Don’t Pay for Information. Pay for Mentorship.

In the AI era, the value of training is not more content. It is feedback that builds judgment under uncertainty.

The hesitation is everywhere. AI writes code, explains subsystems, and answers almost any technical question in seconds — so upskilling feels pointless, and paying for training feels riskier still. The advice most engineers receive is the safe option: pay no one; it is all free on YouTube.

It is half right. The wrong half is the part that determines careers.

Never pay for information

The right half first. Nearly every kernel subsystem has a recorded talk, every concept has a dozen explanations, and AI will produce a new one on demand. A paid course that is a collection of recorded videos behind a login is selling information, and information no longer commands a price. On this point the advisors are correct — the fear stems from sound judgment, not weakness.

But the advice rests on one assumption: that learning is the collection of information. Under that assumption, AI really has eliminated the argument for upskilling. The assumption is wrong.

What a career is actually built on

The skill that pays a salary is not stored information. It is judgment under uncertainty: the ability to begin when the problem is unfamiliar, identify the useful cues, state your assumptions, test the right hypothesis, and defend the diagnosis. In embedded Linux work the difference appears quickly — a board that does not boot, a driver that behaves inconsistently, a device-tree change with a side effect nobody predicted. More explanations improve your vocabulary. They do not, by themselves, build the judgment needed to isolate the signal.

Information is abundant. Judgment under uncertainty is not.

So the practical question changes. For years, engineers asked where to learn from — which course, which channel, which book. That question made sense when good explanations were scarce. They no longer are: explanations are everywhere, instant, and mostly adequate. Collecting more of them is no longer the bottleneck. The question that now separates engineers who grow from engineers who stall is a different one — how is judgment actually formed, and is anything in your week deliberately forming it?

Judgment is built through repeated attempts, feedback, and reflection — engaging with the problem rather than jumping to the answer.

Each failed attempt, examined and improved, is one repetition. And repetitions do not happen on their own — they need a structure, a learning system, that makes them happen on ordinary days and not only on motivated ones.

Now test the free-video path against that requirement. A feed supplies content without limit, but the next video is chosen by engagement, not by what your reasoning needs next. Hours accumulate, familiarity accumulates, and the ability to begin on a failure you have not seen before stays where it was.

A feed is not a learning system. It is someone else’s system, tuned for watch time.

The component no free path supplies

Suppose you accept this and build your own system — a fixed daily slot, real projects, code you write rather than watch. That is far better than the feed, and it costs nothing. But self-directed practice can become an open loop without external review: you repeat what you already do well, avoid what exposes you, and cannot see either happening from inside. A video cannot see you at all. AI helps with explanation, but it does not reliably diagnose the blind spots behind your question, and it carries no memory of where your reasoning failed last month.

A responsible mentor supplies exactly this, and it has little to do with delivering content. A mentor watches how you engage a problem, tells you plainly where you are weak, points your next weeks of practice at that weakness, and makes you reflect on the attempt — what you assumed, what you skipped, what you would do differently. The correction is personal, specific, and often uncomfortable. No feed has an incentive to provide that; no tool built to satisfy you will volunteer it.

A mentor closes the loop: someone who watches your reasoning, not your progress bar.

What to pay for, if you pay at all

The decision test for any paid programme is simple. Does the money buy time with a mentor — a person who examines your work, diagnoses your gaps, and redirects your practice? Then you are paying for the scarce component. If it buys recorded videos, you are paying for information, and you know its price.

There are free routes to the same component, and they deserve a sincere recommendation. A senior at work who reviews your thinking is a mentor. So is an upstream maintainer — patch review on a kernel mailing list is some of the most exacting feedback available in this profession, and it costs persistence instead of money. If you can take that route, take it. Its limits are practical, not theoretical: review begins only after you can already produce a credible patch, it examines the patch rather than how you reasoned your way to it, and much of embedded work — vendor BSPs under NDA, internal product code — never reaches a public list. The gap those limits leave is the space where a dedicated mentor matters.

There is no easy substitute for sustained, high-quality feedback — and that is the standard a serious programme should meet.

That standard means not charging for access to explanations that are freely available anyway, but creating repeated opportunities for problem engagement, guided practice, and correction.

That is the principle behind TECH VEDA’s training: live mentor-led sessions where attendance is not optional, built around working through problem scenarios rather than watching explanations — and the strongest feedback from learners is rarely about a specific command or module. The lasting change they report is the confidence to face unfamiliar issues at work and reason towards a solution instead of waiting for one to appear.

The real test of learning is not whether you can reproduce a solution. It is whether you can begin when no solution is visible.

AI has made information cheaper, and in doing so it has made judgment, disciplined practice, and honest feedback more valuable, not less. The engineers pausing their upskilling are misreading what changed: they price training as if it were content, and content has indeed fallen to zero. But what training was always meant to build — the ability to reason through problems that arrive without answers — is now scarcer in the market and better rewarded than it has ever been.

So keep the caution; it is well placed. Refuse to pay for recorded explanations, this year and every year. But do not let that caution stop you from investing in the one asset the AI era has made more valuable. Find the feedback — a senior, a maintainer, a mentor. Put a structure around your practice. And judge every learning decision, free or paid, by a single measure: is it strengthening how you think when the problem is new?

— Raghu Bharadwaj