let it read your Slack, your CRM, your databases, your docs. Give it the architecture to act, not just answer. That's what NemoClaw enables at the infrastructure level. And that's what Death by Clawd is joking about at the surface level.

I'm currently working with startups, PE firms, scale-ups who are asking me one question: how do we replace parts of our SaaS stack with AI agents? Not all of it. Not overnight. But the parts that are just moving data from A to B, formatting reports, sending updates, managing workflows.

The answer keeps being the same. MCP + CLI.

I said it in my community this morning: this might be as big as the ChatGPT moment in November 2022. Not because of one announcement. But because the pieces just clicked together. The models are good enough. The tooling is ready. The infrastructure is there. And the economics make it obvious. I'm not saying every SaaS company is dead. Good products with real moats will adapt. But the "we charge $50/seat/month to do something an agent can do in 4 minutes" category? That's a rough place to be right now. The SaaSpocalypse isn't coming. It's already being planned in boardrooms I sit in every week. Or maybe that's just my bubble. But it's getting louder in here 🫧

For most, upskilling is a buzzword. At Growth Tribe, we turn it into an engine of growth—no fluff. 1000+ teams serviced.
👉👉 www.growthtribe.io/teams

Andreia Saraiva

I ran an edtech company for years.

never once did I build things this useful

I always wanted to build custom interactive onboarding for our users. Every time, same story. Need a designer. Need a developer. Sprint planning. Three weeks minimum. By the time it shipped, the energy was gone. Couple weeks ago I built a full animated onboarding for my GenAI Circle community in a couple hours. 15 interactive scenes. Member names floating across the screen. A click-through journey that takes someone from "I just joined, now what?" to "I know exactly what to do" in 5 minutes.

No designer. No developer. No sprint. Just me describing what I wanted and iterating until it felt right. And that's just the onboarding. I'm building an entire ecosystem for our members right now. 100+ actionable playbooks. Step by step. Tested by the community before they go live. A full repository of Claude Code skills that members can install with one command. Monthly live BrainTrusts where people screen share, ask questions, solve real problems together. 7 WhatsApp groups running daily. A curated MCP master list so people can connect their tools without spending days figuring it out. Bi-weekly recaps so even the busiest members stay in the loop.

All of this built and maintained by one person. Me. With AI.

The part that kills me is this was literally my job before. Interactive learning. Custom onboarding. Building educational products. We had a team. We had budget. We never shipped anything this fast or this personal.

The bottleneck today is just imagination (revenge of the "idea guy" :P )

What a crazy time to be alive. So full of possibilities. Curious what you're building that used to feel impossible....

I built client dashboards for all clients now
with real data, security, auth etc etc
SaaS (dashboard) companies should be terrified right now

Context: I do fractional growth work. Every engagement starts the same way. "We need better visibility on what's working." That used to mean a monthly PDF deck that was outdated the moment I sent it. So I connected Claude Code to Replit via MCP, pointed it at my existing growth engine dashboard as reference, told it to build a version for this specific client. Connected to their actual data sources. 45 minutes later (okay, okay, I'm exaggerating a bit): live dashboard. Real KPIs. Auto-updating. OAuth login, role-based access. The VP sees strategic recommendations. The marketing lead sees content performance. The sales team sees pipeline. Everyone gets the same URL, different views based on who they are.

not a prototype. it's in production. of course the one I'm showing here is anonymized and "dumbed down"

What it tracks:
→ pipeline velocity by stage, with conversion rates between each one
→ content performance across webinars, whitepapers, case studies, email
→ retention cohorts (color coded, week by week)
→ churn risk signals with suggested next actions per account
→ weekly goals vs actuals across 6 metrics
→ MRR with expansion and churned revenue broken out

But the part that surprised them wasn't the data. It was the layer on top.

The dashboard surfaces strategic recommendations automatically. "Double down on LinkedIn, 3.2x higher conversion than other channels." "3 deals stuck in review for 14+ days, here's what to do." It's not just numbers. It tells you what to do next.

Click any module and you see exactly which tools feed it (HubSpot, Analytics, Stripe), when each one last synced, what the key insights are. Full transparency on where every number comes from.

Replit handles the infrastructure. Connectors manage auth to their tools. Hosting, SSL, OAuth all built in. Autoscaling handles Monday morning traffic when 15 people check it at once. TTheir IT team barely had to touch
 anything.

been running 3 weeks. zero maintenance, only additions. Your next move = one of these for every new engagement. first session, strategy. second session, live dashboard.

Crazy time to be alive...

(I can't code... built with Claude Code and Replit [.] com )

Europe before the “editing” part is pretty crazy.

