How Do We Put the Human at the Center of AI?

SECTION 1

Introduction

The hype around AI, generative, agentic or otherwise, is such that it's easy to lose sight of the fundamental purpose of this technology: helping people. In reality, that's the purpose of all technology, to improve people's lives.

This is a point that expert Tiankai Feng, Global Head of Data and AI Strategy, is keen to stress to the business world. In July 2024, he published his first book, Humanizing Data Strategy, which explores how to approach and leverage data by putting people first.

Just over a year later, he wrote a sequel: Humanizing AI Strategy, addressing how to design and implement an AI strategy that is sustainable, effective and fundamentally human. We've pulled together the core of his thinking into an operational lens that any organization running generative and agentic AI programs today can put to use directly.

SECTION 2

Why Humanize AI Strategy: A Question of Priority, Not Technology

### What a "Non-Humanized" AI Strategy Looks Like

A humanized AI strategy, according to Tiankai Feng, means insisting on prioritizing human values and needs, in contrast to an approach that only considers technological advances. A non-humanized AI strategy is one that fails to account for the implications of how employees will interact with AI tools, how AI should or shouldn't be used for customer interactions, and the potential positive and negative impacts it could have on society. Ignoring these questions means risking harm that will be very difficult to undo once a system has been deployed at scale.

### Lessons From the First Book, Humanizing Data Strategy

Building on Humanizing Data Strategy, Feng advocates for learning from past mistakes and experiences in data management, and for proactively putting principles and guardrails in place before an incident makes them unavoidable. When it comes to AI, he wants everyone working on AI initiatives within organizations to ask not just "can we do it?" but also "should we do it?" and "how can we do it properly?" That deceptively simple triple question actually structures his entire method.

SECTION 3

Why AI Projects Fail: Speed Versus Diligence

The AI project failures Tiankai Feng examined all share one thing in common: they prioritized speed over diligence and guardrails. He identifies three starting points for preventing further failures.

### The Right Human at the Right Time

This means building in the concept of the "right human at the right time": being rigorous about setting up trigger points and decision nodes to determine when human-in-the-loop involvement is mandatory, particularly around regulatory, ethical and business risks. In an age of autonomous agents capable of chaining multiple actions without direct supervision, this principle becomes an architectural requirement rather than just a best practice.

### Multidisciplinary Expertise From the Start

AI experts don't have exhaustive knowledge of every risk surrounding AI applications. Bringing in experts from legal, diversity and inclusion, cybersecurity and other relevant functions early in the development process helps identify these risks quickly, before they become costly to fix, and helps design appropriate guardrails during the scoping phase rather than in reaction to an incident.

### Testing as Rigorous as a Pentest

Just like penetration testing in cybersecurity, more rigorous testing to detect malicious intent and drifting behavior within an AI application helps identify vulnerabilities and prevent real-world failures before it's too late. For agentic systems going into production in 2026, that means regular red-teaming campaigns, adversarial test sets, and incident-simulation scenarios, on par with a standard application security audit.

@cite:les-sept-peches-capitaux-de-la-transformation-ia-lecons

SECTION 4

The 5C Framework: Rooting Strategy in Human Needs

At the heart of Tiankai Feng's approach to a humanized AI strategy is the 5C framework, all rooted in human needs and traits: competence, collaboration, communication, creativity and consciousness. Competence refers to a team's ability to genuinely understand what an AI system is doing, beyond its interface. Collaboration describes how humans and AI agents divide up tasks without either erasing the other's contribution. Communication covers transparency about automated decisions and their limits, toward employees as much as toward citizens or end customers. Creativity is a reminder that AI should expand the space of possible solutions rather than over-standardize it. Consciousness, finally, is the collective ability to anticipate the consequences of a deployment before they materialize.

If an organization focuses on these five aspects throughout an AI application's lifecycle, scoping, design, deployment, operation, retirement, it can hope to become more human-centered intuitively and durably, rather than at the cost of a permanent, expensive governance effort.

SECTION 5

Staying the Course Without Losing Agility: A Directional, Not Instructional, Strategy

### Vision From Leadership, Autonomy on the Ground

For Tiankai Feng, any technology-related strategy needs to be directional rather than instructional: a clear target picture should be provided in terms of business drivers, operating model, architectures and processes, but with enough built-in flexibility to adapt to technological, industry or societal changes. It also depends on a cultural element where every detailed decision doesn't need to be made top-down: leadership provides the vision and direction, while the people on the ground, who best understand the operational realities, operationalize it by making autonomous decisions within the defined framework.

This logic connects directly to the scaling challenges organizations face today as they move from exploratory AI use to enterprise-wide deployments, where the sheer number of use cases, teams, and agents involved makes centralized, detail-level control impossible.

@cite:naviguer-le-scaling-de-l-ia

SECTION 6

The "No AI at All" Argument: Acknowledging the Risks Without Giving Up the Benefits

Some argue that the only truly humanized AI strategy is no AI at all: the impact on jobs, creativity, and the environment all make AI contrary to human interests. Tiankai Feng responds that, as with any technological advance, we need to maintain a healthy, balanced relationship with it. Despite the real risks to jobs, intellectual property, or the energy footprint of large models, we also need to recognize the benefits AI can bring: greater advances in medical research, more personalized and impactful education, or reducing repetitive manual tasks. The future he envisions is one where humans and AI coexist and collaborate, each strengthening the other's capabilities rather than purely replacing them.

SECTION 7

Writing With AI: What Tiankai Feng's Experience Reveals About Human-AI Collaboration

One detail illustrates the book's thesis particularly well. Tiankai Feng wrote his first book entirely without AI assistance, but decided to use generative AI to help him write the second one, essentially modeling, in his own words, good collaboration with AI. The tool helped him structure his thoughts, critically examine his content, and overcome writer's block at certain points, without ever replacing his voice or his final editorial judgment.

This personal experience concretely illustrates the 5C framework: the competence to know when to call on the tool, the collaboration between author and assistant, the communication of a result that stayed true to his thinking, the creativity preserved rather than standardized, and the consciousness of the limits of what AI could reasonably contribute. His takeaway is simple: the clarity of his thought and his writing style improved compared to the first book, and learning, he points out, remains one of the best human traits.

SECTION 8

Conclusion: What This Means in Practice for Organizations

Humanizing AI Strategy is written for anyone who works with AI or is interested in working with it. Decision-makers will find actionable guidance to frame their programs; teams working with AI day to day will find ways to make conscious choices about how to collaborate with it, and useful feedback to feed back to those who build AI applications.

For an organization shaping its AI governance today, the operational translation of these principles comes down to a few simple habits: explicitly documenting the decision points where a human must remain in the loop, involving legal, security and business functions during scoping rather than at final sign-off, testing agentic systems the way you'd test an attack surface, and regularly measuring the five dimensions of the 5C framework rather than adoption rate alone.

@cite:s-organiser-pour-l-ia-de-l-experimentation

Thanks to Tiankai Feng for taking the time to share his thinking. Learn more about Humanizing AI Strategy.

Disclaimer: The statements and opinions expressed in this article are those of the author(s) and do not necessarily reflect the positions of Adservio.

FAQ

Frequently asked questions