]> git.otsuka.systems Git - cotsuka.github.io/commitdiff
add inference draft 2026-23
authorCameron Otsuka <cameron@otsuka.haus>
Wed, 3 Jun 2026 17:00:05 +0000 (10:00 -0700)
committerCameron Otsuka <cameron@otsuka.haus>
Wed, 3 Jun 2026 17:00:05 +0000 (10:00 -0700)
content/articles/tokenmaxxing-as-management-theory/index.mdx [new file with mode: 0644]
content/articles/tokenmaxxing-as-management-theory/tokenmaxxing-as-management-theory.png [new file with mode: 0644]

diff --git a/content/articles/tokenmaxxing-as-management-theory/index.mdx b/content/articles/tokenmaxxing-as-management-theory/index.mdx
new file mode 100644 (file)
index 0000000..1d89c25
--- /dev/null
@@ -0,0 +1,63 @@
+---
+title: 'Tokenmaxxing as Management Theory'
+type: newsletter
+date: 2026-06-02
+modified: 2026-06-02
+description: or, I sure hope finance doesn't this bill...
+publication:
+  name: Inference Draft
+  issue: 2026
+  volume: 23
+tags:
+  - ai
+posse:
+  Substack: https://inferencedraft.substack.com/p/tokenmaxxing-as-management-theory
+---
+
+import { Picture } from 'astro:assets';
+import YouTube from '@components/youtube.astro';
+import tokenmaxxing from './tokenmaxxing-as-management-theory.png';
+
+<Picture src={tokenmaxxing} alt="Corporate tech hub in cyberpunk workspace" />
+
+Rolling out new technologies, procedures, and patterns within a company is historically difficult. An [NBER study](https://www.nber.org/system/files/working_papers/w9287/w9287.pdf) found that by the end of 2000, 88.6% of companies had adopted the internet into their business, but **only 12.6% found it enhanced their business processes**.
+
+This process is playing out once again with AI, with similar levels of companies adopting AI and varying reports of enhancement: [5%](https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf), [15%](https://www.deloitte.com/global/en/issues/ai/ai-roi-the-paradox-of-rising-investment-and-elusive-returns.html), etc. Just like the internet era, companies have a desired outcome: use the new technology in ways that increase productivity, improve quality, and develop new ideas.
+
+If you are management at one of these companies, how do you push the necessary tool, process, and hierarchy changes to enable the enhancement? One way is to announce to all employees that they need to be "tokenmaxxing," or using as many tokens as possible with the hopes that it leads to enhancement. An Anthropic employee [reportedly used $150k of tokens in a month](https://www.nytimes.com/2026/03/20/technology/tokenmaxxing-ai-agents.html). Jensen Huang [wants $500k employees to be spending at least $250k in tokens per year](https://youtu.be/gwW8GKwHB3I?t=1468). [Disney](https://www.businessinsider.com/how-disney-tech-employees-are-using-ai-claude-cursor-tokens-2026-4) and [JPMorgan](https://www.businessinsider.com/jpmorgan-track-software-engineers-ai-use-dashboards-2026-4) have AI usage dashboards.
+
+But what do you do once your company has fully adopted? Now you need to focus on whittling down the use cases to those that enhance productivity and quality. And this is where we're starting to see tokenmaxxing fatigue, with companies [accidentally blowing $500MM](https://www.fastcompany.com/91550884/claude-ai-costs-climb-company-spent-half-a-billion-dollars-in-a-single-month-report), Salesforce attempting to create a [metric to measure tokens and compute in relation to work completed](https://www.axios.com/2026/04/15/tokenmaxxing-ai-roi-metrics), and Uber's CTO realizing [they already blew their 2026 budget four months into the year](https://www.theinformation.com/newsletters/applied-ai/uber-cto-shows-claude-code-can-blow-ai-budgets), while Uber's COO is [not seeing productivity gains](https://www.businessinsider.com/tokenmaxxing-debate-uber-exec-viral-ai-costs-2026-5).