Hundreds of princes, city-states, bishops etc etc all convinced they were permanent. Then somehow Germany happened. All of these turned into one country by 1871. Same with Italy. Same with France (just bloodier and slower).
Crazy part is how ‘normal ‘it feels now. kids learn “Germany” like it’s a fact of nature… 300 territories that got absorbed and for the most we simply forgot they existed. Collective fictions (Harrari coined this I think) work like that. Once enough people agree,l the rest just kinda disappears from memory…

I ran a LinkedIn masterclass yesterday.

here's everything we covered...

We spent an hour going through what we found after analyzing 60000+ LinkedIn posts. I'm sharing the full deck in the slides below. 35 slides. Screenshot them, save them, do whatever you want with them.

The short version is that a lot of what the gurus have been teaching doesn't hold up. We checked the actual data and most of the popular advice either overpromises or is flat out wrong. Some findings were obvious. But some genuinely surprised me.

The biggest one is that LinkedIn completely rewired its algorithm. They replaced five separate ranking systems with a single AI model called Brew 360. Your profile now acts as a classifier. It decides what interest bubble you belong to and only shows your content to people inside that bubble. Post off topic and the algorithm literally doesn't know who to send it to. Your profile is no longer just a landing page. It's a routing signal. This changes everything about how you should think about your headline, your about section, your top five skills. All of it.

On hooks, curiosity gap wins by a massive margin. 66% win rate in our dataset. The worst performer was list promises. "7 ways to..." had literally 0% win rate. Zero. That format is done. What works instead is specificity. Posts with exact numbers in the hook are twice as likely to perform well. Posts that start with "I" outperform by 38%. People connect with people, not with advice from a distance.

Longer posts perform better too. 1,250 to 3,000 characters, about 40% better than short ones. Counterintuitive. But the algorithm rewards dwell time. The longer someone stays on your post, the further it gets pushed. Break up the text with white space, write simply, keep it at a grade five to seven reading level. I hate that last one but the data is pretty clear.

Small accounts actually have a huge advantage right now. Under 1k followers? Your engagement rate is 400% higher than accounts with 100k+. The algorithm rewards genuine engagement over raw follower count. And the top 1% of creators reply to comments 741% more than average. That's 286 replies per week. Replying is not optional anymore.

The single biggest finding from the entire study though is that consistency beats everything. We saw a 149x difference between a creator's best post and their average. 80% of posts only generate 10% of total impressions. You can do everything right and some posts just won't land. Every post is a lottery ticket. You can't predict which one hits. You just need to keep buying them.

We built viralbrain [.] ai so that you don't have to remember all these tips... they're all cooked into our tool. Post consistently. Create posts that win.

Full deck is in the slides. It's directional, not gospel. But the patterns are real and they're consistent across thousands of posts.

Secrets Of The LinkedIn Algorithm Masterclass Feb 26

shadow AI at its finest…

Let’s go ahead and share ALL company data with Anthropic

Nobody asks what fractional AI consultants actually do all day.

so here's a random Monday from last week.

Morning. I'm inside a PE-backed manufacturing group that's been acquiring companies across Europe. Eleven brands in, more in the pipeline. Every acquisition comes with its own CRM, its own CMS, its own external providers burning budget with zero compounding value. My job: build a repeatable integration engine so brand #3 costs a fraction of brand #1. The playbook gets faster with every deal. Not a consulting project. An operating system.

Afternoon. B2B SaaS company, big GTM team and the headcount is shrinking. Not because of layoffs. Because of what we built together. Data sources that monitor movements and buying across their entire target market. Strategy generators that also produce campaign assets in minutes instead of weeks. Performance marketing campaigns with learning loops, that auto-correct. A "GroundControl" dashboard their VP now opens before coffee. The question isn't "how many AI tools do we have." It's "how many work units get processed per human."

Evening. I'm scoping a new engagement. A company that sells AI externally but can't figure out how to use it internally. Sound familiar? They've got ambassadors, workshops, a community of enthusiasts. But no system. No connective tissue between the experiments. Tomorrow I start building the framework that ties it all together.

Three different problems. Same methodology.

Find the worst process. Map it completely. Build the system. Transfer ownership. Leave. I don't stay. That's the whole point. Everything gets documented, all IP stays with the client. When their team is ready, I'm out.

Been doing this for about 18 months now. Mostly B2B, mostly enterprise, mostly teams that already tried "just buy ChatGPT Enterprise" and realized the tool wasn't the bottleneck. The thinking was. What surprises me is how similar the problems are across industries. A PE-backed industrial group and an enterprise SaaS sound completely different. But both had the same gap: smart people, good tools, no system connecting them.

