"Bots control bots." a16z co-founder reports the advent of AGI - ForkLog: cryptocurrencies, AI, singularity, the future

# “Bots control bots.” a16z co-founder says the arrival of AGI

Artificial intelligence has already crossed the AGI threshold. This was stated by Marc Andreessen, co-founder of Andreessen Horowitz (a16z), in a three-hour episode of The Joe Rogan Experience.

According to the investor, the turning point happened “about three months ago” after the release of the latest generations of models from OpenAI, Anthropic, Google, and xAI — GPT-5.5, Claude 4.6, Gemini 3, and Grok 4.3.

The investor did not reference an industry consensus or a scientific definition of AGI. In the podcast, it sounded like his personal assessment of what is happening.

Andreessen argues that today’s models “in 99% of cases” produce answers better than almost any expert available. He calls the new versions of LLMs a “qualitative leap” and a continuous cognitive amplifier — a tool for writing text, building arguments, breaking down complex topics, and finding solutions.

The expert described his own approach to working with AI as a multi-step simplification of the response: first “explain it like to a ten-year-old,” then — “like to a five-year-old,” then — “like to a two-year-old.”

Medicine

The a16z co-founder said that during his illness he used chatbots and called such systems “the best doctor in the history of the world” — because they are available around the clock and can guide a patient step by step.

He also said that doctors are already using ChatGPT at scale. “Everyone” after seeing a patient asks the LLM “what’s going on with this person.” Some input the data right during the consultation.

Earlier, the American Medical Association reported that 81% of doctors in the United States use AI in their professional activities. The organization emphasized that such systems should remain auxiliary tools and be applied in compliance with transparency and safety requirements.

Andreessen also described multimodal medical scenarios: a user uploads blood test results, after which the neural network “will tell you what’s wrong with you.” As an example, he cited an acquaintance who assembled a personal AI health panel “in the spirit of Star Trek.” The investor also mentioned the decline in the cost of whole-genome sequencing to roughly $200.

Programming and AI agents

According to Andreessen, in Silicon Valley the current “cutting-edge level of development” is working with about 20 AI bots at the same time. They perform tasks in parallel 24/7, while a person “checks the result every 10 minutes” and provides feedback.

The next stage is agent hierarchies. Each bot will have its own “sub-bots,” and above them — bot managers. Within a year, Andreessen predicts, a model with “10–20 virtual assistants, each of which then has another 10–20 helpers” will become routine.

Later, this will grow into multi-level structures where “bots control bots that control other bots.” In such a configuration, one developer could oversee up to 1000 agents and become “1000 times more productive.”

Science

Andreessen said that AI is progressing especially quickly in areas with “provably correct answers” — mathematics, physics, chemistry, biology, and programming. According to him, systems are already solving problems that humans could not handle.

Next are creating new drugs, treating cancer, accelerating space research, and discovering “new physics.”

Recall that in May, an OpenAI AI model refuted Paul Erdős’s 80-year-old hypothesis about unit distances.