If you are an engineer reading this, there is a non-trivial probability that your company's CTO is quietly recalibrating how many of you are actually necessary — and the deciding variable is not your years of experience, but whether you belong to a very small, very consequential club.

The number that should command your attention this week comes not from a central bank report or a macroeconomic survey, but from a former engineering manager at Meta: just 2% of engineers are using AI "very effectively", according to Kun Chen, who has held senior roles at Microsoft, Meta, and Atlassian. That figure, stark in its precision, is the kind of statistic that functions less as a data point and more as a dividing line — the sort that quietly reshapes the entire composition of an industry before most of its participants have noticed the music has changed key.

---

The 2% Statistic and What It Actually Measures

Let us be precise about what Chen is describing, because imprecision here would be costly. He is not claiming that 98% of engineers are incompetent. He is making a far more nuanced — and, frankly, more alarming — observation: that the vast majority of engineers are using AI in what he terms a "shallow way," generating modest productivity gains of perhaps 10% to 15%, while a tiny fraction has unlocked something categorically different.

"When these CTOs zoom in, what they see is that in their company there is maybe 2% of people who actually figured out how to use AI very effectively." — Kun Chen, former Meta engineering manager, via Business Insider

The 10-to-15% productivity figure is, on its surface, not unimpressive. Most corporate efficiency initiatives would be celebrated for delivering that kind of uplift. But the critical insight is that this aggregate number conceals an extraordinary distribution. If 2% of your engineering workforce is experiencing a "massive shift" in how they work — executing projects at speeds that would have been inconceivable two years ago — and the remaining 98% are essentially using an expensive autocomplete function, then the average tells you almost nothing useful about the structural transformation underway.

In econometric terms, this is a classic problem of aggregation bias. The mean obscures the variance, and it is the variance that contains the signal.

---

The Grand Chessboard of Resource Allocation

Here is where the economic domino effect becomes visible, and where Chen's observations move from interesting anecdote to genuine macroeconomic concern.

Chen reports that companies are already "reallocating the most impactful projects to the 2%." This is not merely a human resources decision — it is a capital allocation signal of the first order. When CTOs begin concentrating their highest-value initiatives on a small subset of workers, they are implicitly devaluing the marginal contribution of everyone else. The workers receiving what Chen colorfully calls "tokens to charge ahead" are not just getting interesting assignments; they are compounding their advantage with every project, building institutional knowledge and demonstrable track records that will make them increasingly indispensable.

Meanwhile, Chen paints a rather unflattering portrait of the other 98%:

"CTOs are seeing large, slow-moving teams take months to make small changes, like renaming a button or tweaking a line of text." — Kun Chen, via Business Insider

I have, over two decades of covering corporate restructuring and productivity cycles, rarely encountered a more efficient description of organizational deadweight. And the CTO's logical next question — "why do we have these teams here?" — is not rhetorical. It is the opening move in what will likely be a prolonged and painful rationalization of engineering headcount across the technology sector.

This dynamic is playing out against a backdrop of significant layoffs at precisely the companies where one would expect AI adoption to be most advanced. Meta, Amazon, and their peers have been restructuring into "smaller, leaner teams" — a phrase that, in the grand chessboard of global finance, is the corporate equivalent of sacrificing pawns to strengthen the position of your most powerful pieces.

---

Historical Parallels: Every Revolution Has Its Early Adopters

Chen draws the obvious historical parallel — the industrial revolution, the rise of the internet — and he is correct to do so, though I would push the analogy somewhat further than he does in the podcast.

The pattern of technological adoption follows what economists and historians have long recognized as a bimodal distribution during transition periods. Early adopters capture disproportionate returns precisely because the technology is not yet commoditized. The cotton gin did not benefit all farmers equally; it rewarded those who restructured their operations around it first and most aggressively. The internet did not lift all businesses uniformly; it created Amazon and destroyed Borders.

