🎙 We held the eighth Lecture Club call on Stanford MS&E 435: Economics of the AI Supercycle.
Lecture 8 — Guillermo Rauch (Vercel). The topic was the economics of applications and AI coding: vibe coding and the future of software development. The lecture seemed marketing-oriented (Rauch spends half the talk selling Vercel), technically nothing new — so we quickly moved to economics and the main debate: can agents replace people?
Some thoughts from the discussion:
• Nikita (me, led the call) — Technically, nothing new in the lecture, Rauch markets himself for half the talk. But the deploy layer is economically underestimated: the profit pie has shifted towards Vercel, whose value is hiding server abstractions from vibe coders ('you change card details for access to an elastic server'). I'm already a user: Claude defaults to deploying on Vercel.
• Stepan (ML product, GigaChat) — Software is becoming almost free, and everyone who used to make software is now making agents: Software 2.0 → 3.0. Vibe coding at work every day and suggests the term 'vibeworking'. Like Tilda and Wix didn't kill programmers — the entry threshold just falls, and the demand for ML engineers will only grow.
• Irina (customer success, payment service) — Disagrees that support can be entirely handed over to agents: from a certain check, a client needs a live responsible person, and B2B is much more secure than B2C (a payment delay means 100 people without salaries). About coffee: a robot will brew more consistently, but people return to 'their' barista for the ritual.
• Philip (Applied Sciences, Zillow) — Through The Goal: simplify deployment — the bottleneck moves, and once you start paying, the same Claude will help you move cheaper, so in a year or two, the main beneficiary of the whole story is Vercel itself. Support is a matter of interface: recently, LLM support answered him in a minute.
• Jean (futurist) — Half of the speaker's numbers don't add up, 'Vercel = AWS for agents' is self-positioning. Software has become free and disposable, trust in institutions is at a historic low — that's why robots are trusted more. But vibe coding in its current form is 'info-gypsy', it will disappear in a year or two; the bet is that Stepan's office will grow from 10 to 20–30, developers will return.
• Nikita (robots and service) — Through Naval's podcast: interest fades in a robotic cafe — at equal price and quality, people choose a live barista for the ritual and networking. Robotics is the next wave after LLM, but the humanoid form is more likely a mistake; and as soon as a device starts working well, it stops being called a 'robot'.
• Nikita (closing) — Both hardware and token cost historically tend to zero — expensive only relatively. Two surveys of 100 people: they want a live barista, but a robotic taxi (routine → robot, emotions → human). Vibeworking will spread people across professions, but the club suggested not to lose touch — meet every two weeks.
📺 Recording of the call:
https://youtu.be/QMnwC7JYO88
