Thinking about
what we build
On AI engineering, physics simulation, and building software at scale.
Recent posts
Browse the archiveWhy Simulate the World
When thinking and building become cheap, proof from the real world becomes the slowest step in progress. A second world, built from the laws of physics, makes proving fast: machines test designs there and build only the winners.
Simulating the World
Language predicts a storm without resolving a single molecule. First-principles simulation resolves every molecule but cannot reach a single living cell. The way to simulate the world is a new tokenization of physics.
A Physics Company
The laws of physics have been known for a century. What those laws do, at scale, has never been computed — and that gap is where an entire industry is still waiting to be built.
We built a layout viewer for AI agents. Then an agent scaled it to 38 million shapes.
How a goal-driven loop turned PhotonForge's browser LiveViewer from a half-million-shape viewer into a 38-million-shape, 60 FPS tool for humans and agents.
Learning Auto-Routing by Building: From Brute Force to an Auto-Design Agent
How I learned auto-routing for photonic chips — a failed brute-force attempt, AI as a learning partner, interactive HTML arenas, a PhotonForge router, and an agent that iterates 27 designs in under three minutes.
Predicting Peak Memory for an Electromagnetic Mode Solver
How we replaced a heuristic memory estimate with a calibrated model for Tidy3D mode solver workloads, eliminating under-predictions across the calibration set.
Can AI Agents Autonomously Design Components on Photonic Chips?
We gave AI agents a photonic simulator, a DRC engine, and four design challenges. They autonomously designed waveguide bends, crossings, splitters, and demultiplexers — some reaching near-perfect performance.
Designing a Photonic Chip Component with ~45 Lines of Python
A compact introduction to photonic inverse design with Tidy3D, using a pre-built simulation and a ~45-line optimization loop.
"What Should We Work On Next?"
The story of building an 80,000-line autodiff library almost entirely through AI agents — and the verification infrastructure that made it possible.
The Agent Control Loop — Engineering for Tolerance
Why agent reliability isn't magic model behavior — it's an environment where correctness is continuously verified. A framework for deciding when and how to delegate to AI agents.