AI Has Started Building Itself: Development Speeds Up, and the Main Human Skill Shifts to Choosing the Direction
What remains the central human skill when AI begins helping to create the next version of itself?
Understand why direction-setting remains human as development accelerates: which problem to solve, which constraints to set, and which result is acceptable.
What to watch for
Key takeaways
The decision in “As Claude itself is working, part 1/2” depends on one criterion: this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.
For the “Max” scene, the decisive point is this: a benchmark measures a narrow capability; working value requires repeatability, a clear cost, and control over errors.
The boundary of the “80% of the code created AI, part 1/2” case is defined by this point: this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.
The decision in “The main human skills in the AI era” depends on one criterion: this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.
The boundary of the “The leaders are separated from the market. Who lost part 1/2” case is defined by this point: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.
The decision in “3 scenarios for future AI from Anthropic” depends on one criterion: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.
The boundary of the “6 AI trends in the Silicon Valley” case is defined by this point: the risk depends on the scope of access, the scale of the consequences, and whether the system can be stopped and its actions reconstructed.
The practical meaning of “Exhibition” is that the risk depends on the scope of access, the scale of the consequences, and whether the system can be stopped and its actions reconstructed.
In the context of “OpenAI, DeepMind, Apple: different strategies in the AI race,” this criterion applies: this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.
What this episode is about
Anthropic acknowledges that Claude participates in developing its own systems, models write as much as 80% of the code in some projects, and OpenAI and DeepMind are building similar acceleration loops. Humans are not disappearing from the process, but mechanical production matters less while task definition and verification matter more.
When Claude writes code for the next version of Anthropic’s tools, AI becomes part of its own production cycle. This is not yet autonomous self-improvement in which a model defines the goal and releases itself without people. But the loop already exists: the current system accelerates the engineers building the next one.
The figure of eighty percent of code sounds like replacement of the programmer, but it describes a different distribution of work. The model generates large volumes, while a person defines the architecture, reviews, corrects, and decides what should be built at all. The more code AI produces, the more dangerous a bad direction becomes.
The central human skill is therefore not “write faster than the model.” It is formulating the problem, choosing constraints, and seeing where an answer looks convincing but leads in the wrong direction. This applies beyond developers. Marketing, product, editing, and management receive the same new division of labor.
Market leaders pull ahead because OpenAI, Anthropic, xAI, and Google have models, compute, and an internal environment in which AI helps create AI. Companies without such a system lose not one release, but the speed of every subsequent iteration.
Anthropic’s scenarios range from controlled acceleration to a system that becomes difficult to control. Apple, DeepMind, and OpenAI choose different strategies, but all arrive at the same loop.
The practical conclusion for an individual is that AI cannot remain a thirty-minute weekly exercise. It has to become a constant work tool, or the productivity gap will grow with every new model generation.
The practical conclusion for an individual is that AI will not remain a thirty-minute weekly exercise. As a result, it has to become a constant work tool, or the productivity gap will grow with every new model generation.
Episode transcript
The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 4 segments: 3 identified, 0 mixed, 0 marked with ✓, and 1 unresolved.
Loading…