The End of 'Technical' vs 'Non-Technical'

When Andrej Karpathy (former Director of AI at Tesla and founding member of OpenAI) says he's never felt this behind as a programmer, the rest of us should sit up and pay attention.

That's the starting point of a brilliant reflection by Nate B Jones, whose recent video unpacks what he calls a "phase transition in technical leverage." If you're a leader trying to understand how AI changes your workforce (not just your engineering team, but everyone), this is essential viewing.

The Old World is Breaking

For decades, technical competence meant one thing: writing correct instructions faster than other people on problems that mattered. You internalised abstractions, mastered your tools, and shaped deterministic systems. When something broke, you could trace causality, step through the behaviour, and find what was wrong.

In that world, authorship and authority were tightly linked. If you wrote the code, you owned the behaviour. You had both the authority and the knowledge to fix it. This is why the conventional wisdom has always been to keep your founding engineer around (they know where all the skeletons are buried).

That entire system of assumptions is now changing.

The Shift: From Authorship to Orchestration

The unit of leverage has shifted from writing code to orchestrating intelligence. And this isn't a buzzword. An LLM is a probabilistic token generator (it produces plausible sequences conditioned on inputs). You can't single-step through its reasoning. You can't rubber-duck it. You can't reliably reproduce its outputs.

As Jones puts it: "We didn't just get a better Python library. We got a new kind of machine in the loop."

Three things broke simultaneously. First, control is no longer the default. You don't author behaviour anymore (you condition it). Mastery shifts from "make it do exactly what I want" to "steer reliably toward outcomes and detect when it drifts."

Second, effort no longer maps to output. Someone who knows how to set up a delegation loop will outperform someone grinding manually (despite equal intelligence). The skill is delegation, not execution.

Third, the abstraction stack inverted. We used to collapse intention downward into implementation. Now we define intent, let the system generate artefacts, and verify the output. The job moves from construction to supervising a construction crew.

The New Skill Tree (For Everyone)

Here's where Jones' framework becomes genuinely useful for leaders. He lays out a hierarchy of capabilities that aren't just for engineers (they're for anyone who needs to tell probabilistic machines how to generate useful work).

Level 1: Conditioning

The foundation is learning to steer. This means tight intent specification (ambiguity is "gasoline on the fire" for probabilistic systems), context engineering (what goes in, what stays out, what gets summarised), and constraint design (output formats, schemas, rubrics, token budgets). Without constraints, you have a slot machine. With them, you have a reliable component.

Level 2: Authority

This is the difference between "I used AI" and "I know how to operate an AI system responsibly." It requires verification design (how does truth enter the loop?), provenance and chain of custody (where did claims come from?), and permission envelopes (the model cannot be your security boundary). You can ship something correct without understanding why (and ship something wrong that looks correct). The craft becomes designing workflows where you delegate generation but ensure human authority over what actually ships.

Level 3: Workflows

This is where compounding leverage emerges. You stop treating the model like a chatbot and start treating it as a pipeline component. You build intermediate artefacts, create checkpoints, develop a failure mode taxonomy (was context missing? retrieval wrong? tool failure? hallucination?), and make the surrounding system extremely observable. You're taking auditability skills and scaling them.

Level 4: Compounding

The highest leverage comes from evaluation harnesses (without evals, you're just improvising faster), feedback loops (draft, critique, revise, recheck, ship), and drift management. You treat your AI work like production infrastructure (versioning, auditability, governance) because you're operating under continuous change without losing control.

The Factorio Metaphor

Jones uses the video game Factorio as a perfect training metaphor. You land on a planet and start by handcrafting basic items, but the system pushes you into automation. You improve mining, install conveyor belts, route outputs into more factories, automate the supply chain.

The instincts that scale in Factorio are the instincts that scale in AI work: decomposing problems, modularity, observability, understanding bottlenecks, blast radius estimation. Nobody cares if you personally crafted a gear. What matters is that the system produces gears at scale that do useful work.

The Real Divide

The most important divide is no longer engineer vs non-engineer. It's someone who can delegate vs someone who can't. A lawyer building a contract review workflow and an engineer building a debugging agent are climbing the same skill tree. Different artefacts, same hierarchy of capabilities.

As Jones concludes: "The new hierarchy won't be based on who codes the fastest. It will be based on who can orchestrate uncertainty without losing authority. That's what technical means now. It's for everyone."

What This Means for Leaders

If you feel behind, it's not that you're failing (you're correctly perceiving that the stack is different now). The way forward isn't frantic tool-chasing or denial. It's recognising that every knowledge worker in your organisation needs to develop these capabilities, and building the structures to help them do so deliberately.

Organisations that figure this out (that take this skill tree, detail it for their context, and scale it across their workforce) will realise 10x speedups. Those that insist on old hierarchies of technical vs non-technical, that keep people locked in narrow job hats, won't.

The choice is yours.