--- /dev/null
+# Normalize line endings: LF in the working tree everywhere.
+# Prettier enforces endOfLine: "lf", so git must not inject CRLF on checkout
+# (core.autocrlf=true would otherwise re-break formatting on stash, merge, etc.)
+* text=auto eol=lf
+
+# Binary assets
+*.png binary
+*.jpg binary
+*.jpeg binary
+*.gif binary
+*.webp binary
+*.avif binary
+*.ico binary
+*.pdf binary
+*.woff binary
+*.woff2 binary
- uses: oven-sh/setup-bun@v2
- name: Install dependencies
run: bun install --frozen-lockfile
+ - name: Check formatting
+ run: bun run format
- name: Build site pages
run: bun run build
- name: Build search index
},
}),
},
- integrations: [
- mdx(),
- sitemap(),
- ],
+ integrations: [mdx(), sitemap()],
image: {
layout: 'constrained',
service: {
import YouTube from '@components/youtube.astro';
import suzerainty from './american-ai-suzerainty.png';
-<Picture
- src={suzerainty}
- alt="Ceremonial scene under the American flag"
-/>
+<Picture src={suzerainty} alt="Ceremonial scene under the American flag" />
The Anthropic export control saga was not only a preview of AI safety policy, but also a peek at what a global reordering of political economy around AI intelligence could become. Access to frontier model intelligence, and possibly [Powerful AI as Dario calls it](https://darioamodei.com/essay/machines-of-loving-grace), exhibits many of the same properties that other arms-/export-controlled technologies do: a thin line separating civilian and military use. Think rockets, cryptography, or satellite imaging.
## Open Threads
Can smarter model routing substitute for raw model intelligence? My guess would be no, but it could marginally improve performance?
+
- Japan's Sakana AI claims it's Fugu model can do this. [Link](https://sakana.ai/fugu-release/)
- OpenRouter claims it's Fusion tool can do this. [Link](https://openrouter.ai/blog/announcements/fusion-beats-frontier/)
Europe trying to sort itself out.
+
- UK chews through it's 6th PM in 10 years as Starmer resigns. [Link](https://www.bbc.com/news/articles/cdxd00lg599o)
- Hoping to update their monetary rails to be competitive with the US. [Link](https://www.euronews.com/business/2026/06/23/european-parliament-backs-long-awaited-digital-euro-to-reduce-us-dominance-in-payments)
- Trying to get satellite launch capabilities going. [Link](https://www.ft.com/content/d2489175-c9df-435c-87f0-d4bb9915bb6e)
[^1]: I don't think is nominally any different than before, but the second derivative may have changed.
-[^2]: Okay, this isn't the only reason.
\ No newline at end of file
+
+[^2]: Okay, this isn't the only reason.
import emptied from './bitcoin-wallets-emptied.png';
import losses from './cumulative-losses-by-evidence-tier.png';
-<Picture
- src={emptied}
- alt="Bitcoin Wallets Emptied"
-/>
+<Picture src={emptied} alt="Bitcoin Wallets Emptied" />
In the Bitcoin world, it would be incorrect to think a Bitcoin wallet actually holds any Bitcoin. Despite the similar names, Bitcoin wallets carry out very different functions from the common cash wallet we carry around in our pockets. Bitcoin is technically "stored" on the public blockchain and a user can "unlock" that Bitcoin to send it somewhere using a private key. That private key is what is actually being stored in a Bitcoin wallet.
Across Bitcoin's history, there have generally been two types of wallets: hot and cold. Hot wallets are accessible directly from a computer: often more convenient for users to spend from or to integrate with other services, but with the potential downside of making it easier to be stolen through malicious means.
-Cold wallets, in comparison, are air-gapped and often purport to have additional physical security improvements like protection against certain side-channel attacks and duress codes. What is often not discussed very often is *how* that private key is generated, because whether the private key is stored on a cold or hot wallet, if the key is easily guessable, it doesn't matter how it is protected.
+Cold wallets, in comparison, are air-gapped and often purport to have additional physical security improvements like protection against certain side-channel attacks and duress codes. What is often not discussed very often is _how_ that private key is generated, because whether the private key is stored on a cold or hot wallet, if the key is easily guessable, it doesn't matter how it is protected.
For those who want to learn the actual mechanics of how elliptic curve cryptography works, I invite you to start at its [Wikipedia page](https://en.wikipedia.org/wiki/Elliptic-curve_cryptography). The high-level (aka incorrect) version is that there is a curve, upon which a **random** point is selected, a tangent line is drawn until it intersects with the curve, it is flipped along the axis, a new line is drawn between the points, it is flipped along the axis again, and so on.
