Just five years ago, a programmer's efficiency was measured in a way that any engineer from the 1950s would understand. You get a task. You estimate complexity. You deliver. You account for execution. Errors are deviations from the norm — the fewer, the better. Speed is currency — the more you do per unit of time, the higher you were valued.
This logic didn't apply only to programmers. It applied to everyone. An accountant who closed the balance sheet without mistakes. A surgeon who performed flawless operations. An architect who designed a building matching specifications. Efficiency meant: output divided by time, penalised for errors.
It was simple. It was measurable. And it built the entire modern business world.
Three pillars underpinned every evaluation, every promotion, every raise:
Do more — scale, productivity, throughput. The more you produce, the better. Factory, office, line of code.
Do better — quality, precision, zero defects. Six Sigma. Continuous improvement. Kaizen.
Do differently — innovation, differentiation, competitive advantage. Find a way that others haven't found.
Every company, every employee was assessed across these three dimensions. And for decades, it worked. Because tools were limited, and the human mind was what gave them value.
And then AI arrived
It didn't come slowly. It didn't knock on the door. It kicked it in.
In just two years — from late 2022 to today — AI tools went from technological curiosity to everyday work instruments. Programmers talk with agents that write code for them. Lawyers use models that analyse thousands of pages of documents in minutes. Doctors consult diagnoses with systems that have read more medical papers than any human in history.
And suddenly, the old three pillars — more, better, differently — stopped being the exclusive domain of humans.
More? AI can generate content, code, analyses at a scale incomparable to humans. A hundred solution variants per minute. A thousand pages of documentation per hour. Scale that Ford's factory could only dream of.
Better? In many tasks, AI already makes fewer errors than humans. Not because it's smarter. Because it's not tired at five in the afternoon. It's not distracted by emails. It doesn't have bad days.
Differently? AI can combine patterns from different fields, creating solutions humans wouldn't have invented. Not through genius — through scale. It sees connections in billions of data points that the human mind could never grasp.
And here we reach the crux. If the machine meets all three conditions of the old definition of efficiency — then humans cease to be efficient in the old sense of the word.
Not because they are worse.
Because the framework has collapsed.
New question: what does "effective conversation with AI" mean?
Since old measures don't work, we need new ones. And the first intuition is: efficiency in the AI era is the quality of interaction with the machine. The ability to conduct a conversation. To formulate instructions. To decompose problems.
But this immediately raises further questions.
Does conversation style matter? Does someone who formulates instructions concisely get a different result than someone who provides rich context? Is "the ability to talk to AI" a new professional competency — as real as the ability to write code or negotiate?
The answer is: yes. And the differences are enormous. Same model, same task — but an instruction written by someone who understands the problem deeply yields incomparably better results than one written by someone who "just wants AI to do it for them."
Does technical and interdisciplinary knowledge help? Is a programmer talking to AI about code more effective than a layperson? Does someone who understands the domain — medicine, law, finance — plus technology, extract more from AI?
Again: yes. Interdisciplinarity becomes a multiplier. Not additive — multiplicative. Whoever understands both the problem and the tool achieves results unavailable to specialists in just one.
Is it better to decompose tasks or give broad instructions? Does breaking things into atomic steps yield better results? Or does AI perform better when given wide context and left to decompose on its own?
The answer here is more complex. But one thing is certain: decomposition is a competency. The ability to break a problem into parts isn't a technique for formulating instructions — it's a way of thinking. And those who've mastered it produce better results regardless of the tool.
Where are we really?
Let's step back from philosophy for a moment and look at the facts.
"Build me system X described in a few sentences." Does that work yet? In a limited sense — yes. Simple applications, websites, automation scripts — AI can generate them from a short description. But a banking system? A medical platform? An ERP? No. We're not even close.
"Build me a system like system X." This is even harder. Because it requires understanding not just code, but business context, processes, exceptions, edge cases, security policies, legal regulations. AI doesn't know the context nobody gave it. And context is precisely what humans bring.
Where does the magic of the instruction end and the need for human intellect begin? Exactly where the problem stops being typical. Where there's no pattern in the training data. Where you need to make a decision that follows from no algorithm — only from understanding situations, people, risk.
Limits nobody talks about
And here we come to something the tech industry is silent about. Because it doesn't fit the narrative of unlimited progress.
Human intellect is a finite resource.
It's not about IQ. It's about cognitive energy. The capacity for focus, analysis, decision-making. It's one thing to perform simple, repetitive tasks for eight hours. Entering data. Writing reports by template. Filling in forms.
It's quite another to warm up your brain for eight hours of intelligent instruction formulation for AI.
Effective work with AI isn't clicking and waiting for results. It's constant decision-making. Evaluating outcomes. Course correction. Decomposition on the fly. Critical analysis of what AI returned. Reformulating the question. Changing strategy. This is an enormous intellectual effort.
And you can't do this all day. The human brain isn't built for eight hours of uninterrupted analytical work. Cognitive psychology is clear: after three to four hours of intensive mental work, decision quality drops dramatically.
Are these new limits of industriousness? Not physical — like in a factory. Not even "mental" — like in an office with repetitive tasks. But purely intellectual. Limits of the capacity for deep thinking.
Perhaps three to four hours of truly good, deep work with AI is the maximum. And this is precisely the new "working day."
How does this change the definition of efficiency? Perhaps it's the unused limits — on tokens, on queries, on time with AI — that become the measure of how much intellectual fuel remains in the tank? Not "how many hours did you work," but "how many deep interactions were you able to conduct"?
The chasm between thinking and laziness
There's writing a good, intellectually deep instruction. One that provides precise context, sets the right constraints, defines the expected result. One that requires a person to truly think through the problem before formulating it.
And there's cutting corners. "Just do something." "Write this for me, I can't be bothered to think." "Generate code for this feature."
The difference in results is enormous. And it grows with each generation of models. The better the tool, the more visible the quality of the question becomes. A good instruction isn't a technique — it's proof of thinking.
And perhaps this is the only true measure of efficiency in the AI era: the quality of thinking invested in the interaction. Not quantity. Not speed. Quality.
The circle closes
And here we arrive at the moment where this whole story comes full circle.
We started with old rules. Do more. Do better. Do differently. Then AI came and shattered those rules. We searched for new ones. We tried to define efficiency through the quality of instructions, through the ability to converse with machines, through new competencies.
And then we discovered that human intellect is finite. That you can't work with AI at full throttle for eight hours. That the real difference doesn't lie in the tool, but in what the human brings to the interaction. In the depth of thinking. In the ability to ask questions. In understanding context.
And suddenly we realise something surprising.
Human intellect is valuable in exactly the same way it has always been.
Once a chisel. Then a keyboard. Now a conversation with a machine. Tools change — and will keep changing. But what truly matters — the ability to think — hasn't changed one iota.
Old virtues return. Depth of thinking. Interdisciplinarity. Patience. Precision. The ability to understand things the machine doesn't understand — not because it can't, but because nobody asked it to.
We're not in a new world. We're in the same world with new tools.
We're not searching for a new definition of efficiency. We're destroying the old one — and discovering beneath it the same truth that was always there.
The more powerful the tools, the more valuable the human who knows how to think.
We don't write this to glorify AI. We don't write this to demonise it. We write this to ask a question that gets lost in the noise of the technological revolution:
Has what truly matters really changed? Or have the tools simply become powerful enough for us to finally see what efficiency has always been — the ability to think?