Building scientific AGI to accelerate materials discovery

Executive overview

Materials R&D underpins every major industry — automotive, aerospace, semiconductors, energy — yet a novel discovery still takes 10+ years to reach commercial scale. AI agents that can read millions of papers, simulate billions of materials, and run thousands of experiments simultaneously can compress that timeline dramatically.

Radical AI is building full-stack scientific AGI starting in materials science, targeting inverse design: defining the problem first, then designing the material to solve it — rather than discovering materials and searching for applications.

The core insight: AI doesn't replace scientists — it removes the indexing bottleneck, freeing researchers to focus on what to build next rather than how to search for it.

Why materials science is the right target

• Every critical industry — manufacturing, defense, semiconductors, climate — is gated by materials R&D
• Current process: 10+ years and enormous cost from novel discovery to scaled material system
• Fragmentation of literature and slow experimental cycles make this an ideal AI problem
• AI can read across disciplines simultaneously; human scientists cannot

The AlphaGo analogy for science

• AlphaGo indexed more games than any human could study, then made a move no human had conceived
• Applied to science: AI indexes millions of publications, simulates billions of materials, tests thousands simultaneously
• This unlocks inverse design — start from the hardest unsolved problem, design the material from there
• Human scientists can do this, but only over very long timeframes; AI can do it fast

How Radical AI was formed

• Co-founder Jorge Gonzalez was tracking AI architectures at AlleyCorp and asked: why is AI only solving small problems?
• Joseph Krauss identified materials science as the field with the biggest gap between AI potential and current practice
• Third co-founder Herd Seder had already built an autonomous scientific lab at Lawrence Berkeley National Lab
• All three agreed this shift was inevitable — and that they had to be the ones to build it
• Pre-seed round raised in 45 minutes; sole investor was AlleyCorp's Kevin Ryan

Operating principles at Radical AI

• Mission over job: every hire is asked whether they want a job or a mission — wrong answer disqualifies
• 51% rule (adapted from SpaceX): when confidence reaches 51%, make the decision; debate beyond that costs more than the occasional wrong call
• Two filters on 51%: how significant is the decision, and what is the downside if wrong
• Failure is treated as a normal output of the discovery process, not a judgment on the person or project
• Culture stays aligned through shared belief in the mission, not process or hierarchy

Bias to action as a competitive advantage

• Kevin Ryan's lesson: strategy without action leads to dead ends
• Best entrepreneurs get things done and watch results unfold — then iterate
• Startups are hard; the edge comes from doing more, knowing more, and holding a thesis no one else has
• Joseph's approach: keep moving forward, connect the dots in retrospect (Steve Jobs quote on connecting dots backwards)