SWE is past the elbow of the exponential kickoff. I watched it happen in real time. Other fields are next.

Two years ago I was writing every line of code. A year ago I was prompting and reviewing. Six months ago I was running multi-turn loops manually - plan, implement, verify, fix, repeat. Last week I ran 63 automated steps on a complex codebase and walked away. Came back to 20,000 lines of working code. That's not an anecdote. That's three distinct 10x jumps in less than two years, and I lived through each one. Here's how the stack looks: Layer 1 - The models. Opus 4.6 and GPT-5.4 are not incrementally better than what we had in 2023.