Say What You Want, and AI Will Try to Design the Protein
AI·October 28, 2026

Software for designing proteins and genomes has long demanded specialist know-how. Users typically have to set up sequence constraints, scoring functions or model inputs before a tool produces anything useful. A new report suggests that barrier is starting to fall: these tools can now take a plain-language description of the protein someone wants and work from that.
The idea is simple. Rather than translating a research goal into technical parameters, a scientist might describe the function, structure or behavior they are after, and the AI system turns that request into design work. The report does not make clear how closely the output tracks the request, or how much checking a researcher still has to do.
That caveat matters. Protein design models can produce candidates that look convincing on a screen and then fail in the lab, so a conversational front end mainly changes who can ask the questions, not whether the answers need testing. Lower barriers could bring more labs and smaller teams into genome and protein work. They could also put more designs in front of people without the background to judge them.
The headline's "vibe genes" appears to be a nod to vibe coding, the trend in which developers describe software in everyday words and let AI write the code. Whether that approach carries over cleanly to biology, where a flawed design can have real consequences, is the question to watch.
Reporting based on an external source.