Andreessen Horowitz cofounder Marc Andreessen says AI beats humans in coding because 'AI never gets drunk, sick, or high' and never files 'HR complaints.'

Marc Andreessen has a reputation for not watering down his takes. On a recent episode of the Joe Rogan Experience, the venture capitalist and co-founder of Andreessen Horowitz made his position on AI coding agents clear: bots, he said, are just better workers, at least on paper.

Andreessen told Rogan, laughing, “The bots never get frustrated with you.” A bot, he added, “never gets drunk, never gets sick, never gets high” and does not file HR complaints. That’s a catchy soundbite, but the argument behind it is something the tech industry has been slowly building towards for years.

Running 20 bots like a mini corporation
According to Andreessen, the gold standard in Silicon Valley right now is running about 20 AI coding agents at the same time. Each bot works independently on a task and checks back with its human manager for feedback every 10 minutes or so. This results in a near-continuous output loop that runs 24/7, no sick days, no lunch breaks, no end-of-day fatigue.


“This is why people can't go to sleep,” he said. “You've got 20 AI bots that are all as good as the best programmer in the world, doing exactly what you tell them to do.”

This agrees with a large-scale study published to SSRN, where randomized controlled trials were run across Microsoft, Accenture, and a Fortune 100 company with a total of close to 5,000 developers. According to the study, developers who had access to GitHub Copilot saw a 26% average increase in productivity, the equivalent of turning an eight-hour day into ten hours of output, with no measurable decrease in code quality.

The emotional math of AI versus humans
Speed isn’t the only argument Andreessen made. He described a common workplace scenario: a developer spends two weeks building something, is told he's wrong, fixes it, then is told the original was better after all. “The guy gets pissed at you,” Andreessen said. The bot? It says no problem and tries 12 more times.

There is something true in this observation. Human frustration, ego and emotional investment in work are features of creativity, but they can also slow things down. The output of an AI agent is not owned by it. That’s not a small thing when you’re iterating fast.

This is just the beginning
Andreessen sees this escalating quickly. He said he thinks that within a year, it will be normal for each person to have 10 to 20 AI agents, and each of those agents will have 10 to 20 sub-agents. Basically, a full corporate org chart powered by code.

Nvidia CEO Jensen Huang had this vision echoed in a recent earnings call when he predicted a future full of billions of AI agents, each of whom uses tools as people use PCs today.

And it won’t end with coders. “It's going to be every writer, every lawyer, every doctor,” Andreessen said.

That prediction is already backed by institutions. The McKinsey Global Institute’s November 2025 report, ‘Agents, Robots, and Us,’ proposes that current technologies could theoretically automate activities that make up roughly 57% of US work hours, and the demand for AI fluency in US job postings has increased sevenfold in just two years.

So what does that leave us with?
Tools such as OpenAI’s Codex and Anthropic’s Claude Code are already changing the way software engineers work and live. Some developers have strange hours, working late to take care of their agents, as if they were factory floor managers on the night shift. They even have a nickname: “AI vampires.”

The transition is happening fast, and Andreessen’s framing of bots as tireless, emotionless, always-available teammates captures something the numbers are beginning to confirm. For workers everywhere, the question is not whether AI is coming for some tasks. The question is, are they going to be running the bots or waiting to be replaced by them.