Reading has always been a safe space for me. A room inside a room where the noise stops. I read two to three books a year — not enough, and I know it. But right now, a very loud argument is being made that none of it matters anymore. That AI has made knowledge a commodity. That degrees are obsolete. That learning programming is pointless. I want to give an honest response to that — not a defensive one.
- AI commoditised information retrieval. The ability to evaluate, connect, and apply what is retrieved was never the commodity — and it is now the differentiator.
- What twenty-four years of learning actually bought: the frameworks decayed, the fundamentals compounded — and the fundamentals are what let you direct AI today.
- An eight-book stack mapped to the AI concepts that matter: prompts, LLMs, context, RAG, MCP, and agents.
What It Feels Like to Actually Know Something
There is a specific feeling I associate with having genuinely learned something — not retrieved it from a search, but actually learned it. It is a feeling of being settled inside a subject. Of knowing where your knowledge ends and where uncertainty begins. Of being able to walk into a difficult conversation, a production crisis, an architectural review — and feel, underneath the pressure, something solid.
A cosy room. A window. Rain outside. A cup of coffee, a candle, and a book open in your lap. Something about that scene is not just comfortable — it is the physical version of what learning actually produces inside you. A sense of being grounded. Of knowing your own ground. That feeling is not nostalgia. It is the interior architecture of genuine competence — and it is the most useful thing you can build in a career.
I want to say this with some humility, because I am not a perfect learner and I do not read as much as I should. But I notice, directly, the difference between periods when I am reading and thinking deeply and periods when I am not. The quality of my judgment changes. The confidence is quieter but more solid. You know your stuff. Not loudly — just clearly. That is what AI cannot manufacture for you.
Information, Knowledge, Wisdom — They Are Not the Same Thing
Before we evaluate any claim about AI making learning unnecessary, we need to be precise about what we mean by "learning" and "knowledge." Because the argument only holds if you collapse three very different things into one word.
The claim that AI makes learning unnecessary is accurate only at Level 01. It confuses the retrieval of information with the possession of understanding. And that confusion is not innocent — for a lot of people, it is genuinely costly.
What Twenty-Four Years of Learning Actually Bought
I wrote my first production code in 2002. Nearly every specific technology I learned that year is gone. The application server is discontinued. The framework configuration files I could write from memory are a historical curiosity. The certification I studied for tests knowledge that no longer has a job attached to it. If learning were about the technology of the year, my first decade would have been a write-off.
EJB entity beans. Struts configuration. SOAP toolkits. The build tool before the build tool before the current one. Every framework-specific skill had a half-life of three to five years — and the half-lives are getting shorter, not longer.
How databases commit and fail. How networks behave under load. How to decompose a domain into parts that can change independently. How to read code I did not write and find the assumption that breaks it. None of this has decayed in twenty-four years. All of it has compounded.
That distinction is the entire answer to "AI makes learning unnecessary." When an agent generates a service for me today, the model knows more syntax than I ever will — and that costs me nothing, because syntax was always in the decaying column. What the model cannot do is know which of its own outputs to distrust. The fundamentals I built in the compounding column are precisely what let me catch the things AI gets confidently wrong: the transaction boundary it ignored, the failure mode it never considered, the domain rule it violated while passing every test. The learning that mattered was never the technology of the year. It was the engine underneath — and the engine is what AI amplifies.
The Fake Narrative — and Why It Is Worth Naming Directly
There is a meaningful difference between thoughtful people making debatable claims about the future of learning, and the content machine that has built an industry around telling people what they want to hear. I want to be direct about the second category.
"5 AI Tools to Make $100K — No Skills Required"
This content pattern is dishonest. Not wrong in a debatable way — dishonest in a deliberate way. It combines a real technology with a false narrative about wealth being accessible through shortcuts, with the purpose of generating attention and selling courses. The people publishing these videos know the claim is not reproducible. The people watching them are building neither skills nor wealth — they are building dependency on the next shortcut. I have no patience for it, and I think it deserves to be called what it is.
