Marc Andreessen

Marc Andreessen is a software entrepreneur and cofounder of Andreessen Horowitz whose AI work includes investment and public advocacy. He cocreated Mosaic and cofounded Netscape and Loudcloud. In 2026, he argued that reasoning models, coding agents and self-improvement were making AI practically consequential, and connected a16z’s investment in Cognition to his belief that automating software production would expand markets.[2][1][8]

In his April 3, 2026 Latent Space interview with Shawn Wang (swyx) and Alessio Fanelli, Andreessen called contemporary AI an “80 year overnight success”: a sudden practical payoff from decades of research. He recalled coding in Lisp in 1989 and rejected the idea that his interest in AI began with generative AI. He regarded reasoning, coding, agents and recursive self-improvement as successive breakthroughs, while acknowledging that enthusiasm and disappointment still move in cycles.[1]

His enthusiasm for Pi and OpenClaw centered on their combination of language models with familiar Unix mechanisms: shell access, a filesystem, markdown files and a recurring loop that wakes the agent. In his explanation, existing commands and command-line interfaces expose substantial computing capability without requiring every integration to be rebuilt around a new protocol.[1]

Andreessen argued that file-based state makes an agent less dependent on its underlying model: replacing the model can preserve memories and capabilities stored in files, although the agent’s personality may change. He also described moving agents between compatible execution environments and instructing them to inspect and rewrite their own files to add capabilities. He qualified that assessment by calling the systems prototypes with security issues still to resolve.[1]

Andreessen’s examples included a friend who reportedly used OpenClaw to rewrite firmware for a robot dog, then had the agent revise the code when problems arose. He distinguished that friend’s experiment from his own ownership of a robot whose conversational LLM was disconnected from its movement controls. The story illustrated his expectation that coding agents could repair existing devices as well as create new applications. His enthusiasm for people running unrestricted agent experiments included the qualification, “I’m not doing it myself to be clear”.[1]

He supported open models for both access and learning: freely available code and papers let others understand and reproduce techniques. He also expected local inference to matter because of compute shortages, cost, trust and latency. His examples ranged from tasks needing a modest local model rather than a powerful cloud model to wearable devices where a network round trip would be undesirable.[1]

Andreessen expected model capabilities to keep improving, but separated technical progress from adoption. He warned that new models could erase some companies’ claimed value while arguing that businesses would still be needed to bring AI into complicated institutions and everyday use. He also proposed that founders equipped with AI for paperwork, reports and other managerial work might run companies at greater scale without surrendering their inventive role to professional management.[1]

On September 9, 2026, Andreessen published “Investing in Cognition”, presenting a16z’s backing of Scott Wu and the Cognition team as an application of that outlook. He wrote that the firm had known Wu for over seven years and previously backed him at Lunchclub. His investment rationale was that cheaper software production makes previously uneconomical uses viable, increasing demand; automating software itself could therefore accelerate industries that depend on it.[8]

Software has been eating the world at the speed of human hands. It is about to eat the world at the speed of compute.[8]

Andreessen’s comparisons between agents and earlier computing systems draw on his browser work in the 1990s. Looking back in the 2026 interview, he emphasized human-readable protocols, readable HTML and view source as ways developers could understand systems and learn to build on them.[1]

Andreessen cocreated Mosaic at NCSA, where he was a student working on the project. Mosaic built on earlier browsers; NCSA’s history credits Eric Bina’s programming skill with making it easier to get running. In Tim Berners-Lee’s account in Weaving the Web, reproduced by NCSA, Andreessen maintained a near-constant presence in web newsgroups, listened to requested features and critiques, programmed improvements and repeatedly released versions for people to try.[5][2]

Andreessen then cofounded Netscape with Jim Clark. In his 2011 retrospective, Ben Horowitz recalled that they initially planned a video-distribution business for a proprietary network. During planning, they instead chose to improve browser security, functionality and usability, believing this could make the Internet the network of the future.[6]

Andreessen served at Netscape from April 1994 to March 1999, including as chief technology officer and executive vice president of products, and on its board. He then served as AOL’s chief technology officer from March to September 1999. He holds a BS in computer science from the University of Illinois at Urbana-Champaign.[7]

He cofounded Loudcloud, later Opsware, and became chairman in September 1999. His fellow cofounders included Ben Horowitz, president and CEO, and Tim Howes, CTO. After the company sold its managed-services business to EDS on August 15, 2002, it developed automation technology originally used internally into standalone software. That software automated tasks such as provisioning, configuring and patching servers and network devices. Opsware subsequently sold to Hewlett-Packard for $1.6 billion.[7][2]

As a venture investor, Andreessen set out his AI agenda in “Why AI Will Save the World”, published June 6, 2023. He argued that AI could augment human intelligence through tutors, assistants and scientific collaborators, and advocated rapid development and competition among established companies, startups and open source projects. He acknowledged that AI would help malicious actors, proposing enforcement against criminal uses and investment in defensive AI, including cyberdefense and biological defense.[3]

His Techno-Optimist Manifesto, published October 16, 2023, broadened that position into an argument for technological acceleration, markets and material abundance. Andreessen presented intelligence and energy as foundations for growth and explicitly endorsed augmented intelligence alongside artificial intelligence. These were statements of his philosophy and expectations about technology’s benefits.[4]