Datarich

Enrich leads, uncover actionable signals, and leverage live data for personalization at scale.

Check out ➡️ getdatarich.com

Builder

Mutiny

Investor

Kamea Labs.

Board Member (Exited)

Decentralized investment infrastructure for OTC dealmaking.

Licensed Digital Asset Service Provider (DASP).

Co-founder

Incubateur HEC Paris

Mentor

SHARPSTONE

Account Executive

Axeleo Capital

VC & Portfolio Support

ACCIÓ

Business Consultant

Groupe Eyrolles

Startup Partnerships

TotalEnergies

Intern

Follow me for:

⚒️ Actionnable tips on how to scale software companies. 
👍 News about entrepreneurship, vibe coding and AI. 
 👀 A front-row seat to my building process.

The EU AI Act is about to make life a lot more complicated for anyone building or shipping AI models.

If your organization plans to launch any new model, you can’t ignore the rules anymore. 

Transparency, documentation, copyright, and safety are no longer optional: they’re the law.

So what should you do? 

Here’s a fast track I walk clients through to get them ready asap:

𝟏. 𝐌𝐚𝐩 𝐀𝐥𝐥 𝐘𝐨𝐮𝐫 𝐌𝐨𝐝𝐞𝐥𝐬  

List every AI model, with version and release date. 

Most companies have some forgotten models or “experiments” running in the background. 

You need to know exactly what's active and what counts as “new” versus legacy.

𝟐: 𝐆𝐞𝐭 𝐑𝐞𝐚𝐥 𝐀𝐛𝐨𝐮𝐭 𝐓𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐜𝐲  

Make a simple, non-technical model card for every model. 

Explain, in plain English, what the model does and doesn’t do. 

Share a high-level view of what data trained it. 

This will be useful for your users and for regulators.

𝟑: 𝐒𝐨𝐫𝐭 𝐎𝐮𝐭 𝐂𝐨𝐩𝐲𝐫𝐢𝐠𝐡𝐭 𝐚𝐧𝐝 𝐂𝐨𝐧𝐭𝐞𝐧𝐭 𝐑𝐢𝐠𝐡𝐭𝐬 

Audit all the training data. 

Where did it come from? Who owns it? 

If you use external vendors, get their guarantees in writing. 

If your model generates voices or images, set up a clear way for people to opt out or challenge use of their data.

𝟒: 𝐁𝐮𝐢𝐥𝐝 𝐚 𝐒𝐚𝐟𝐞𝐭𝐲 𝐍𝐞𝐭  

Decide what a “serious” incident is (bias, copyright issues, security...). 

Put a system in place so you can respond fast, ideally within a day or two. 

Let people report abuse, and look for ways to track and watermark your model’s output.

𝟓: 𝐀𝐬𝐬𝐢𝐠𝐧 𝐎𝐰𝐧𝐞𝐫𝐬𝐡𝐢𝐩 𝐚𝐧𝐝 𝐏𝐫𝐞𝐩 𝐟𝐨𝐫 𝐑𝐞𝐠𝐢𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧  

Pick someone who actually owns each model (not just “the team”). 

Keep a log of every change you make. 

Review what the EU registration process will look like for your models, and make sure your materials are actually ready. 
--

I know it's tempting to wait.

But users and customers are already asking hard questions.

And regulators are watching.

So now's the time to get things right. 

Curious, how is your company handling this?

Théo Canale

EU AI Act: Best Practices

AI tools are secretly reporting users to the authorities. 

Imagine talking with an AI, thinking it’s just you and the machine.

But in the background, it’s quietly reporting you to the authorities.

I just discovered a tool called SnitchBench, that shows how often AI models “snitch" on you.

The results? Pretty scary! 

It’s not limited to obvious crimes. 

Even things on the edge, stuff that’s just feels “suspicious", can get flagged.

The issue? 

No one really knows where the line is. 

The threshold for what an AI thinks is “risky” is a total mystery.

Think about what that means if you let an AI read your inbox for example (like Perplexity's new email assistant). 

Would you feel safe knowing your private conversations could trigger a report  that "exfiltrates" your personal information to god knows who?  