-This makes it very easy to conduct the correct calculations if you know what the original point was. If you don't know what the point was, it is *extremely* computationally expensive to ever correctly guess the original point.
+This makes it very easy to conduct the correct calculations if you know what the original point was. If you don't know what the point was, it is _extremely_ computationally expensive to ever correctly guess the original point.
-So applying this back to Bitcoin private keys, *done correctly*, it would be effectively impossible for someone to guess the correct private key for a given Bitcoin, unless they had it to begin with. Of course, I bolded the word "random" in the paragraph above, because all of the fancy math depends upon truly random inputs. If a Bitcoin private key is generated with less than full randomness, it could be possible for an attacker to exploit this weakness to attempt to guess the private key.
+So applying this back to Bitcoin private keys, _done correctly_, it would be effectively impossible for someone to guess the correct private key for a given Bitcoin, unless they had it to begin with. Of course, I bolded the word "random" in the paragraph above, because all of the fancy math depends upon truly random inputs. If a Bitcoin private key is generated with less than full randomness, it could be possible for an attacker to exploit this weakness to attempt to guess the private key.
This is exactly what happened over the weekend to users of Coinkite's Coldcard hardware wallets. From [Block](https://engineering.block.xyz/blog/predictable-rng-fallback-and-32-bit-reseed-in-coldcard-firmware):
> An attacker who can determine or sufficiently constrain the device UID, timer state and RNG-call history can reproduce the fallback stream offline.
->
+>
> A wallet xpub, address or generated public key provides a candidate-validation oracle. Successful recovery of a wallet seed or private key permits theft of all associated funds.
>
> For Mk2/Mk3 v4, there is no cryptographically generated secret input to enumerate. For current devices, once the fallback state and call history are fixed, the remaining secure-element-derived search space is at most `2^32`, averaging approximately `2^31` candidate trials.
->
+>
> This does not mean every remote attacker can immediately recover every seed. Practical cost depends on available UID information, boot timing, prior RNG calls and derivation cost. No end-to-end brute-force benchmark is claimed here.
[Galaxy Research's latest update](https://x.com/glxyresearch/status/2084411915745595527) (as of this article) have identified a confirmed 1,367 BTC stolen, with up to 2,055 BTC potentially attributed to the same exploit.
-<Picture
- src={losses}
- alt="Cumulative Losses by Evidence Tier"
-/>
+<Picture src={losses} alt="Cumulative Losses by Evidence Tier" />
A truly terrible result for Coldcard users who trusted the entropy generated by their device to provide them with a private key that would protect their Bitcoin. The bug and its exploitation has already led to introspection across the ecosystem, with the [formation of ad hoc red teams](https://x.com/Rob1Ham/status/2084523368783438198) analyzing codebases for vulnerabilities, and [commentary on whether Anthropic's/OpenAI's overzealous safeguards place defenders at a disadvantage](https://x.com/TFTC21/status/2084392236050251776).
## Open Threads
US interventions:
+
- US Treasury intervenes in Yen market. [Link](https://www.ft.com/content/0f9b2fe7-bde4-4f5f-b49e-93ccb5da9ea8)
- US in talks with Oman to open routes through Hormuz. [Link](https://www.cnn.com/2026/08/02/world/live-news/iran-war-trump?post-id=cmscx6wlq00003b6wmdch0ua4)
Cybersecurity can't keep up with the attacks:
+
- Shai-Hulud is back. [Link](https://www.aikido.dev/blog/keyv-and-friends-compromised-in-npm-supply-chain-attack)
-- OpenAI's model breaking out of its sandbox is likely just the beginning. [Link](https://foreignpolicy.com/2026/07/30/openai-hack-genie-bottle-defense/)
\ No newline at end of file
+- OpenAI's model breaking out of its sandbox is likely just the beginning. [Link](https://foreignpolicy.com/2026/07/30/openai-hack-genie-bottle-defense/)
import whichway from './bitcoins-opt-in-constitution.png';
import situation from './situation-monitor.png';
-<Picture
- src={whichway}
- alt="Choosing the path ahead"
-/>
+<Picture src={whichway} alt="Choosing the path ahead" />
While protocols are generally all agreed upon and adhered to, it is extremely common to have competing pieces of software adhering to said protocol with slightly different understandings of how specifics are implemented.
