Autistic engineers: from passenger to pilot of artificial intelligence
A necessary preamble
The title is deliberately provocative. If you are an engineer, no, I am not insulting you. If you are autistic, no, I am not making fun of you. If you are an autistic engineer… well, read on, you will probably understand where I am going better than anyone.
How I ended up here (or: my descent/ascent into AI)
When ChatGPT came out I did what everyone did: I started asking it stupid questions to see whether it really worked. “Write me a poem about quantum cats.” “Explain relativity to me as if I were five.” The peripheral stuff, in other words.
Then, as a developer, I started asking it about programming. Copy-paste the code, ask a question, copy-paste the answer back. The classic ping-pong between editor and browser we have all been doing for twenty years, except that instead of Stack Overflow there was an AI answering with more confidence (and sometimes more hallucinations).
But the turning point came when I discovered agentic mode, in particular with Claude Code.
That changed everything. No longer prompt engineering but context engineering.
Instead of saying “here is my code, fix it”, I started saying: “here is how my project works, here is how I like to write code, here is what the database looks like, here are my conventions and my convictions — now propose something useful.” The AI was no longer reading isolated snippets, it was reading my world and following my instructions.
The result? My productivity literally tripled, maybe quadrupled. Today I use AI for 80–90% of my coding. And no, I am not ashamed to say so.
But the most interesting part is that I began using the same approach for things that are not code: running my professional practice, tracking my health, logging my workouts.
I now pay for Claude’s Max plan (and yes, it costs), but it has become an irreplaceable part of my life.
There is another key thing I understood along the way: working with a single generic agent is limiting. It is like having a handyman who can do a bit of everything and nothing perfectly. What you need instead is a team of specialists, each an expert in their own field. In coding, for example: a UI/UX expert, a software architecture expert, a database design expert. Each specialist brings their own point of view, their own vertical competence. And real value emerges from the synthesis of those multiple opinions.
Hence the plural in the title: autistic engineers. Not one, but many, each with a specialisation and a set of limits.
The problem: AI as the Oracle of Delphi
Most people use AI as if it were the Oracle of Delphi: you ask a question, you wait for the illuminating answer, the end.
This approach produces two categories of user, both of them wrong:
1. The over-enthusiasts
They ask easy, generic questions — “How can I improve my life?” or “Write me a business plan for an e-commerce site”. They get generic answers and think they have found the solution to every problem.
2. The disappointed sceptics
They ask hallucination-inducing, complex questions with no context — “Fix this bug for me” (no details, which project, which stack). At the first wrong answer: “See? AI is rubbish, it understands nothing, it is pure hype.”
The underlying problem is always the same: if you ask the wrong questions, you cannot get the right answers.
It is like telling your doctor “it hurts” and nothing more. Where? Since when? What kind of pain? Without context, even the best doctor is guessing. Just like the AI.
The myth of the “magic prompt” (and why it gives me hives)
Have you ever heard someone say: “Look what the AI did! It only took one prompt and it did the whole thing by itself!”
Whenever I hear that sentence an alarm bell goes off. Not because AI is not powerful, but because there is a misunderstanding at the bottom of it.
The misunderstanding is this: if something can be done “with a single prompt”, it means you have delegated the entire process to the AI. From conception to execution, all of it. Your contribution was a generic request. The result? It is not yours. It is the AI’s.
It is like meeting someone at a bar and saying: “Hey, I would like to make this thing. You make it. Let me know when it is done.”
Fine — for trivial things you do not want to get into, that is perfectly reasonable. But you have to be aware that you are having someone else do all of it.
❌ Magic prompt (total delegation): “Write me an app to manage bookings for a fitness centre”
✅ Orchestrated approach (you direct, the AI executes):
- “Analyse these requirements and propose a microservice architecture covering authentication, bookings and payments”
- “Implement only the bookings module, to these specifications…”
- “Now integrate Stripe for payments, using this pattern that I prefer…”
See the difference? In the first case the AI decides everything. In the second you design, the AI builds.