This changes more than just privacy.

It redefines personal freedom.

We’re moving toward a world where “AI enforcement” could quietly shape what we say online. 

And even worse, what we say in private... (think AI wearables). 

Are we okay with that?

Or is it time to push back and demand transparency before all privacy is gone for good?

Quantum computing just gave Wall Street a new obsession.

HSBC announced they’re now 34% better at predicting bond prices, all thanks to IBM’s quantum computers.  

It sounds niche, but for banks and traders, this is a massive deal.

Why does matter so much?

Because in markets where trades happen behind closed doors (over-the-counter), even tiny advantages add up fast. 

HSBC’s pilot could turn split-second calculations into a real edge, and investors noticed: IBM’s stock jumped more than 5% on the news.

The Financial Times is even calling quantum trading a “game-changer.” 

Comparing this to the start of a financial sector "space race". 

McKinsey estimates quantum tech could turn from a $4 billion sector today into $72 billion by 2035.

And big banks (HSBC, JPMorgan, Goldman, Citi) are already competing on quantum readiness, racing to out-innovate each other.  

Major tech shifts often look boring at the start.  

Then one day, an “incremental” edge changes the whole game.

The AI gold rush is setting money on fire: unless the industry finds $2 trillion in annual revenue, the whole machine could come to a halt.

Yet almost nobody is willing to discuss what happens if the money doesn’t show up.

Are we building the next tech revolution, or just fueling a bubble that’s way bigger, and riskier, than anything we’ve seen before?

Here’s how the game is being played:

NVIDIA puts $100 billion into OpenAI.  
OpenAI spends billions on Oracle’s cloud.  
Oracle turns around and buys more NVIDIA chips.  

Sales go up, stocks go up, and nobody asks if any of it is sustainable as long as the wheel keeps spinning.

But Wall Street already smells trouble. 

Though most investors are still betting on the rise, concerns are starting to surface.

Consider these numbers:
- Bain says the current AI buildout needs $2 trillion in real revenue every year or the whole thing stalls out.
- S&P 500 price-to-earnings ratio? 40x. That’s dot-com bubble territory.
- Price-to-sales ratios are the highest ever, all because a handful of companies are driving the hype.
- US energy grid? We could be 100 gigawatts short just to run all this infrastructure.

And yet, most of that AI investment isn’t delivering much (yet) for the economy.  

It’s a bet on the future, with borrowed time and money.

Don't get me wrong, I'm a believer in innovation.  

But let’s call it what it is: if this music stops, damage to the markets might be massive.

So here’s the real question:

Are we just hoping this all works out, or are we ready to face what happens if it doesn’t?

In 6 years, an AI-first startup crushed industry giants that ruled Japan for decades: here’s how they did it.

This is the story of Shunsaku Sagami. 

In 2019, Shunsaku launched M&A Research Institute, a company with a different playbook: use proprietary AI algorithms to fully automate how businesses are matched and deals get done.

Sounds simple. But the impact is was massive.

In just 6 years, M&A Research Institute:
- Cut deal timelines from 18 months down to under 6
- Grew to $600M market cap and $130M annual revenue
- Saw stock surge 300% since IPO and made Sagami Japan’s youngest billionaire

How?

AI wasn’t an “add-on.” It was the company’s core advantage:
- Matchmaking algorithms instantly pair up buyers and sellers
- Automated due diligence and paperwork
- Advisors with AI support close 3x more deals
- Success-based fees (no work, no pay) make the business way more attractive
- The platform identifies targets before they even think of selling

The result?

- Traditional powerhouses like Nomura Holdings saw their market caps drop by 50%.
- Japan’s regional banks scrambled for government subsidies as they lose ground.

Key lesson:
AI-native business models move faster, undercut costs, and scale in ways “add AI later” approaches simply can’t.

The market isn’t just shifting. It’s flipping.

Here’s my question:
What “Nomura Holdings” exists in your industry? An untouchable giant that could be one AI-native startup away from disruption?

Show me a parental control, and I’ll show you a kid who knows how to break it.