The honest version of the AI opportunity looks nothing like that. It is not a shortcut to wealth. It is a new layer of capability available to people who already have the judgment to use it. AI amplifies what you bring to it. If you bring expertise and clear thinking, it amplifies that. If you bring nothing but a prompt, it amplifies nothing. The shortcut content skips this entirely, because admitting it would dissolve the product.
How AI Actually Works in Practice
I use AI every day. It is genuinely useful. But its usefulness is proportional to what I already understand — and that relationship is not incidental. Here is an honest picture of where AI creates real value.
Use AI to go deeper on a topic you are already studying, get explanations calibrated to your level, surface adjacent concepts, and test your own understanding through dialogue. It accelerates learning. It does not substitute for it — because the comprehension still has to happen inside your head.
AI handles first drafts, boilerplate, research aggregation, and repetitive structure. This frees you for judgment work — the decisions, the architecture, the review, the communication that requires real expertise. But you need the expertise to know what to do with the time it saves.
AI can make you faster, more consistent, and more thorough in what you deliver. The judgment about what the customer actually needs — the diagnosis, the trust, the relationship — still requires a human who genuinely knows the domain.
It does not know your specific context, your specific customer, or your specific constraints. It cannot tell you when it is wrong with any reliable confidence. Catching that — directing it well and knowing when not to trust it — requires the knowledge it supposedly makes unnecessary.
After AI, Emotional Intelligence Matters More Than Ever
There is one dimension of this conversation that the industry almost never addresses, because it does not map cleanly onto the AI story: as AI handles increasingly sophisticated cognitive tasks, the human capacities that remain irreplaceable are not technical. They are emotional.
High EQ Matters More Than High IQ in Almost Every Real-World Situation
I believe this. AI is rapidly closing the gap on IQ-adjacent tasks — pattern recognition, information synthesis, logical derivation. But the ability to read a room, to know when a colleague is struggling, to earn trust over time, to give feedback someone can actually receive, to navigate a difficult client relationship, to lead a team through uncertainty — none of this is threatened by AI. All of it is becoming more distinctively valuable as AI takes more of the cognitive baseline. If you are early in your career and wondering what to develop: technical skills, yes. But invest in your emotional intelligence with the same seriousness.
Books Worth Reading for This Moment
Two reading lists for this specific moment. The first two are about building the kind of depth that makes any tool powerful. The next six map directly to the AI stack: one book per core concept, from how models think to how they act. For a more extensive list — including the habits, thinking, and persistence shelf — visit the full AI reading shelf.
Three more in the same spirit — Atomic Habits (Clear), Think Again (Grant), and Grit (Duckworth) — are on the reading shelf with notes.
Who I Actually Read — and How I Check the Claims
Fable 5 was released this week, and I have been using it for two days. For the first time in my career, I feel that something is more intelligent than I am in my own domain — software development and cloud architecture. Not faster at typing. Not better at retrieval. More intelligent, in the part of the work I considered mine. Look at METR’s time-horizon numbers and you will see the same thing measured: the task length these systems can complete keeps doubling on an exponential trend.
That moment is exactly why this article exists — and why the shelf below matters more now, not less. When the tools outpace you, the differentiator is knowing whose signal to trust and how to verify a capability claim yourself. Curate your inputs the way you curate your code reviews.
The people who stop reading because AI can summarise will become the people who can only prompt. The people who keep reading will be the ones who know what to ask — and what the answer actually means.
I am going to double my reading this year. Then triple it. Not to perform intellectual seriousness on social media. But because I have lived long enough in this field to know, without any ambiguity, that the quality of thinking I bring to hard problems is directly connected to how much I am reading and learning outside of those problems. That relationship does not change because a model can retrieve any fact in two seconds. It might even strengthen.
One thing I have not said yet: none of this infrastructure is free. The companies building AI are not doing it as a public service. They are making the largest capital bets in modern technology history — and the revenue model is consumption. Per token. Per seat. Per API call. Every developer, every team, every enterprise deploying these tools is already inside a meter that is running. Understanding what that costs, and deciding whether the return justifies it, is the second conversation. That is the next article.