scripod.com
86% of What Coding Agents Do Is Just Reading — Not Solving | Alexander Whedon of Subquadratic

Overview

Shownote

Highlights

Transcript

Chapters

Pins

86% of What Coding Agents Do Is Just Reading — Not Solving | Alexander Whedon of Subquadratic

Eye On A.I.

21 HOURS AGO
Eye On A.I.

Eye On A.I.

21 HOURS AGO
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

Every AI model in production today has the same hidden tax: doubling the context window quadruples the compute. That's what quadratic compute complexity means in practice, and it's the reason enterprises are spending most of their AI engineering budget on ...

Highlights

Craig S. Smith speaks with Alexander Whedon, co-founder and CTO of Subquadratic, about making million-token AI systems faster, cheaper, and more useful for enterprise data, coding, finance, security, and robotics.
06:56
86% of coding-agent steps are reading
10:45
Large context is still a product and alignment frontier
13:43
Dynamic attention makes sparse context practical
19:20
RAG will be transformed, not eliminated
23:04
Long context can make defensive security much stronger
29:32
Less curation can unlock enterprise data
40:25
Long-context reasoning is asymmetric across industries
47:10
LLMs are helpful, but not sufficient
50:53
Robotics needs a vast context window

Chapters

Why long context requires a new attention architecture
00:00
From narrow retrieval to million-token workflows
07:32
Training long-context models for enterprise reasoning
19:47
Reducing enterprise curation and expanding agents
28:33
Uneven reasoning, access, and the path to deployment
37:30
A longer-term vision for architecture and robotics
47:35

Transcript

Craig S. Smith: In Transformer models, the cost of attention grows with the square of the context length, So doubling context makes the compute roughly four times more expensive. And Alex and his team have solved this issue. Alexander Whedon: When we've d...