This Fraught Moment
The fog comes on little cat feet, goes the poem. Sometimes the storm doesn’t make much more noise. When that quiet arrival coincides with a great deal of tangential commotion and chaos, it can be nearly impossible to hear. And there’s no doubt that the current moment is saturated with both of those things. The commotion may be full of hype from the hypesters and denial from the deniers, but the chaos is real. The upheaval is real. The effect on people’s lives is real.
And yet the volume of those things is drowning out the undramatic arrival of a moment predicted to occur with significantly more fanfare. The Singularity† has been played up as an event that will be unmistakable in its Dolby Surround and Technicolor self-announcement. But it’s here, and it’s busy getting to work while we’re waiting with heads cocked for the dramatic swell of music.
Then What Is It?
Let’s start with a couple things it isn’t. It’s not cyberpunk’s hard integration of tech and tissue, and it’s not Skynet’s beginning to learn at a geometric rate. If not those things, then what?
It’s when machine intelligence becomes capable of true iterative collaboration with human intelligence in a way that interleaves the strengths of both.
You don’t need a brain implant‡ to become a human/machine hybrid mind. You just need to use the technology in a way that amplifies your capability. Pairing with another human on a code or design project doesn’t require you to pipe your skulls together, and this isn’t any different. The synthesis of two intelligences via a rich enough communication channel produces the conditions for better outcomes than either could achieve alone.
And we’re here. It’s expanding people’s abilities. This is a wholly separate claim from “1X -> 10X programmer with AI!”. It’s not about increasing production volume (at dubious quality). It’s about expanding the human’s meaningfully addressable problem solving space. And this isn’t a static extension, like lengthening your reach with a grabber. It’s a continuing, dynamic expansion. Every new domain explored adds to the human’s cognitive toolkit. When AI is the pan-subject-matter expert and the human is the source of judgment, intuition, and taste, you have a hybrid intellect. This intellect can not only do more now, it can bootstrap itself to further capabilities at need or desire.
And the first problem it can help you solve is the problem of itself.
Computers Failed At The Most Important Task Until Now
On all apps the help menu is ubiquitously present and universally insufficient. Some applications have gone so far as to have fully interactive, animated introduction systems (at ruinous cost to implement, let alone maintain). Ultimately, every pre-built help system has fallen short of truly solving the problem. This is not because U/X designers are insufficiently imaginative, or because programmers are lazy§. It’s because trying to anticipate what every user will want to know and what presentation they will find intuitive is a doomed undertaking.
I worked on an input device capable of tracking detailed hand and finger movements. We tried to create a library of basic gestures that could be connected to standard computer functions (volume up/down, browser back, change apps, etc.). We ended up having to scale our expectations way back because there was an utterly unmanageable variety in what people considered to be intuitive for even the simplest described gesture. There is no fixed solution to the human cognition variety problem.
One of the biggest desires in the history of technology has been for help from the machine in its own use. It was a desire unrealized in the forty-plus years since the first users unboxed the first personal computers. But its fulfillment does not come in the form of a machine that can teach you how to use it. It comes in the form of a system that can help you teach yourself how to do anything.
The Big Mistake
Getting back to the image of us collectively cocking our heads for the dramatic music indicating that The Singularity is nigh, there’s an irony playing out right now. The soft singularity is here, but people are looking for the wrong change. Hybrid intelligences are accomplishing things that neither the organic nor the technological components could have done alone. I have personal experience with this very phenomenon. The inference lab bench project I’ve been working on has made heavy use of AI’s pan-subject-matter expertise. The speed with which it was built (and at which I learned) would not have been possible without AI. But the quality of that project in architecture and execution would have been entirely out of reach for the AI without my contribution. It’s GPU-accelerated and data-driven. It generates diagrams from the model architecture data and loads models from safetensors directories as well as GGUF files. Its inference quality is verifiable against a known-good reference implementation, and it does all this in less than 11K lines of clean, comprehensible Go code‖.
