Karpathy is right, software is about to move at lightspeed.
the interface layer of the internet gets sanded down and rewired as agents become the primary users. It'll go headless fast (API/MCP becomes the product). I see this happening in four ways:
1) Fully headless - high intent flows like travel booking, hotels, flights, ride hailing, delivery, reservations, commodity ecommerce (electronics, household goods), bill pay, insurance quotes, etc, not to mention most of SaaS. These UIs are mostly forms and funnels. Agents fill forms better than humans.
2) Thin UIs - amazon-style browsing, real estate search, job marketplaces, trading platforms, app stores and plugin marketplaces. Agents handle execution. Humans still want some oversight and discovery.
3) UI remains core (experience is the product) - social networks, games, streaming and sports fandom, fashion, luxury, beauty, dating, creator platforms, brand-driven commerce, high-consideration purchases (cars, homes). Taste still needs a surface.
Then a new category appears:
4) agent-native networks - agents discover tools, negotiate pricing, route tasks, build reputation, coordinate with other agents. Humans mostly watch. Moltbook is an early glimpse.
tl;dr a lot of websites become "inventory" for agents. Brands become trust layers on top of APIs
Caught up with
@karpathy for a new
@NoPriorsPod: on the phase shift in engineering, AI psychosis, claws, AutoResearch, the opportunity for a SETI-at-Home like movement in AI, the model landscape, and second order effects
02:55 - What Capability Limits Remain?
06:15 - What Mastery of Coding Agents Looks Like
11:16 - Second Order Effects of Coding Agents
15:51 - Why AutoResearch
22:45 - Relevant Skills in the AI Era
28:25 - Model Speciation
32:30 - Collaboration Surfaces for Humans and AI
37:28 - Analysis of Jobs Market Data
48:25 - Open vs. Closed Source Models
53:51 - Autonomous Robotics and Atoms
1:00:59 - MicroGPT and Agentic Education
1:05:40 - End Thoughts