What changed
Anthropic told investors on November 10 that its Mythos AI wrote almost all the code needed to run Claude on AMD chips in about three days. That same job used to take six to twelve months with human engineers. Claude handled the migration with almost no human help.
The work covered prompt processing for one model on AMD's MI355X chips. Anthropic said Claude wrote or ported nearly 100 percent of the specialized code, called kernels.
What you can do with it this week
Look at the technical tasks your team avoids because they take too long. Migrating data between systems. Rewriting a process to work with new software. Setting up integrations nobody has time for.
The tools that did this exist today. Claude, ChatGPT and other frontier models can read documentation, write code, test it, and fix what breaks. You don't need a developer on staff. You need someone on your team willing to describe the problem clearly and check whether the output works.
Start small: pick one annoying technical chore this week and give an AI the instructions a junior hire would get. See what comes back.
What to keep in your own hands
Anthropic had expert engineers watching. You should too, or at least someone who knows what good looks like in your context.
Don't run AI-generated code that touches customer data, financial records, or anything regulated until a person has read it line by line. And keep a manual backup: if the AI-built integration breaks, you need a way to do the task by hand while you fix it. The speed is real, but the responsibility for what it builds stays with you.