A review of Tiankai Feng’s Humanizing AI Strategy – a book about AI strategy that gradually reveals itself to be about something deeper: how responsibility moves, fragments, and sometimes disappears when systems begin to act on our behalf.

Tiankai Feng is one of those genuinely good people you occasionally meet in this field. He’s thoughtful, curious, and genuinely interested in exchanging ideas rather than just broadcasting them. He’s also someone with whom you can disagree without the conversation becoming hostile, which, in the current state of AI discussions, makes him rare. That matters more than it sounds.

There’s an optimism that runs through Feng’s work. You can feel it in his writing, on LinkedIn, and in conversation. I’ve always appreciated this quality in him. It’s not a manic, hyped-up optimism, but rather a more grounded belief that organizations can choose to act responsibly and that people usually want to do the right thing if given the right structures. I don’t always share the same level of optimism, which is probably why his optimism has such an impact on me. It gives me a boost every now and then.

So yes, I came to Humanizing AI Strategy with some skepticism. After all, it’s just another AI book. Another one promising to “put humans at the center.” I’ve read enough of those to know how easily human-centric language can become mere ornamentation – something we say after the architecture is already fixed. But my skepticism didn’t last long.

I had a similar experience when I read Feng’s previous book, Humanizing Data Strategy. What struck me then, and what strikes me again now, is that Feng doesn’t treat “the human” as a soft counterweight to technology. He treats humanity as the terrain that strategy must navigate. In the data book, this idea was expressed through stories of alienation, translation, and learning to bridge worlds. In the AI book, the lens has shifted, but the logic remains the same: systems fail when humans are absent, misaligned, or silenced.

What Feng does particularly well in Humanizing AI Strategy is to relocate responsibility. When AI fails, he argues, it is almost never because the math was wrong. Rather, it fails because humans were rushed, overconfident, poorly coordinated, or unwilling to ask uncomfortable questions. This framing matters. It resists the tendency to conveniently blame the model or the technology, instead insisting that strategy is a human act, even when execution is automated.

The book’s backbone is the 5Cs framework; Competence, Collaboration, Communication, Creativity, and Conscience. On paper, that could easily collapse into management shorthand. In practice, however, it works because it mirrors how organizations actually break down. Not due to a single bad system, but due to literacy gaps, siloed ownership, vague narratives, sameness disguised as efficiency, and moral ambiguity hidden behind speed. Feng is especially strong on competence, not as tooling proficiency but as judgment. AI literacy here isn’t about knowing how to prompt. It’s about knowing when to trust, when to question, and when to stop.

There is a clear throughline from the data book here. Where Humanizing Data Strategy focused on bridging gaps between technical systems and human meaning, Humanizing AI Strategy raises the stakes by acknowledging that AI does not just inform decisions, but rather, it acts. And once systems act, the cost of human absence rises sharply. Feng’s insistence on “the right human at the right time” is not a compliance mantra, but an architectural one.

I also appreciated what the book refuses to do. There’s no moral panic, no apocalyptic framing, and no GenAI hype. Generative models are treated as powerful, unstable, and context-sensitive, never as something magical. Traditional AI isn’t dismissed as obsolete. Human-in-the-Loop isn’t reduced to a checkbox. This is practitioner thinking, written by someone who has seen strategies fail for very human reasons: miscommunication, incentives pulling in different directions, and confidence outrunning competence.

If I have a point of friction, it’s a productive one. “Human-centric” can sometimes slide into the language of alignment rather than obligation. Conscience, as Feng presents it, is thoughtful and pragmatic, but I occasionally wanted it sharpened further. Less about balance, more about limits. Less about values, more about duties. That isn’t a flaw so much as a philosophical difference. Feng is building something organizations can realistically adopt. I am often more interested in what organizations must not do, even when it is efficient or profitable. Those perspectives are adjacent, not opposed.

Still, this is exactly the kind of book many AI leaders need right now. Not a technical manual. Not a regulatory checklist. And not an abstract ethics treatise. It’s a companion for people who sense that something important is at stake, but are tired of being told the answer is either “move faster” or “add more guardrails.

Like Humanizing Data Strategy, this book ultimately makes a quiet but firm claim: technology does not need to become more human. Humans need to remain present. Present in decisions. Present in language. Present in responsibility. In an age where fluent systems increasingly speak for us, that reminder is neither soft nor sentimental. It is strategic.

A review of Ole Olesen-Bagneux’s Fundamentals of Metadata Management – a book about metadata that gradually reveals itself to be about something deeper: how organizations try, and often fail, to understand themselves.

As I began reading Fundamentals of Metadata Management, I didn’t expect to see myself in it. I thought it would be one of those professional books I read: pen in hand, nodding at familiar frameworks, underlining useful distinctions, and then moving on. But a few pages in, I felt that familiar discomfort when a book doesn’t introduce something new, but rather points out something you’ve already been experiencing.

Large organizations are often clouded by a strange, polite uncertainty. The lists of systems that no one quite dares to shut down. The outdated diagrams that everyone still uses. Conversations that circle around “the landscape” as if it were a stable object when, in reality, it keeps shifting depending on who is describing it.

Ole Olesen-Bagneux calls this metadata. He does not mean “data about data” in the narrow sense, but rather the language organizations use to describe themselves, including their systems, processes, risks, ownership structures, rules, and knowledge. Reading this, I found myself thinking, “Yes, that’s exactly it.“ This is not a technical layer. It’s how an organization knows what it is doing. Or thinks it does.

