Portrait of Anjali Shrivastava

Hi, I'm Anjali

I'm researching inference economics through the Analogue Group.

My work focuses on AI's intrinsic cost unpredictability, and what that means for unit economics, capacity planning, and financial measurement.

Before this, I led monetization data science for Adobe's new products team and GTM Analytics for Tana.

Selected Projects

Full collection of side projects, experiments, and half-finished things lives in /vintage

Essays

  1. A token is not a fixed unit of cost (Google Doc)
    Tim O'Reilly Podcast DSP Podcast Tweet
    TL;DR
    • The Nth token in a conversation is an order of magnitude more expensive than the first
    • Variable costs destabilize per-token API pricing, and Cursor and Anthropic's pricing changes indicate this
    • Heterogeneous usage leads to fat tailed risks that compound with usage growth
    • Tail risk represents lower margins, service degradation and system outages in extreme cases
  2. Why fat tails emerge at scale (Google Doc)
    TL;DR
    • A few long-context requests can explode memory use since each carries its own non-shareable KV cache
    • Because real-world traffic has near-infinite variance, the mean cost per request doesn’t converge
    • Without live telemetry on cache pressure, batching logic and P/D ratios, providers can’t optimize unit economics
    • The emerging pricing fix is priority contracts that turn unpredictable per-request costs into a forecastable aggregate-capacity problem
  3. Intelligence is not an API Call (Suggestions welcome)
  4. Notes on where value will accrue in AI (Suggestions welcome)
  5. Opportunities for mechanism design (Suggestions welcome)

I also write about things beyond pricing. See vintage/writing.

Diagrams

I used to build data viz professionally; more of that work lives in /vintage/work.

Some things I 100% recommend:


Open Research


I'm treating inference economics as an open research question: how AI businesses should measure cost, value, usage, and margin when the work being sold has no stable atomic unit. The best part of public-facing work is when others push my ideas into new territory. Like when:

I also publish my drafts in public Google Docs. You can see revisions, debates in the comments, and where my perspective changes.


If you're thinking about any of this, I'd love to compare notes. I'm at anjali.shrivastava99@gmail.com and on Twitter @anjali_shriva.

View the old version of this site at /vintage