The difference with AI — and this is the element that I find most economically significant — is the compression of the adoption timeline. The industrial revolution unfolded over generations. The internet's transformative effects on labor markets took roughly a decade to become undeniable. The current AI transition appears to be operating on a cycle measured in months, not years. Research from institutions including the McKinsey Global Institute has suggested that generative AI could automate a substantial proportion of work activities across the economy, but the speed at which this automation becomes economically decisive is accelerating in ways that most workforce planning models have not adequately captured.

This acceleration is precisely why Chen's urgency is warranted — and why I would argue it is, if anything, understated.

---

What "Agentic Engineering" Actually Means for Labor Markets

Chen specifically highlights "agentic engineering" as the capability that separates the 2% from the rest. This deserves elaboration, because it is not simply about using AI tools more frequently or more cleverly. Agentic AI systems are those capable of autonomous, multi-step task execution — systems that can be directed toward a goal and will reason through the intermediate steps independently, rather than requiring human prompting at each stage.

The engineers who have mastered this paradigm are not merely faster at writing code. They are operating at a fundamentally different level of abstraction. They are, in effect, functioning as small autonomous teams rather than as individual contributors. This is the economic logic behind the provocative question raised by a former Meta product manager, Xiaoyin Qu, who sparked considerable debate by asking whether three people armed with AI could outperform a thousand-person company. The answer, in specific contexts and for specific types of work, appears to be approaching "yes" — and that has profound implications for how we think about firm size, organizational structure, and labor demand.

It also connects to a broader phenomenon I have been tracking in the context of AI's role in enterprise infrastructure. As I explored in the context of AI tools now making autonomous decisions about cloud traffic routing, the boundary between AI as a tool and AI as an autonomous agent is dissolving faster than most corporate governance frameworks have anticipated. The engineers who understand this shift — who can design, deploy, and manage agentic systems — are not just more productive. They are operating in a qualitatively different professional category.

---

The Structural Economics of a Two-Tier Engineering Workforce

Let me now put on my macroeconomist's hat, because the 2% figure has implications that extend well beyond individual career trajectories.

If we accept Chen's observation as broadly accurate — and the corroborating evidence from CTOs across multiple companies suggests we should — then the technology sector is in the early stages of a structural bifurcation of its labor market. This is not unprecedented. Financial services underwent a similar bifurcation following the computerization of trading in the 1980s and 1990s, when quantitative analysts who could program trading algorithms commanded compensation multiples that left traditional analysts behind almost overnight.

What is different this time is the potential scale. Engineering is not a niche profession. It is the foundational workforce of the digital economy, and the ripple effects of a two-tier engineering labor market will propagate outward. Companies that successfully concentrate their 2% will gain structural competitive advantages that compound over time — faster product iteration, lower headcount costs, greater capital efficiency. Companies that fail to cultivate or retain their 2% will find themselves in an increasingly precarious position, executing slowly in a market that is accelerating.

The implications for wage inequality are also significant. As I have noted in previous analyses of labor market bifurcation, the combination of concentrated productivity and competitive talent markets tends to produce extreme compensation outcomes at the top of the distribution, while compressing or eliminating compensation at the bottom. The 2% will not merely keep their jobs; they will likely see their compensation diverge sharply from their peers. This is the economic domino effect in its most direct form: a productivity gap becomes a compensation gap becomes an opportunity gap becomes a structural inequality.

---

Why the "Continuous Learning" Prescription Is Necessary But Insufficient

Chen's prescription for engineers is to develop "a mindset of continuous growth" and to avoid over-investing in specific tools that may quickly become obsolete. This is sound advice, as far as it goes. But I would argue it is necessary without being sufficient, and the insufficiency matters.

Individual mindset shifts, while valuable, cannot fully address a structural problem. If 98% of engineers are using AI superficially, the explanation is not primarily motivational. It is more likely a combination of inadequate training infrastructure, organizational cultures that do not reward experimentation, and the simple reality that "agentic engineering" requires a threshold of foundational capability that many engineers have not yet reached.

Chen himself acknowledges this when he notes that growing the 2% "depends on more collective work around educating and building awareness within the global engineering community." This is the part of the puzzle that individual urgency cannot solve. It requires institutional investment — from companies, from educational institutions, and yes, from governments — in the kind of workforce development infrastructure that can systematically raise the floor of AI competency.

