byte | sumit sute

██████╗ ██╗ ██╗████████╗███████╗

██╔══██╗╚██╗██╔╝╚══██╔══╝██╔════╝

██████╔╝ ╚████╔╝ ██║ █████╗

██╔══██╗ ╚██╔╝ ██║ ██╔══╝

╚█████╔╝ ██║ ██║ ███████╗

╚════╝ ╚═╝ ╚═╝ ╚══════╝

Some thoughts don't need essays. They just need somewhere to land. Bytes are the small ones: thoughts that appear mid-commute, mid-debug, mid-coffee. I post them from Telegram or the terminal, whichever is open, and whichever I'm probably procrastinating in when the brain buffer overflows. Stay connected on the telegram channel for latest bytes.

Most agent regressions are process regressions, not intelligence regressions. Better scaffolding for repository conventions, explicit tool/library preferences and tight self-review gates helpsapr 20

xxx

Writing about my agentic experiences has made a plateau clearer: still operating in a rudimentary loop. One step at a time, occasional plans, occasional one-shot or sometimes planned orchestration, but little structure that carries forward.apr 17

xxx

With agentic workflows, instincts atrophy. Not because the knowledge disappears, but because the activation pattern changes from a shift in when and how it gets activated.apr 16

xxx

Goodhart's law shows up everyday now. The system keeps producing, the bar just moves. Once the output becomes the target, quality stops being the constraint.mar 31

xxx

Careful reading as a skill is like returning home in the vibe-or-ngmi age. Eventually, interfaces will become truly multimodal, until then, really slowing the fuck down. to read. To read carefully. Building tighter workflows.mar 31

xxx

Consuming generated content, especially writing, is still not normalized. It's a little murkey whether that's because the output feels plastic or unnecessarily verbose, or because it now takes longer to read something than it did to produce it.mar 31

xxx

When something arrives already structured and reasonable, it quietly removes the impulse to interrogate it. That absense of questioning is harder to notice than the mistake itself.mar 27

xxx

Asking the agent not just for the output, but about the assumptions it carries, the boundaries it ignores, and the part of the system it quietly takes ownership of. That's where usually the real gaps aremar 27

xxx

There is kind of debugging loop with AI where you keep asking follow-ups until the problem disappears. It works, but it also shifts the focus away from understanding what was actually happening.mar 27

xxx

The difference between asking follow-up questions to improve a solution and asking them to improve your understanding is subtle, but it compounds quickly.mar 27

xxx