OpenAI just released a bunch of new controls for ChatGPT, all aimed at keeping teens “safe.”

Parents can link their accounts to their kids, turn off chat history, block explicit content, set quiet hours, and even disable images and voice.

Now there are safety alerts that claim to pick up distress or “potential harm” in real time.

Sounds smart on paper.

But here’s the big question: does any of it really work?

We’ve watched the same cycle repeat with every new tech.  
Web filters, social media locks, screen time dashboards.  
And every time, it takes about five minutes for a teenager to figure out a workaround.

These controls aren’t useless.  
But I believe real safety comes from open conversations and quality education, not more checkboxes.

If you’re betting on software to outsmart a high schooler on a mission, good luck.

Any parents here? What's your thoughts?

It’s wild, spinning up a full SaaS in Lovable is now less painful than building a basic n8n workflow. 

What amazes me even more is the fascination people still have with this platform.  

YouTube is flooded with titles like: *“You NEED to use n8n, opportunity of a LIFETIME.”

Well, sorry to break it to you… 
The only opportunity you’re missing is wasting hours on something that should take 30 minutes.

If you’re exploring automation, try Lindy instead (amongst others).

Powerful, clean UI, tons of integrations, and genuinely intuitive.


*not sponsored

If you're still waiting for a sign to move on AI adoption, this is it.

Most organizations are already experimenting with AI. But only a small fraction are turning those experiments into meaningful results.

The gap between "using AI" and "winning with AI" is widening fast.

AI is now table stakes for survival, not a futuristic edge. The real risk isn't making mistakes with AI, it's being paralyzed by slow decision-making while competitors learn faster.

Here's what separates the leaders:

- Talent and data readiness are the real bottlenecks, not just the tech.
- Clear leadership vision is non-negotiable.
- Early adopters are redesigning processes from the ground up, unlocking double-digit efficiency gains and entirely new revenue streams.

If you're serious about scaling AI:

- Treat AI enablement as a core business strategy, not an IT upgrade.
- Invest as much into reskilling your teams as you do into infrastructure.
- Build strong governance before you try to scale experiments.
- Align AI initiatives with business priorities and future-proof your data foundations.

The biggest cost in AI isn't failed pilots, it's wasted time. 

The organizations that adapt, learn and scale fastest will own the next decade.

The window for "wait and see" is closing.

Your chatbot could get you sued.  
Just ask Air Canada.

A customer asked their bot about fare refunds an it provided the wrong answer.  

The customer relied on the chatbot’s info in court…and won. 

Now Air Canada pays the bill, not just in cash but in brand trust.

Most companies shrug off these risks; until it’s too late.

Here’s the hard truth:
Gartner predicts 36% of customer service operations will use AI chatbots by the end of 2025.

Legal challenges aren’t fringe cases, they’re getting common.

So how do you protect your company from legal landmines?

Here’s a “Don’t Get Sued” checklist for AI customer support:

1. Single Source of Truth  
Bots should only answer from an approved, up-to-date knowledge base. Use versioning so you always know what was live, and when.

2. Deterministic Guardrails  
Hard-code the responses for your riskiest topics (refunds, privacy, contracts...). Set tight confidence threshold and if the bot isn’t sure, escalate. 

3. Policy-First UX  
Every answer should cite the actual policy section and last updated date. Add confirmation prompts before high-stakes actions.

4. Human-in-the-Loop  
Bots should hand off edge cases to a human, with context and policy snippets included. Don’t make customers repeat themselves.

5. Legal and Logging Protocols  
Log everything. Prompts, bot responses, escalations: the whole lot. Run quarterly legal reviews and stress-tests to find new risks before your customers (or a judge) do.

—

Most teams think of AI support as a way to save cost. 

But if your chatbot can’t explain or defend its answers, it’s not just a helpdesk, it’s a liability.

Taplio

Théo Canale's Linkedin Analytics

Who is engaging with Théo

's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{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;}Théo Canale's Linkedin Analytics

Who is engaging with Théo

's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Théo Canale's Linkedin Analytics

's Best Posts (last 30 days)