The threshold point in AI-powered coding
Copilot and Cursor always started with optimism and never sustained a predictable outcome. On 13th Feb that changed — and six months later it has held.
I used Copilot and Cursor before 13th Feb 2026 and tried to solve some specific non-trivial software and hardware problems. It always started with a lot of optimism — however, it could not sustain a predictable outcome.
Come 12th Feb, a great friend, Tarun Beri, asked me to try Claude Opus (4.6 at that time). Rest is history. The next day I subscribed to Claude Pro, and in two days upgraded to the Max plan, and never looked back.
Now I am in weekend founder mode, and finding that Claude is not only a software development powerhouse — there are unconventional applications where it scales superbly.
Taking hints from my experience, and anticipating near-future trend shifts in the embedded domain with AI disruptions, I am close to finishing an AI workbench for hardware and electronics engineers that allows closed-loop development using Claude or other AI tools.
I came a long way on how to structurally apply Claude Fable 5, Opus X, and Sonnet X to get the best out of the buck.
Here is what my personal git repo history looks like after 13th Feb.
