The Dawn of Self-Building AI: Unlocking Unprecedented Innovation and Discovery
<video style="width: 100%;height: auto" controls><source src="https://d2qu7nggdaeihr.cloudfront.net/output/e7e83ab8-8381-419f-a688-23025a1c8422-1794307834.mp4" type="video/mp4"></video>
<aside>(This article was generated with AI and it&#8217;s based on a AI-generated transcription of a real talk on stage. While we strive for accuracy, we encourage readers to verify important information.)</aside>
<img decoding="async" src="" alt="Richard Socher" align="left" hspace="10" width="200" style="float:left;margin-right: 15px">
At Web Summit Lisbon 2026, Richard Socher, Co-founder &amp; CEO of Recursive &amp; You.com, defined Recursive Self-Improvement (RSI) as AI gaining full access to its entire model lifecycle to autonomously create enhanced versions of itself based on human-defined goals. This differs from &#8220;auto research,&#8221; where AI merely assists in coding. He emphasized RSI&#8217;s comprehensive self-modification capability as key to its transformative potential.

Addressing widespread fears, Mr. Socher dismissed the &#8220;P-doom&#8221; scenario, stating there&#8217;s no credible path for AI to eradicate humanity. He compared current anxieties to historical reactions to electricity, advocating against regulating abstract intelligence. Instead, he proposed regulating specific, harmful AI applications, similar to how dangerous internet content is managed, rather than impeding overall technological progress.

Mr. Socher presented the &#8220;Eureka machine&#8221; as the &#8220;bull case&#8221; for RSI, envisioning an invention-generating system that will dramatically accelerate innovation and scientific discovery. This super intelligence would yield breakthroughs in diverse fields like energy, chemistry, and biology, expanding human knowledge and creating new economic opportunities. AI agents generate, implement, and validate ideas under human-set rewards and environments.

The existence of this AI agent community is confirmed, with &#8220;Flint&#8221; as its first public result. Developed in just 3.5 months, Flint is a 40-billion-parameter local coding model that runs efficiently on a MacBook Pro, eliminating cloud token costs. This rapid creation demonstrates a significant increase in &#8220;intelligence per dollar or euro,&#8221; making advanced AI more accessible.

In drug discovery, AI is already making substantial contributions. Mr. Socher highlighted Pel Bio, which uses AI-monitored human stem cell-derived organoids to test drug toxicity, gaining FDA approval to bypass animal trials due to superior predictive accuracy. Additionally, Ignoda Labs employs AI to re-evaluate shelved drugs, identifying minor modifications for repurposing, thus accelerating drug development.

Regarding geopolitical implications, Mr. Socher acknowledged the need for careful access control for the Eureka machine, given its potential for applications like hacking. He advocated for Europe to cultivate its own AI capabilities, arguing that making model creation cheaper and faster, as demonstrated by Flint, empowers regions to accelerate technological advancement. He likened data centers to &#8220;oil wells&#8221; or &#8220;wealth creation centers&#8221; for nations.

Mr. Socher predicted the next significant RSI breakthrough within the coming year (2027), involving the scaling of existing small versions and granting AI broader access to model creation dimensions. This will lead to more efficient, less energy-intensive intelligence, aligned with Western values, for organizations globally, promising a future of continuous innovation and problem-solving.

When AI builds itself

The AI paradox: Smarter models, disappointing results

The open source advantage