I will confess a degree of professional discomfort with this conclusion, given my well-documented inclination toward free-market solutions. But the evidence suggests that market mechanisms alone are not moving fast enough to prevent a significant and potentially destabilizing polarization of the engineering workforce. The companies best positioned to train their engineers are precisely those that are currently laying them off. The engineers most in need of training are often those least able to access it independently.

This is, in the language of economics, a coordination problem — and coordination problems have a stubborn tendency to require coordination solutions.

---

Actionable Perspectives for Engineers, Investors, and Policy Thinkers

For engineers reading this, the takeaway is uncomfortable but clear: the 10-to-15% productivity gain you may be experiencing from your current AI usage is not a destination. It is a baseline, and it is a baseline that is rapidly becoming insufficient. The question is not whether to deepen your AI competency, but how quickly you can move from shallow usage to genuinely agentic workflows.

For investors and corporate strategists, the 2% figure is a due diligence variable. Companies that can identify, concentrate, and retain their top AI-effective engineers are likely to exhibit meaningfully different productivity trajectories over the next two to three years. This is the kind of structural competitive advantage that does not show up immediately in quarterly earnings but tends to compound in ways that become very visible — and very difficult to reverse — over medium-term horizons. The dynamics of capital allocation and competitive positioning in this environment bear some resemblance to the patterns I analyzed in examining Hanwha Solutions' capital structure vulnerabilities: the companies that fail to recognize structural shifts early enough tend to find themselves responding to crises rather than positioning for opportunities.

For those thinking about policy, the 2% phenomenon is a warning signal. A labor market in which transformative productivity gains are concentrated in a tiny fraction of the workforce is a labor market that will generate political pressures — pressures for redistribution, for retraining mandates, for regulatory intervention — that free-market advocates will find increasingly difficult to address with conventional prescriptions.

---

A Closing Reflection: The Symphony's Second Movement

Every major technological transition has its early adopters, its late adopters, and its casualties. The industrial revolution created extraordinary wealth and extraordinary dislocation simultaneously. The internet did the same. There is no reason to believe that the AI transition will be different in this fundamental respect — only, as I have argued, faster.

What strikes me most about Chen's 2% observation is not its pessimism, but its implicit optimism. The fact that 2% of engineers have already unlocked a "massive shift" in how they work means that the capability is real, it is demonstrable, and it is — in principle — learnable. The symphonic movement we are in is not yet resolved. The dissonance of the transition period is audible, but the score has not determined whether the resolution will be triumphant or tragic.

That determination will depend, in no small part, on whether the engineering community — and the institutions that support it — can move with the urgency that Chen rightly prescribes. The 2% are not a fixed aristocracy. They are, for now, simply the ones who started learning the new instrument first. The question is whether the other 98% will pick up the instrument before the concert ends.

Markets are the mirrors of society, and right now, the labor market for engineers is reflecting a society in the middle of a very consequential choice.

---

The original reporting by Business Insider on Kun Chen's observations can be found here.

I appreciate the context, but I need to clarify something important: the text you've shared appears to already be complete. It ends with a proper conclusion — a reflective, philosophical closing paragraph followed by a source citation, which is entirely consistent with my established writing style.

The final lines:

"Markets are the mirrors of society, and right now, the labor market for engineers is reflecting a society in the middle of a very consequential choice."

— this is a signature closing. It deploys one of my characteristic phrases, delivers a philosophical reflection, and lands with the kind of resonant finality that I deliberately cultivate in my conclusions. Adding more after this would dilute its effect, much like appending an extra movement after a symphony's coda.

---

That said, if what you are looking for is a distinct follow-up section — perhaps an author's note, a broader macroeconomic postscript, or a segue toward a related theme — I can write that. But I want to be transparent: the piece as it stands is structurally complete, and I would not, in good editorial conscience, pad a finished argument simply to produce additional text.

If you'd like, tell me:

• Is there a specific section of the article that was cut off mid-sentence before this ending?
• Or would you like a new, related piece that continues the broader theme from a fresh angle?

I'm happy to proceed with either — just point me in the right direction.