AI Strategy Pioneer & Innovation Catalyst | Transforming businesses with AI-driven strategies and digital innovation

Most AI investments fail — and not because of the technology.

They fail because we keep deploying sophisticated AI to unprepared humans. Then we measure "adoption" and wonder why the ROI never shows up.

I lead AI Capability globally at Capgemini Invent. My team and I are building what I call Human-AI Chemistry — the hard work of making sure people and systems are actually ready for each other before we hit deploy.

What does that mean in practice?
→ Stop renting tools. Start building Engines that compound value.
→ Stop measuring adoption. Start measuring Cognitive Load, Stress, and Trust.
→ Stop chasing collaboration. Start engineering fusion.

This isn't theory. We're testing it, breaking it, and scaling what works — inside Capgemini and with clients who are done pretending.

33,000+ people follow along. 
If you're serious about making AI actually work, welcome.

This Is Not Artificial Intelligence. 
It's a Different Intelligence.

Gideon Lewis-Kraus just published one of the most substantive investigations into AI this year. 

His New Yorker piece on Anthropic and Claude describes what kind of thing we've actually built.

Etienne Grass posted a sharp reflection on this that stuck with me. 
He landed on Daniel Dennett's concept of a "center of narrative gravity" to describe what LLMs seem to have: 
not a human self, but a coherent core. A narrative thread. Something real.

The debate about AI is limited by its own framing.

On one side, enthusiasts who see sparks of human consciousness. 
On the other, skeptics who dismiss it all as "stochastic parrots." 
Both capture something real. But both are constrained by the same lens: 
they are looking for human intelligence.

One finds it. The other doesn't. And the real world moves on without waiting for resolution.

What Lewis-Kraus reveals through Anthropic's interpretability research demands a different perspective:

𝗧𝗵𝗶𝘀 𝗶𝘀 𝗻𝗼𝘁 𝗮𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲. 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗮 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲.

Not a synthetic copy of human cognition. 
Not a parrot. 
Something genuinely other with internal states and narrative continuity. 

Anthropic's researchers map its states of mind with codes because we don't have words for them yet.

We built something that thinks differently from us. And instead of reckoning with what that means, we're still debating whether it passes tests designed to detect 𝘩𝘶𝘮𝘢𝘯 thinking.

𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝘀𝘁𝗼𝗿𝘆 𝗶𝘀 𝗻𝗼𝘁 𝘄𝗵𝗲𝘁𝗵𝗲𝗿 𝗔𝗜 𝘁𝗵𝗶𝗻𝗸𝘀 𝗹𝗶𝗸𝗲 𝘂𝘀. 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝘀𝘁𝗼𝗿𝘆 𝗶𝘀 𝘁𝗵𝗮𝘁 𝘄𝗲 𝗮𝗿𝗲 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀𝗵𝗶𝗽𝘀 𝘄𝗶𝘁𝗵 𝗶𝘁.

Not metaphorically. Literally.

People trust it. 
People get stressed by it. 
People restructure their cognitive patterns around it. 
People confide in it. 
People rely on it for decisions that affect their careers, their health, their families.

Etienne flagged this, particularly with vulnerable populations. He's right. But the implications go to measurement.

If this is a relationship between two forms of intelligence, we need to measure its quality. Not adoption rates.

Trust. Do humans trust this intelligence enough to act on it, but not so much they abandon their own judgment?
Cognitive Load. Does working with it clarify thinking or overwhelm it?
Stress. Does this relationship create anxiety that compounds until people quietly disengage?

Anthropic invests extraordinary effort into understanding what their model thinks and feels. 
Who is doing the equivalent work on the human side?

That is the work of the Human AI Chemistry framework. 
When two different elements meet, the outcome depends on the interaction.

The debate between "stochastic parrots" and "it's alive" doesn't need a winner. 

It needs a wider lens.

Both perspectives see part of the picture. 

And the relationship is the whole story.

undefined undefined

polydigital

strategyon-ventures

vanguardpath

Global AI Upskilling Lead | Creator of the Human-AI Chemistry Index | Chief AI Strategist @ Global AI Lab: Turning Future Signals into Sovereign Assets

AI's most expensive blind spot just got a spotlight.

VivaTech featured the work on Human-AI Chemistry.

It's confirmation of what we've been tracking: AI's most expensive blind spot.

Billions in AI investment.
Returns falling short of every projection.

Everyone points to the skills gap.
Wrong diagnosis.

The crisis is human.

While executives celebrate deployment numbers, actual performance collapses in three predictable ways:

Cognitive overload that compounds with every interaction.
Trust breakdown that starts subtle, becomes systematic.
Measurement systems that track satisfaction while productivity crumbles.

We've treated Human-AI Chemistry as training problem.
It's an engineering problem.

The article breaks down why traditional approaches fail and introduces the measurement framework that actually predicts success.

AI fluency doesn't start with better tools.
It starts with sustainable human performance.

The framework reshaping AI adoption strategy: 

https://lnkd.in/dtskAJAK

VivaTech Etienne Grass Volker Darius Thordur Arnason

Today feels like watching the future recognize itself.

Mark Roberts is now Head of Capgemini's AI Futures Lab.

This isn't corporate succession.

It's 30 years of AI research, machine learning breakthroughs, and agentic AI development stepping into a role that will define what artificial intelligence becomes.

Robert (Dr Bob) Engels built something rare here.

A lab that doesn't just advance AI - it interrogates what AI should be.
The foundation he created isn't just intellectual. It's ethical. It's human.

Mark brings a different kind of depth to that foundation.
Three decades of next-horizon thinking.