+
+We've known [measuring developer productivity by lines of code is a poor measurement since the 80s](https://www.folklore.org/Negative_2000_Lines_Of_Code.html) and more broadly, managing via output metrics is a surefire way to mistake the output for the outcome. Some Amazon employees put together an informal leaderboard to track token usage and what do you know: [the company asked them to deprecate it](https://www.businessinsider.com/amazon-ai-leaderboard-tokenmaxxing-2026-5).
+
+The strange part of the last wave of AI buzz is that companies seem to have temporarily forgotten about comparative advantages. "Anyone can write software now! Software companies are dead," misses the point that the software was just one part of the actual product these companies sell: there is ongoing support, feature development, integrations, and 100% focus on the software. Companies did the cost-benefit analyses and decided that paying a third-party to hire engineers and develop the software was cheaper than attempting to do it in-house.
+
+While AI has certainly brought the cost of building software internally down, there is evidence that [models provide better answers](https://arxiv.org/abs/2407.14333v2) to [more sophisticated users](https://x.com/TheStalwart/status/2057078123611464181). Specialists will continue to extract more value per token from these models, and if the studies hold true, they should diverge even more because these specialists will know which questions can be routed towards cheaper models versus frontier models, improving cost efficiencies.
+
+So ultimately, the tokenmaxxing craze will likely be a passing fad. Companies needed to quickly adopt AI and rather than cautiously analyze the highest value use cases, they pushed their teams to *just. get. on. board. now.* As inference costs add up, the pendulum is likely to swing back towards the cost-benefit analyses: is the token spend justified by the outcomes?
+
+---
+
+## Mine Print Hash
+
+The EU is in a tough position, with the Hormuz energy shock now weighing heavily upon their sovereign bond market. Matt Dines and I review the ECB's Financial Stability Review report, where they warn of an "abrupt repricing" in sovereign bonds. We also go through Japan's decision to integrate dollar stablecoins, which has massive implications for Eurodollar and SWIFT monetary rails.
+
+<YouTube
+  title="The Yen Battlefront: Europe's Weak Hand and the Dollar Stablecoin Hierarchy"
+  videoURL="https://www.youtube.com/watch?v=8gZXTUm53zE"
+/>
+
+---
+
+## Open Threads
+
+What will crop yields look like next year?
+- Farmers substituting for cheaper fertilizers. [Link](https://www.seattletimes.com/business/iran-war-forces-farmers-seek-fertilizer-alternatives-from-cow-dung-to-compost/)
+- Will El NiƱo weather cause crop failures? [Link](https://www.reuters.com/sustainability/cop/strong-el-nino-may-be-imminent-climate-change-will-make-its-effects-worse-2026-06-02/)
+
+Across the globe, countries are tightening their hold over critical industries and resources as a way to maintain sovereignty:
+- China expanding foreign investment controls. [Link](https://www.reuters.com/world/asia-pacific/china-toughens-rules-outbound-investment-after-meta-manus-contention-2026-06-01/)
+- EU discussing China foreign investment controls. [Link](https://www.theguardian.com/world/2026/may/28/eu-discuss-restrictions-chinese-imports-fears-overreliance)
+- Japan bolstering foreign investment controls. [Link](https://english.kyodonews.net/articles/-/76942)
+- Indonesia implementing export controls. [Link](https://www.bloomberg.com/news/articles/2026-05-31/indonesia-advances-export-control-plan-despite-uncertainty)
+- EU is possibly getting access to Anthropic's Mythos but importantly, at the discretion of the US government. [Link](https://www.ft.com/content/f88d62e3-5b67-4aed-ad69-6f38b62559b3)
\ No newline at end of file
diff --git a/content/articles/tokenmaxxing-as-management-theory/tokenmaxxing-as-management-theory.png b/content/articles/tokenmaxxing-as-management-theory/tokenmaxxing-as-management-theory.png
new file mode 100644 (file)
index 0000000..8bf11fc
Binary files /dev/null and b/content/articles/tokenmaxxing-as-management-theory/tokenmaxxing-as-management-theory.png differ