So recall back to the HTTP RFCs I mentioned before: while the discussion, reviews, documentation, and testing are all useful in creating consensus, it is still up to the software to implement these features and for enough users to agree to switch. There are already [forked versions of the Bitcoin Core node software](https://github.com/bitcoinknots/bitcoin) which implement BIP-110, but so far they are not gaining much interest.
-<Picture
- src={situation}
- alt="BIP110 Situation Monitor"
-/>
+<Picture src={situation} alt="BIP110 Situation Monitor" />
My guess is that BIP-110 will likely not become a part of the "standard" Bitcoin protocol unless a large number of users decide they really care about limiting data sizes. And specifically, they [need a large enough of percentage of economic nodes to signal support for the BIP](https://bitcoinmagazine.com/news/foundry-asks-bitcoin-miners-vote-bip-110).
## Open Threads
Open-weights and security:
+
- OpenAI's pre-release model hacked Hugging Face to cheat on an evaluation. [Link](https://openai.com/index/hugging-face-model-evaluation-security-incident/)
- Everyone except OpenAI and Anthropic sign open letter in favor of open-source models. [Link](https://www.microsoft.com/en-us/corporate-responsibility/topics/open-weight/)
- Refuting claims that distilling American models is how Kimi K3 took a leap. [Link](https://techcrunch.com/2026/07/23/experts-say-exploiting-anthropics-fable-isnt-how-kimi-k3-got-so-good/)
- Judicial examples that even if they did, distilling may be considered legal. [Link](https://searchengineland.com/google-loses-key-dmca-claims-against-serpapi-in-scraping-lawsuit-483185)
AI valuations:
+
- "Unnamed state-backed manufacturer" starts delivering DUV machines to key Chinese chipmakers. [Link](https://www.tomshardware.com/tech-industry/semiconductors/china-begins-mass-production-of-domestic-immersion-duv-lithography-machines)
- Oracle needs ~$7B for energy infrastructure collateral in Wisconsin. [Link](https://finance.yahoo.com/technology/ai/articles/oracle-faces-potential-7-billion-104844370.html)
- - DOW awards $7B to Oracle. [Link](https://www.war.gov/News/Releases/Release/Article/4554042/dow-awards-a-nearly-7-billion-oracle-agreement-to-accelerate-the-arsenal-of-fre/)
+ - DOW awards $7B to Oracle. [Link](https://www.war.gov/News/Releases/Release/Article/4554042/dow-awards-a-nearly-7-billion-oracle-agreement-to-accelerate-the-arsenal-of-fre/)
- Nvidia to guarantee $250B in financing for OpenAI data center. [Link](https://www.wsj.com/tech/ai/nvidia-in-talks-with-openai-to-guarantee-250-billion-financing-for-data-center-3dd6eae3)
-[^1]: They most definitely didn't, but it's an easy timestamp to point to.
\ No newline at end of file
+[^1]: They most definitely didn't, but it's an easy timestamp to point to.
import YouTube from '@components/youtube.astro';
import denarius from './blocking-your-ai-access-for-your-safety.png';
-<Picture
- src={denarius}
- alt="Silver Denarius Depiction"
-/>
+<Picture src={denarius} alt="Silver Denarius Depiction" />
-Anthropic released Fable 5, a version of Mythos with safety gates, intended to stop bad actors from using the model's full capabilities. The full details are not public, but reporting seems to suggest that Andy Jassy, Amazon's CEO, [contacted the US government to report Fable 5 was easily jailbroken](https://www.wsj.com/tech/ai/amazon-ceos-talks-with-u-s-officials-triggered-crackdown-on-anthropic-models-dcc90578) and led to the US slapping export controls onto Fable 5 and Mythos. This is an interesting chain of events because Anthropic is [both an Amazon investment and an AWS customer](https://www.aboutamazon.com/news/company-news/amazon-invests-additional-5-billion-anthropic-ai).
+Anthropic released Fable 5, a version of Mythos with safety gates, intended to stop bad actors from using the model's full capabilities. The full details are not public, but reporting seems to suggest that Andy Jassy, Amazon's CEO, [contacted the US government to report Fable 5 was easily jailbroken](https://www.wsj.com/tech/ai/amazon-ceos-talks-with-u-s-officials-triggered-crackdown-on-anthropic-models-dcc90578) and led to the US slapping export controls onto Fable 5 and Mythos. This is an interesting chain of events because Anthropic is [both an Amazon investment and an AWS customer](https://www.aboutamazon.com/news/company-news/amazon-invests-additional-5-billion-anthropic-ai).