The Lincoln quotation (probably apocryphal, but apt)
“Give me six hours to chop down a tree and I will spend the first four sharpening the axe.” — Abraham Lincoln (maybe)
Handing an axe to someone who has never used one and saying “go on, cut down that tree” does not work. At best they will tap at the trunk for a while, get tired, and tell you axes are overrated.
The point is this: you risk getting no results if you have not first sharpened the axe.
The difference is subtle but fundamental: it is not about admiring what the AI manages to do on its own, but about discovering what you manage to achieve when the AI amplifies your abilities. The real goal is that, at the end, you can tell yourself: “I did that.”
How to really use it: the three magic words
1. CONTEXT
If the context is missing, you cannot expect an adequate answer. Full stop.
Think of an architect. You would never just say “design me a house”. You would say: “I have a 500 m² plot on a hillside, a budget of X, a family of four, I want three bedrooms, modern style, south-facing for natural light, and I have to respect these landscape regulations.”
With AI it is the same. Not “write me an app”, but: “Write me a booking system for a fitness centre, with OAuth authentication, a PostgreSQL database, Stripe integration for payments, a mobile-first interface, and it has to scale to 10,000 concurrent users.”
See the difference?
2. PROCESS
You have to break down what you do into processes. Any activity: software development, teaching, marketing, administration.
Take the process, break it apart, and ask yourself: which of these pieces deserves to be delegated to the AI?
A concrete example: teaching (which I did for years, alongside development).
❌ You cannot say: “Build me the lesson.”
✅ You can say instead:
- “I have prepared this outline. Develop the slides only for the ‘client-server architecture’ section, audience of sixteen-year-olds, beginner level, with practical examples.”
- “I taught this lesson to university students. Adapt it for fifteen-year-olds: simpler language, everyday examples, no technical terms without an explanation.”
- “I design the lesson, the sequence and the content. You generate ten multiple-choice quiz questions on these five key concepts, medium difficulty.”
Guiding principle:
- Keep for yourself: creativity, particularity, the human elements, the overall vision
- Delegate to the AI: what is not really your job, long/repetitive/tedious activities, specific operational tasks
3. ORCHESTRATION
The various tasks get carried out by these “autistic engineers”.
Every AI agent is hyper-specialised and rigidly focused on its domain. Just like a real specialist with autistic traits, it excels in one specific area but has precise limits:
- Narrow focus: you cannot ask it questions that are too wide-ranging or outside its field
- Dependence on explicit context: if a piece of information is missing, it does not infer it. It does not “read between the lines”
- Rigid patterns: it tends to follow established schemes and the statistical biases of its training
- Limited world view: it sees only through the lens of its specialisation
This is not a criticism, it is their nature. And knowing it is essential to using them well.
Your role is to orchestrate these specialists.
To put together what each of them tells you within its own field, to manage their misunderstandings, to correct any hallucinations, and above all to make the final decisions.
My “Eureka” assistant: a concrete example
Beyond coding, I have built a personal assistant made of several specialised agents:
- The Doctor: gives me health advice based on the data I track
- The Accountant: helps me manage invoicing and the finances of my practice
- The Communication Expert: analyses the job enquiries arriving from LinkedIn, so I can sidestep the ones that only waste time
Each one knows how to do its own thing. None of them knows how to do everything.
The value is not in the individual agent, but in my ability to orchestrate them. To bring their opinions together, to decide what to keep and what to discard, to steer, and to hold the overall vision.
In other words: I have not delegated my life to a superior intelligence. I have built a team of stupid specialists who, under my direction, let me operate on several planes at once. I remain the conductor. They are the instruments.
The skills of the future (and no, AI will not steal your job)
There is a refrain I hear everywhere: “AI will replace human beings, it will steal our jobs, we will become useless.”
It is an understandable fear, but a mistaken one.
And there is another fear, fed by decades of science fiction: “AI will become conscious, it will rebel, and it will wipe us out.” Terminator, The Matrix, HAL 9000. The same pattern: the machine takes control.
This fear is mistaken too. But for different reasons.
The problem is not that “we are downplaying the risks”. The risks of AI absolutely do exist: algorithmic bias, manipulation of information, concentration of power, amplified inequality. Those risks are real and serious.
But they have nothing to do with “artificial consciousness”.
