For every property you optimized, every failure mode you mapped, there are a thousand more you didn’t. A single cryptic promoter or off-target effect can derail months or years of work. Any bioengineer will tell you: projects die from blind spots, not just from a lack of optimization effort.
The field is hellbent on optimizing designs for a handful of properties at a time. This misses the most important half of the problem. Can we engineer away unwanted effects systematically? Is negative engineering possible?
Our answer is yes. This mission requires radically new foundation model architectures, a more ambitious scope of training than ever before, and the tools to make all of it usable.
So we’re building it. Stay tuned.