Coding agents won't replace decades of painstaking systems research, argues AI researcher
charles_irl · x · 2026-09-27
charlesirl pushes back on the idea that coding agents make decades of collaborative, painstaking systems research obsolete, arguing the opposite: the expensive human processes supporting that work remain essential in the AI era.
The thread also lists high-quality prior work their team drew on, including NVIDIA FlashInfer's cascade attention, Hydragen attention from Scaling Intelligence Lab, Hellerstein & Stonebraker's selectivity ordering, and IBM's classic System R paper.
Key takeaway: rigorous research methodology and the human systems behind it are not replaced by agents—they're amplified by them.
More from AGI Musings
- LLMs Are 'Stupid' Yet Productive at Code: Brute Force Beats Intelligence — ivan_bezdomny · 2026-09-27
- Yacine: AI UX shelf life is about a month — learn Unix terminal, build your own tooling — yacineMTB · 2026-09-27
- CS professor: what matters from a CS degree is learning how to think — CSProfKGD · 2026-09-27
- Why You Shouldn't Let Claude Write for You: Lived Experience Is Irreplaceable — vykthur · 2026-09-27
- CS professor: manual coding will become secondary for the vast majority — CSProfKGD · 2026-09-27
- AI bubble talk cools, but $3 trillion annual revenue bar still looms — josh_wills · 2026-09-27