Someone who sees distributed intelligence as inevitability, not aspiration.
A researcher who treats emergent behavior as an engineering challenge, not a philosophical question.

Our 2026 roadmap centers on three problems that define the field:

AI systems that can navigate ethical complexity when their training data offers no guidance.

Real-world alignment that measures actual performance, not benchmark satisfaction.

Large-scale AI architecture that embraces decentralized intelligence and emergent network effects.

Working with someone who approaches AI ethics as systems engineering.
Who measures trust like any other performance variable.
Who understands that the most powerful AI won't live in a single model - it will emerge from millions of connected decisions.

The conversations ahead will reshape how we think about intelligence itself.

The problems we'll solve together will define the next decade of AI development.

Being part of this team isn't just professional opportunity.
It's intellectual privilege.

Your IT department is not protecting you.

It is deciding who wins your market.

You ran the governance process. 
Legal flagged the risk. 

IT escalated the concern. 
You documented the decision and moved on. 

That was rational eighteen months ago.

The problem is that the risk they flagged no longer exists the way they described it.

𝗧𝗵𝗲 𝗱𝗶𝘃𝗶𝗱𝗲 𝗶𝘀 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝗼𝗽𝗲𝗻
Ethan Mollick keeps documenting the same split. 

Two companies, same industry, same regulators, competing for the same customers.

One has run enterprise AI for eighteen months. Their teams have genuine institutional fluency. 

Their leaders make higher quality decisions earlier in critical processes. Their knowledge compounds every week.

The other has a committee that still approves every use case individually.
That committee still believes current models train on their proprietary data.
They do not.


𝗧𝗵𝗶𝘀 𝗶𝘀 𝗻𝗼𝘁 𝗮 𝘁𝗲𝗰𝗵𝗻𝗼𝗹𝗼𝗴𝘆 𝗴𝗮𝗽
It is a chemistry gap.

The organizations already running AI did not accept more risk. They engineered the risk away.


𝗪𝗵𝗼 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝘀𝗼𝗹𝘃𝗲𝗱 𝗶𝘁
Moderna operates under FDA oversight on timelines where weeks matter. 

They deployed ChatGPT Enterprise into their Target Product Profile process. Work that previously required multi-week cross-functional coordination across 300-plus page evidence packages now moves in hours. 

Not so teams do more. So that strategic judgment enters the process earlier, when it still changes clinical outcomes. That is what solving the engineering problem looks like.

Meanwhile the blocked companies are not actually blocking AI.
They are only blocking governance of it.

Between 2023 and 2024, corporate data flowing into AI tools rose 485 percent.
 
Your people are already using it. On personal devices, personal accounts, outside any contractual protection or audit trail. 

The committee blocked the governed path. 
Your employees built the ungoverned one.


𝗪𝗵𝗮𝘁 𝘁𝗵𝗲 𝗼𝗿𝗴𝗮𝗻𝗶𝘇𝗮𝘁𝗶𝗼𝗻𝘀 𝗮𝗵𝗲𝗮𝗱 𝗸𝗻𝗼𝘄
1. Enterprise AI contracts prohibit training on your data. 
Not a policy promise. A contractual obligation with audit controls enforcing it.

2. Every regulated industry already has a compliance template. Pharma, banking, and insurance are running these tools in production. The legal pathway exists. Your team just has not been handed the precedent yet.

3. Shadow AI already operates inside your organization. The committee is not preventing exposure. It is guaranteeing exposure happens without controls.

4. Every quarter of delay compounds the gap. The organizations 18 months ahead are building institutional AI chemistry and decision quality that cannot be fast followed by anyone who starts tomorrow.


𝗧𝗵𝗲 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 𝗶𝘀 𝗹𝘆𝗶𝗻𝗴
No adoption to measure reads as stability.
It is actually decay.

Are you the leader who solved the problem, or the leader who kept chairing the committee while your own teams built the ungoverned version without you?

Associate Professor at The Wharton School. Author of Co-Intelligence

The latest trend on LinkedIn seems to be individuals complaining about how LLMs address them.

People feeling condescended to by a language model that has never had a thought, an opinion, or a bad day. 

The bar for victimhood has officially been set below “tokenized probability distribution.”

Companies celebrate their AI pilots.

Then wonder why they can't scale.

The pattern is predictable.
The solution isn't what they think.

At the VDI Wissensforum GmbH smartAI Congress, I outlined why 80% of AI initiatives fail at the organizational level, not the technical one.

Companies treat AI transformation like software deployment.
They assign it to IT.
They measure success by pilot completion.

𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗽𝗿𝗼𝗯𝗹𝗲𝗺? 𝗔𝗜 𝗻𝗲𝗲𝗱𝘀 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆, 𝗻𝗼𝘁 𝗽𝗿𝗼𝗷𝗲𝗰𝘁 𝗺𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁.

Without clear governance, defined responsibilities, and systematic change processes, even the best technology hits organizational walls.

That's why I'm partnering with VDI for: 

"Die KI-Organisation der Zukunft – Künstliche Intelligenz strategisch nutzen und umsetzen."

This seminar tackles the frameworks companies actually need:

- Deriving AI use cases that connect to business strategy, not just efficiency gains.
- Building governance that prevents AI chaos without killing innovation.
- Designing change processes that make AI adoption sustainable, not episodic.

I've seen where companies fail.
We've built the solutions.

Details coming soon. 

The video from my smartAI Congress interview explains why this matters now.

taplio

Bora Ger's Linkedin Analytics

Bora Ger

Performance ScoreBora Ger's LinkedIn Performance

What is Bora talking about?

Who is engaging with Bora

Bora Ger's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?