I have bandwidth opening up end of Q2. And it's not just strategy. I work with a small team that builds and implements the systems we design. Architecture, automation, integrations. From diagnostic to production. If this sounds like your company.. happy to chat about what a diagnostic sprint looks like. No pitch deck. Just a conversation about where the bottleneck actually is.

15 people you should follow/connect with on LinkedIn.

Real practitioners doing exceptional work. Not mega influencers with ghostwriters..These people are shaping how I think (I'm extremely grateful).. and they can do the same for you.

Special shout out to Walid Boulanouar, my co-founder. The person I call when something doesn't make sense yet. Helps me live in the future. We build together. If you care about AI..you must follow him...

in no particular order

Luísa Lima, if you want to understand agentic security before it becomes everyone's problem. She ships faster than most teams. The ai architect and security brain you want in your corner.

Jan-Willem Bobbink, For me, he is the world's best SEO/GEO expert. Highly Practical, no hyperbole, all results..on what is still the world's most important channel. 

Damian Nomura if you need cool calm but highly effective ai adoption... He'll walk you through terminal workflows, explain how he does things and be honest about what he doesn't know. The kind of person who lifts everyone around him.

Roald Larsen, if you want someone who quietly knows more than most rooms he walks into. A builder, a product guy and a marketeer all in one. 

Shaun Thresher if you're building systems with Claude Code. The person who figures out the workflow so you don't have to.

Rhys Fisher if you're interested in agentic swarms and self-optimizing AI systems. He's building frameworks where tasks evolve and improve themselves. His work will be table stakes in 18 months.

Pascal Malengrez, if you care about safe agentic deployments and AI-as-a-Service models. He thinks about work re-organization at a level most people haven't caught up to yet. Cool, calm, composed.

Andreia Cândido Saraiva if you want to see what building with AI actually looks like day to day. Google Workspace CLIs, Slides design systems with design tokens, health simulations.

Chris Parker if you care about AI transformation in practice. He's leading a collaborative handbook built from real practitioner experience...we're releasing it soon ;)

Louise de Sadeleer her LinkedIn and marketing game are off the charts. 

Jean Bonnenfant for anything growth, LinkedIn or just overall growth. He's always been an inspiration for me. On anything LinkedIn and marketing.

Ross Stevenson if you care about learning and development that actually changes behavior. Not just check boxes. He studies what 81,000 people want from AI and turns it into something actionable.

Vladimir de Ziegler if you want strategic clarity and execution quality in the same person. Work with him...you just know the output will be exceptional.

Stas Pavlov ツ if you're always looking for the next tool or resource worth testing. He finds things before everyone else, asks the right questions and shares openly. Curiosity you can learn from.

They are exceptionally good. Follow them. Reach out.. thank me later... surround yourself with people better than you at the things that matter. 

who's missing?

text posts are 65% of LinkedIn. and they lose.

badly.

we analyzed over 1 million posts for ViralBrain. image posts get 2.4x more engagement than text. And they go viral at 3.3x the rate.

The crazy part? again... images are only 35% of all posts. most creators are competing in the most crowded, worst performing format on the platform.

Why does this happen?

LinkedIn's algorithm (BREW360) measures dwell time. how long someone stops scrolling on your content. images naturally hold attention longer. more dwell time = more distribution. they also trigger more "see more" clicks on mobile, which counts as extra engagement.

and polls? 49 average engagement. don't bother.

You don't need a designer. a simple screenshot with annotations already outperforms pure text. You can also use our image generator on viralbrain [.] ai ... anything visual puts you in the top 35%.

(sample: 362,800 image posts vs 653,300 text posts. clean, API defined categories. no guesswork.)

I'm not saying stop writing. i'm saying stop posting naked text when you could spend 2 extra minutes adding a visual. the data is pretty clear on this one.

last tip: vertical images 1200x1500 work best, why? Because they grab more real estate on mobile.

a tale of two models…

ps: nothing says ‘national security’ like a model that makes things up with full confidence

taplio

David Arnoux's Linkedin Analytics

David Arnoux

Performance ScoreDavid Arnoux's LinkedIn Performance

What is David talking about?

David Arnoux's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}David Arnoux's Linkedin Analytics

David Arnoux

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}David Arnoux's LinkedIn Performance

What is David talking about?

David Arnoux's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

David Arnoux's Linkedin Analytics

Performance ScoreDavid Arnoux's LinkedIn Performance

David Arnoux's Best Posts (last 30 days)