We are left with thinking, again

15 May, 2026

Give yourself a break. AI is here now, whether we like it or not. It has a cultural impact that is driving people crazy. People love it, people hate it, some would build statues to Dario Amodei while others literally tried to burn Sam Altman’s house down. the cultural confusion is palpable.

I believe this confusion also flows from the fact that AI is first and foremost a cultural innovation. Cultural innovations are the ones that affect the mind first, then the surroundings. Language is a cultural innovation, as it changed the way communities interact. One of the reasons why AI is so pervasive is exactly that - it changes the mind and the way we interact. We were discussing the other day in the office how meetings changed in the age of AI. People are contributing with context & instructions as they know that transcript will become useful to train an agent. That is wild. What other pieces of software changed the way people communicate? Printing changed reading and writing. The internet changed the way we access information on global scale and opened up access. That’s basically it.

Language. People go around complaining about AI slop, AI fatigue. They are right, but that’s a feature and not a bug. It is a signal of cultural stagnancy. People will unfortunately almost always cut shortcuts to achieve results. The issue here is that results are so painful to read that they break the social contract that prescribes that time spent to consume content should be at least equal to time spent to produce it. in fact, this arbitrage closes pretty quickly as people cry their frustration to just another dodgy blog post. I feel sorry for those who can’t differentiate slop from good content just because AI was used in the process of getting to it.

A way that helps me see AI is through an equation: Output quality = f(Intent, Prompting)

I intentionally keep this equation simple, and before I oversimplify this, I’ll state my assumption. This piece is not a deep dive into the way models convert language into action. There are better places to read about this - for example this is a fancy visualisation of how neural networks visualise language in shapes. Also, as a love letter from investor to founders - please just assume we consider context solved. I know it’s an oversimplification, but context is undifferentiated, and foundation models are already going after it (the OpenAI Deployment Company?). That is why I believe it will be commoditised.

So what are the truly unique, cultural and meritocratic factors that remain?

Prompting. Ah, prompting. the real step change. this little box in which people can type whatever they want and get a machine to spit out value in no time (and compliments! “what a fantastic endeavor, Andrea!” to me asking the best way to wash my gym shorts). Prompting really became the action behind the rise of AI. People used to prompt other humans, but now that has become the language to talk to machines.

A non-programmable language. This is by far the most important change. This is what worries engineers the most and makes normal people feel invincible. Code has always been programmable - people learnt a new language that was built for machines and that is what made the first wave of AI almost lethal to coding. It was built around rules that can be recreated in the world of AI. Yes, some hiccups here and there, but we are increasingly saying goodbye to traditional coding through new code ships like Cursor, Claude Code and Codex.

We are now in the realm of non-programmability. This creates a set of problems. Aside from the major one - the dangerous reward mechanism that allows thoughtless people to obtain value by inputting minimal effort - the major problem is the cultural barrier that is being created. Prompting is an art that at the moment rewards a specific part of the population - English-speaking, Western-educated users. And that is what creates an incredibly strong bias from the people who can’t extrapolate themselves from the population they belong to.

My dear Native English friends - as a non-native English speaker, trust me when I say prompting is not solved at all. Studies have shown LLMs not only produce lower-quality output for non-native English speakers, but actively degrade quality when they detect non-nativeness.

That is about prompting, the communication angle to the machine. But there is a bigger, pre-linguistic issue here at play that I feel nobody is talking about as they can’t properly prompt it in their claude boxes in a self fulfilling loop. It’s intent. The incredibly powerful intent.

Intent comes from the word intendere, meaning stretching towards (→). It’s the combination of two things: direction and discernment. without anywhere to go, you are lost. what is it that you really want? and how do you know when you get it?

When Michelangelo thought about the David, he explained how he “saw the angel waiting [for him] in the marble”. Intent is knowing what we want to build - prompting is the exercise of picking the shovel into the marble. He, like all the artists, could visualize output in his design process. that is what built his success. the output is a derivative of linguistic expression, but the direction and discernment are pre-conditions.

That explains why product designers and owners who have been educated to translate intent thrive in this environment. Because they have always embedded the key ability to communicate, to an audience, direction and discernment. As the hearer becomes a model, they are able to maximise both factors of the equation.

So the issue with the slop is not its presence. It’s that we are suffocating the ability to build and capture intent after years of programmable thinking and laziness of thinking. We need a way to transition back the power to the design process. that has to lead again. it’s the most painful process as most of the time it has no meaning in itself and we might not have the quick feedback loop we have been used to. so a great way to start, in no particular order:

write stuff - document your thinking
grammar, logic and rhetoric should be the basics of any curriculum. read books, study languages.
disagree and use that anger for a thought you disagree with as a base to build your thinking.
thinking takes time as it’s driven by your system 2. also it’s the necessary effort to maintain the social contract with the audience.
study the masters with proximity to stay as close as possible to intent (not output). how come someone who has interviewed a thousand VCs remains a (good) investor yet one who has worked 10 years with a great VC goes out to become a great one?

and by the way we are still scraping the surface of what can be achieved with AI. to put things in perspective, MCP is 18 months old. context wasn’t a thing until then and investors are now pricing that in as commoditised. we’ll directionally solve context, we’ll “scientify” the art of prompting to cater to a wider audience and “cross-collars” (yes please), we’ll find ways to sketch, build, capture intent (also yes please), among many other improvements foundation models will bring. but the key cultural question remains: just as AI will take over pretty much every job (but VCs, luckily. we are safe if Mark says so), the last piece of art and differentiation will be intent. How do we preserve that in each of us?