What stayed with me most was his insistence that missing documentation was not the problem. Rather, it is the excessive amount of documentation produced in isolation. As he explained IT management, data management, information management, and knowledge management, I kept mentally mapping them onto places I’ve worked. There’s the CMDB that claims to know the infrastructure. The enterprise architecture tool that claims to understand applications. The data catalog that claims to know the data. There are also security registers, records systems, learning platforms, and quality systems. Each is maintained by competent people. Each is internally coherent. Yet, each conflicts with the others. Individually, they appear orderly. Together, however, they form something closer to a polite confusion.

I realized how often I have done this without naming it. I ask, “What systems do we have?“ and receive different answers depending on who I ask. Trusting one repository in one meeting and another in the next. I speak about “single sources of truth“ while knowing that, practically, the truth is always plural. I should know. Truth is always subjective. This book provides language for that experience in relation to metadata.

Ole particularly excels at demonstrating that this fragmentation is the result of structure, not a failure of intelligence or effort. Teams form around responsibilities, tools form around teams, and definitions form around tools. Before long, reality itself becomes segmented. It’s like Conway’s Law applied to knowledge. At some point, I stopped reading it as a book about metadata and started reading it as a book about organizations trying to understand themselves using instruments that were never designed for that purpose.

His proposal of a data discovery team seemed oddly delicate to me. It felt almost old-fashioned, borrowing more from the reference librarian than the data engineer. Ole envisions a group whose task is not to declare what is true, but rather to navigate between partial truths. They would know where descriptions exist, how they differ, where they are inaccurate, and how they can be combined without pretending they are the same. It felt less like an architectural solution and more like an ethical stance: humility toward complexity.

The same is true of the Meta Grid. I expected another type of platform, or another box in the system diagram. Instead, he describes something closer to a pattern more than a product. It’s a structure that already exists, latent, between existing repositories. It’s something to uncover rather than create. This idea was surprisingly convincing to me, not because it promises control, but because it doesn’t.

If I have one reservation, it’s that the description of coordination is calmer than I’ve ever seen it unfold. Aligning metadata is difficult and very political. It touches on issues of ownership, budgets, prestige, and professional identity. While some silos may be accidental, others exist because they are useful to someone. Some of these silos are small kingdoms in themselves, for better or worse. But perhaps that is precisely why the book resonates with me. It doesn’t pretend that this is easy. It simply argues that failing to see the problem clearly is worse.

Reading it now, at a time when organizations are rushing to layer generative AI on top of infrastructures they barely understand, the argument feels even more relevant. These systems depend on context, and in companies, context is metadata. When enterprises hallucinate, their machines will too. Their models are either naive or purely designed, and the fractured descriptions of reality they inherit reflect this.

I closed the book with the uncomfortable feeling that it is less about metadata management and more about organizational self-knowledge. It’s about whether we are willing to closely examine the machinery we already live inside before asking it to think on our behalf. After all, most companies do not fail because they lack intelligence, but rather because they stop paying attention. Ultimately, this book is about learning to notice again.

I have always trusted the margins more than the main text. The main text is so confident. It has structure. Someone decided what belonged there. Someone polished the sentences until they looked almost inevitable. Margins are different. They’re where the reader shows up. It’s the place where the pencil hesitates for a moment before writing “Really?” next to a paragraph. Where a small star appears next to a sentence that feels important but not yet fully understood – yet. Where a question mark lives for years because the reader hasn’t decided what to think. Margins are where thinking leaves fingerprints.

When I read non-fiction, I almost always have a pen in my hand. Writing in the margins is how I slow myself down enough to actually engage with the text. It forces me to pause between sentences and ask a very simple question: Do I believe this? To me, that small pause matters.

Hannah Arendt once described thinking as the silent dialogue between oneself and oneself.1 Reading properly often feels like exactly that. A argument unfolding across pages. The author speaks first. Then the reader replies, in small handwritten comments next to the printed words. Without those interruptions, reading becomes something else entirely. Something closer to pure consumption. And consumption is easy now.

We live in a moment where text arrives faster than attention can follow. Articles summarize other articles. Threads summarize books. AI systems summarize everything. The result is a strange form of intellectual smoothness: ideas pass by us constantly, but very few of them actually stay. My margin notes interrupt that flow. They mark the exact place where the text met resistance. Or curiosity. Or disagreement. In other words, they mark the place where thinking happened. This is also why I’ve always liked footnotes.

Footnotes are, in a sense, the author’s margins. The place where the writer admits that the argument didn’t appear out of nowhere. Someone else thought about this before them. Someone else asked the same question earlier. Someone else’s book is sitting underneath the paragraph you are reading right now. Neil Postman argued that print culture disciplines thought because claims must be supported, referenced, and sustained across pages rather than merely performed in the moment.2 Footnotes are part of that discipline. They slow things down, and they invite the reader to follow the trail. A margin note does something similar, but from the other side of the page. It records the reader’s reaction to that trail. And that is the small idea behind this section.

Margin Notes is where I’ll publish my reflections on books I read about technology, AI, philosophy, and occasionally things that don’t quite fit any category at all. Some will be reviews. Some will be disagreements. But they will never just be summaries. The internet already produces more summaries than any of us can realistically read. What I’m interested in instead is the moment where a book actually pushes back. Where it forces the reader to think a little harder, reconsider something, or follow a reference that leads somewhere unexpected. Those are the moments that deserve a margin note.

And over time, if you collect enough of those small notes, you’ll start to see something interesting: a map of your own thinking forming in the margins of other people’s books. Which is, if we are truly honest, where most ideas actually begin.