In Favor of Proving

gravatar
 | 
October 31, 2025
 | 
2 min read
Featured Image

Most AI talks a big game. In science and engineering, talk is cheap. If a model can’t show its work—if it can’t prove why a design passes, a device behaves, or a process stays stable—it’s just an autocomplete with good manners.

Our thesis is simple: Verified Scientific AI will become the default way real-world R&D gets done. Not “trust me” assistants. Not vibes. Systems that combine formal math, physics, and tool-connected agents to deliver results you can audit, rerun, and sign off on.

Why now? Because the bottleneck has moved. Multi-physics stacks, chiplet systems, photonics, thermal, mechanics, control—each adds a failure mode. Teams don’t need more suggestions; they need operators that can search equations, fit models, run sims, drive instruments, and then produce a proof or a back-to-back check that says: this holds. That’s how you take AI from “assistant” to “colleague.”

If you’ve ever shipped silicon or tuned a resonator, you know the last 10% is where schedules die. A verified approach flips that. Embed provers where decisions happen. Bake physics into the loop. Let agents talk to the lab and the CAD, not just the keyboard. The output isn’t a paragraph—it’s a verified step in the flow: an equation that checks out, a model that fits within bounds, a test that closes over the spec. That’s how you cut cycles, not corners.

We’re already seeing this inside our portfolio at Two Small Fish. One team’s lab stack no longer waits for a human to guess the next measurement; the system proposes it, runs it, and justifies it. Another team’s design flow catches protocol and stability violations before a person ever sees a waveform. Different domains, same pattern: prove first, move fast later. The payoff isn’t incremental UX polish; it’s weeks off a program and confidence at tape-out or sign-off that actually holds.

If you’re building toward this—operators that can prove, not just predict—we want to hear from you. At Two Small Fish, we invest in the infrastructure that turns ideas into atoms. Verified Scientific AI is that infrastructure. Let’s make “show your work” the new default.

tsfv-linkedin-1

MetroCentre - King Tower
225 King Street West, Toronto

Espace CDPQ
3 Place Ville Marie, Montreal

120 East 23rd Street - 4th Floor
Manhattan, New York

© Two Small Fish Ventures. 2026

tsfv-wordmark

MetroCentre - King Tower
225 King Street West, Toronto

Espace CDPQ
3 Place Ville Marie, Montreal

120 East 23rd Street - 4th Floor
Manhattan, New York

© Two Small Fish Ventures. 2026

tsfv-wordmark