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

September 8, 2026

AI Summary

5 min read

In a benchmark of frontier coding models, Alexander Whedon’s team at Subquadratic found that 86% of the steps an agent took were read steps — just scanning and engineering context before the actual execution, which accounted for only the last 14%. That ratio captures the central problem Subquadratic is built to solve: the cost of context, in both compute and latency, is strangling the usefulness of large language models, especially as they move into agentic workflows. Whedon argues that the bottleneck isn’t just context window size — it’s context intelligence, cost, and latency, and the trade-offs enterprises make among them.

The Quadratic Bottleneck and Sparse Attention

The technical root is straightforward. In standard transformer attention, every token must compare itself to every other token. Double the context length, and the compute cost roughly quadruples — that is the quadratic complexity. Sparse attention tries to avoid this by only computing a subset of token relationships. The challenge has always been picking the right ones without sacrificing quality. Fixed patterns, like sliding windows that only look at nearby tokens, are brittle because relevance is content-dependent, not positional.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

Never miss an episode of Eye On A.I.

Get every new episode summarized in your inbox — free, ~5 minutes to read.

No spam. Unsubscribe anytime.

What you'll learn

1 Timestamped Outline
2 (00:20) **The 86% Reading Problem** - Alex reveals that benchmarking shows 86% of coding agent steps are just reading, not solving
3 (02:03) **What Subquadratic Does** - Alex introduces himself and explains the company's mission around quadratic compute complexity
4 (02:50) **Sparse Attention Explained** - A clear breakdown of what sparse attention is and why it matters
5 (04:30) **The DeepSeek Problem** - Why existing dynamic sparse attention still has quadratic costs
6 (05:12) **Subquadratic's Unique Approach** - How they solve dynamic selection without the quadratic cost
7 (06:08) **The Four Axes of Context** - Alex reframes the problem beyond just context window size

+ Full timestamped outline available in the app

Show Notes

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 context management rather than on the actual problems they're trying to solve. Alexander Whedon, co-founder and CTO of Subquadratic, joins Craig Smith to explain how SubQ's sparse attention mechanism eliminates that tax, achieving 40 times faster inference and 64 times less compute than standard attention at one million tokens, and what becomes possible when that constraint disappears. The conversation covers striking benchmark findings: 86% of what frontier coding agents do is "read steps," just trying to gather and organize context before the actual problem-solving begins, and frontier models drop well below 50% accuracy on financial document analysis at 500,000 tokens, revealing how asymmetric long context capability actually is across industries.

The most commercially important argument in this episode is about enterprise data. Most large organizations are sitting on hundreds of billions of tokens of data they've never been able to put to work in an AI product, told they need a $10 million data transformation project before they can even start building. Alex's core claim is that SubQ's architecture makes that barrier no longer necessary, enabling enterprises to process far more of their data with far less curation, at a fraction of the cost. He closes with what he describes as the most important and underexplored frontier in AI right now: we are still very far from understanding what users actually want from models reasoning over millions of tokens, and the product and alignment work needed to answer that question has barely begun.

Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.