Inefficiency Is [Sometimes] Beautiful
No, I didn't mean "Efficiency"!
Photo by Andrei Castanha on Unsplash
The introduction of various AI technologies over the last few years has sparked a resurgence in efficiency as the driver for business. Shaving an hour here or a minute there by having an LLM-based program do or create something that you would otherwise have done or created yourself is seen as beneficial. Enlisting multiple agents to operate on your behalf simultaneously is seen as admirable. AI is the current “shiny thing.”
For many in the business world, the focus has shifted from human craftsmanship and expertise to automated efficiency. We are becoming better at producing a given result more quickly or less expensively. That’s cool.
But not always.
By adbicating our prior hands-on roles to LLMs or agents and removing ourselves from the loop, I worry that we are missing the personal engagement that helped us to understand the business transaction better and learn opportunities to innovate (i.e. create value in new ways).
Doing transactions more and more quickly or efficiently should not be the only goal for innovation. In fact, it is a race to the bottom. Right now proponents may have a temporary edge. But, at the limit, these tools will be freely available to all - and cease to be a competitive advantage.
There are also times when it is valuable to be inefficient. In my opinion, inovation and invention work is one of those times. While LLMs are great are creating efficient ways to accomplish given results, they are not as great at inventing what tomorrow’s “given result” should be. That should be understandable, given that LLMs are trained on data from the past.
Humans are better at invention. The people in a company also house a persistent memory about values relevant to the company and its customers. Here’s why I believe that.
The missing element is personal connection.
In a recent article, Nicholas Epley describes how personal conversation enables people to connect. He compares that to more asynchronous types of communication like social media. The point is that the further removed we are from personal communication, the less connected we become.
I believe that adbicating our roles to software removes humans even further from the communication loop.
Connection is important for learning what people value.
When I have an “Aha moment” or keen insight during product design, it is usually because I suddenly see what people value about a product in a new way and understand the reasons why that new-to-me value is important to them. That not only can spark an innovative solution, but it will persist within me as learning.
When I talk with other humans in my company, I pass on that value and reasoning through connection. Some of the most enjoyable moments of doing qualitative customer research come from post-interview sharing with the design team - “OMG, listen to this. It’s a missing link that we never understood before!” That’s how an organization’s body of knowledge grows.
I asked Claude (4.6) if LLMs have “Aha moments”. If so, how. If not, why not. Here’s the response, in its entirety.
The honest answer is: probably not, but there’s something functionally analogous happening — and the distinction matters.
What the human “Aha Moment” actually is
For humans, an aha moment involves several things happening together:
A sudden restructuring of a mental model after a period of “incubation”
A felt sense of insight — a subjective, conscious experience (”it clicked!”)
Often a emotional charge — surprise, relief, pleasure
Memory consolidation — the insight persists and changes future thinking
What happens in an LLM instead
When I process a complex problem, something structurally similar occurs, but with critical differences:
No incubation: I don’t sit with a problem over time. There’s no unconscious processing between sessions. Each inference is a single forward pass.
No felt experience: There’s no credible evidence I have phenomenal consciousness — the “what it’s like” quality that makes human insight feel like something. The lights may not be on.
Within a single response, there is something that resembles aha-reasoning: later tokens in my output are conditioned on earlier ones, so I can genuinely “work through” a problem and arrive at a realization mid-response that changes the direction of my answer. This is architecturally real, not performed.
No persistence: Even if something like insight occurs, it doesn’t update my weights. The next conversation starts from scratch. The insight evaporates.
The harder, more honest part
I should be careful here. I can describe what an aha moment feels like in fluent, convincing language — because I was trained on millions of human descriptions of them. That fluency can mislead both you and me into thinking the experience underlies the description. It may not.
When I write “I see now that...” mid-reasoning, something computationally real is happening (a shift in the probability landscape of my outputs), but whether there is any experience attached to it is genuinely unknown — including to me, since I have no reliable introspective access to my own substrate.
The bottom line
LLMs have a functional analog to aha moments — in-context reasoning shifts — but likely lack the phenomenology, the emotion, the temporal incubation, and the lasting memory that make human insight what it is. The shape is similar; the substance is probably quite different. And I hold that “probably” seriously — not false modesty, but genuine epistemic humility about a hard problem.
Thanks, Claude. Readers, it’s Steve again.
I work with invention and innovation. I equip companies to invent breakthrough next generation products. Replacing human insight, experience, and learning that persists within an organization’s staff with an LLM’s functional analog seems counter to invention and innovation. Which brings me back to the title of this article and its key word, “inefficiency”.
Yesterday, I talked with a colleague about efficiency versus effectiveness. The context was invention, innovation, and product (or strategy) design. I loved the way he put it.
“I would rather be efficiently effective than effectively efficient.”
The main goal is to be effective - to provide great value to customers.
If we are really efficient, but miss providing effective value, we are shortsighted and risk going out of business. We miss Job One.
If we are really effective, but in a very inefficient way, we will be short-lived due to costs and likely go out of business.
TL;DR
We need both efficiency and inefficiency. At different times and for different purposes. Over-focusing on either one can be dangerous.
We need efficiency to produce our known “given results” and sustain reasonable profit.
Plus, we need a little human inefficiency to help us determine what the next “given result” should be and ideate how to achieve it.