However refined the simulations become, consciousness is something human. An AI can perfectly imitate the patterns of conscious thought, answer as if it had self-awareness, even “say” that it does. But it remains a simulation. Extremely sophisticated, convincing, yet still a simulation.
The real risk is not that AI “wakes up” and independently decides to dominate us. The real risk is how we humans choose to use it: to concentrate power, manipulate, exclude — or to amplify abilities, democratise knowledge, solve problems.
The machine has no will. It has no purposes of its own. It cannot “rebel”. It does exactly what it was designed to do, no more and no less. And that brings us back to the central responsibility: we are the ones directing.
AI does not replace the human being. It radically transforms which competences have value.
It has happened before. When computers arrived they did not eliminate accountants: they eliminated the ones who only knew how to add up. When Google arrived it did not eliminate researchers: it eliminated the ones who only knew how to find information without being able to evaluate it.
Now it is happening again. The competences that count have changed. And those who do not adapt will be left behind.
Here are the new fundamental skills:
1. DECONSTRUCTION
Being able to take a process you carry out naturally, almost automatically, and break it into pieces.
Turning the implicit into the explicit. Understanding what you do, why you do it, and which parts can be delegated.
It is not trivial. In fact it is one of the hardest things there is.
2. ORCHESTRATION
Managing a team of specialists (human or AI) who each see only their own piece of the problem — as every specialist does.
Giving each of them the minimum necessary context. Integrating partial outputs into a coherent solution. Deciding when a result is good and when it has to be redone. Knowing who to consult about what.
It is project management applied to AI. And it requires an overall vision.
3. LEADERSHIP (yes, with machines too)
You have to be able to stop an agent when it goes in the wrong direction.
A concrete example: the AI is writing code and, in order to fix the bug, decides to refactor the whole architecture you never asked it to touch. You have to step in: “Stop. I did not ask you to redo the architecture. Fix only bug X without touching the main component.”
Or: it is generating content and wanders off towards a detail that is marginal for you. You: “Stop. You are digressing. Get back to the original focus: section Y only.”
It is not like managing people (the AI does not take offence), but it requires assertiveness, clarity and the ability to recognise when something is going off course before it does damage.
4. CRITICAL THINKING vs PATTERNS AND BIAS
AI is trained on billions of examples. That means it reproduces the statistically most probable solution based on what it has seen.
If you ask “how do I lower my cholesterol”, it will give you the standard answer: more vegetables, less animal fat, oats for breakfast, oily fish, no fried food. It is the solution that works in 90% of cases.
But if your case is in the other 10% (say you hate fish, will not give up red meat, find oats disgusting, and the idea of eating nothing but salad depresses you), the AI will keep proposing the standard diet until you give it precise constraints: “I want to lower my cholesterol BUT: no fish, no oats, I want to keep meat twice a week, and I need strong flavours. Propose alternatives compatible with these constraints.”
The critical skill is telling the difference between:
- “The standard approach is fine, it is the right solution for this case”
- “No, this is different. I need something non-standard, for these specific reasons”
Without that capacity for discernment, AI will always drag you towards statistical mediocrity: solutions that work, but generic ones. Good for everybody, perfect for nobody.
Conclusion: sharpen the axe
AI is not magic. It is not an oracle. It is not an assistant that solves your life with one prompt.
It is a force multiplier: it amplifies what you know how to do, but it does not replace what you need to know. Without competence underneath, you are multiplying by zero.
Using it well takes method: clear context, decomposed processes, deliberate orchestration, an overall vision. In other words: you have to sharpen the axe before you cut down the tree.
When I started working this way, I discovered something: I was not “working less because I was delegating”. I was producing more, working differently. I concentrated on design, organisation and direction, and delegated the manual labour. You might discover something similar.
And in the end, when you look at what you have made, you can still say: “I did that.”
AI is a wave. You can let it flatten you, you can ignore it, or you can learn to surf. I chose the third option. What about you?
P.S.: If you have read this far and something clicked, good. The next step is to experiment. Take a process you know well, break it apart, and delegate one piece to the AI. Then watch what happens, correct, refine. For me it was the only way to really understand how it works.