These are artifacts of human focus, standards, and critical thinking. But layoff-happy CEOs and Main Street enthusiasts have been making the same mistake: thinking that the technological component can do it alone. Management’s attraction to dumping 90% of their expensive (not to mention opinionated) engineers isn’t a mystery. Nor is it puzzling why early adopters would be keen on offloading chores to agent systems. But the truth is that AI really, really sucks at replacing people. Agents have deleted databases, purged inboxes, and indulged in sprees of drunk texting everyone they could reach. AI doesn’t have judgment, comprehension of consequences, or accountability.
The ideal single-human replacement unit isn’t an AI agent. It’s an augmented human.
No One Ever Suggested Firing Luke So R2 Could Fight Alone
The power in hybrid intelligence is not productivity, it’s adaptability. You know — the trait that got us off the menu and to the top of the food chain? We didn’t get to our place in the world order by banging more rocks more quickly. We got there by discerning which rocks made fire, which ones could be made into tools. We ascended by designing equipment, clothing, and shelter that suited the environments we moved into.
Let’s say you have a mobile web software product that makes use of 3D graphics to do stuff your users love. Could be a game, could be a Zen garden designer — doesn’t matter. You hear that Gaussian splats might be able to improve your users’ experience with better performance and faster load times. In the before times you could carve off one of your engineers to spend weeks doing a deep dive. They’d come back with an answer and probably a small demo. Or if you were in a hurry, you could try to find a contractor to bring their expertise right now. Each of these is expensive in its own way, and the cost that sits between question and answer is both large and unavoidable.
But now, in the world of the soft singularity, that question can be answered in a day. And that same engineer can not only provide the answer, they can immediately pivot to becoming your Gaussian splat implementer if the answer is “yes”. The engineer brings their pragmatic experience and critical mind; the AI brings the new subject. And by the time the preview implementation is done, they have already walked a significant distance down the path to their own expertise. There is a tremendous speed advantage extracted, but it is not 10X coding speed, it is 10X pivot speed. The other thing your pivoting engineer has is your business context. Even if you could onboard a world-class expert that very day, they would need time to orient. Your engineer leverages the AI for the tech, but they already know everything else about your product, your customers’ needs, and your business model.
When your team is composed of hybrid minds, they are decisively more capable of addressing the time-critical challenges of success and survival.
The Power In The Partnership
Folklore traditions warn of the hazards of bargaining with otherworldly powers from fae to demons to djinn. Prompting AI and devising agent instructions can be fairly compared to those cautionary tales. The frames of reference between the parties are aligned enough for communication, but not quite enough to ensure comprehension. Wording must be careful, precise, and encompassing. Consequences for insufficient caution can be disastrous. But AI has a trait attributed to no folkloric being: the ability and willingness to help you get exactly what you want.
The sound and fury attendant to the conversations and consequences of powerful AI have buried the lede and masked the quiet arrival of the soft singularity. It’s changing what people can learn and accomplish, and it’s changing how quickly they can do it. The biggest beneficiaries will be those who don’t seek to eliminate the human from the equation, but embrace the complementary strengths of the hybrid mind.
Footnotes
† ↩︎ The origins and definitions are a deep pit. TL;DR: Ray Kurzweil’s book went broad with the term coined by Vernor Vinge.
‡ ↩︎ You don’t want one, either. Seriously. Think of every bug, pop-up ad, virus, or other bad experience you’ve ever had with your phone. There’s no such thing as a hack-proof computing device. And let’s not even get started on social media doom scrolling. Would you really want that inside your head?!
§ ↩︎ We are, but that’s actually a good thing and a different article.
‖ ↩︎ The other Big Mistake made with AI coding is equating volume with productivity. The software engineering profession dismissed LoC (lines of code) as a useful productivity metric decades ago. Code is not an asset, it’s a liability. Capability is the asset. Good engineering is getting the most capability for the least code and complexity, like any other ROI-driven undertaking.