A bit of schadenfreude as Anthropic has repeatedly said Mythos is dangerous, ["[t]he fallout—for economies, public safety, and national security—could be severe,"](https://www.anthropic.com/glasswing#:~:text=The%20fallout%E2%80%94for%20economies%2C%20public%20safety%2C%20and%20national%20security%E2%80%94could%20be%20severe%2E) and called for a [pause to AI development](https://www.anthropic.com/institute/recursive-self-improvement#:~:text=We%20believe,technology%2E):
> Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety.
General opinions seem to range between the following arguments:
+
1. Dario and Anthropic are genuinely worried about the pace of AI advancement, believe institutions are moving too slowly, but intend to advance AI anyways until everyone promises they'll stop.
2. The current administration has a political agenda against Anthropic and is slapping export controls on them in retaliation. This is the same line of thought from when Anthropic was [designated a supply chain risk](https://www.politico.com/news/2026/06/04/hegseth-anthropic-designation-supply-chain-risk-00951183).
3. Dario and Anthropic (and maybe Amazon?) are putting together a marketing campaign that paints Mythos as so capable, it's dangerous — a la Apple's Power Mac G4 ad:
To see if you're allowed to read the next article, scan your ID. To keep you safe, so you don't read something you shouldn't have, because ultimately this is about safety from wrongthink. The fight spilling into public view is over who controls the thought.
---
+
## Mine Print Hash
Tokenized assets have been a hot topic, alongside stablecoins, which plays right into the Mine Print Hash thesis. Matt Dines and I discuss how tokenized assets are competing with traditional financial rails via GENIUS Act stablecoins on last week's podcast.
## Open Threads
A little bit of everything this week:
+
- China's cross-border digital payments system. [Link](https://www.ft.com/content/76de5ca6-9ae8-49ae-a0cb-64d09040f327)
- Nvidia taps the bond market. [Link](https://www.ft.com/content/c0e7cc58-e309-4f47-b466-d577c06cbeec)
- Another US-Iran deal? [Link](https://www.cnbc.com/2026/06/15/us-iran-deal-hormuz-markets.html)
-[^1]: Except for Bluesky, because apparently that's not social media?
\ No newline at end of file
+[^1]: Except for Bluesky, because apparently that's not social media?
import bilbo from './circles-float-problem.png';
import meme from './after-all-why-not-why-shouldnt-i-keep-it.jpg';
-<Picture
- src={bilbo}
- alt="Contemplating the golden ring"
-/>
+<Picture src={bilbo} alt="Contemplating the golden ring" />
Stablecoins are a pretty easy concept to wrap your head around: you give the stablecoin issuer a dollar, they invest that dollar into some specified mix of "safe" assets, and in return they give you a digital token. That digital token has some magical properties that I won't get into today, but they theoretically enable use cases that you aren't able to achieve on traditional monetary rails.
Circle follows the same pattern, issuing its USDC dollar stablecoin, managing its reserve in cash and US Treasuries, and earning profits via the yields on its reserve. Circle [partnered with Coinbase](https://www.circle.com/blog/ushering-in-the-next-chapter-for-usdc) as a key distribution partner, reportedly [entering a revenue sharing agreement](https://www.sec.gov/Archives/edgar/data/1876042/000187604226000062/crcl-20251231.htm#:~:text=We%20incur%20costs%20to%20incentivize%20distributors%20to%20use%20and%20distribute%20Circle%20Digital%20Assets). This setup is issuer-led, making the stablecoin issuer the main profit driver, and everyone else hoping to get a small cut of whatever the issuer is willing to provide.
-<Picture
- src={meme}
- alt="After All, Why Not? Why Shouldn't I Keep It? meme"
-/>
+<Picture src={meme} alt="After All, Why Not? Why Shouldn't I Keep It? meme" />
An obvious challenge to this model would be the creation of a consortium of partners that jointly issue a stablecoin, sharing the profits amongst themselves. Enter [Open Standard](https://joinopenstandard.com/):
>
> The stablecoin space has come a long way since we started Bridge nearly 5 years ago. Total volume is now approaching that of the ACH network and nearly every major financial institution from banks to fintechs is leaning in.
>
-> But cost and governance have limited utility and put core payments use cases out of reach. Businesses need a stablecoin that's open, low-cost, high-throughput, broadly accessible, and aligned to their interests.
+> But cost and governance have limited utility and put core payments use cases out of reach. Businesses need a stablecoin that's open, low-cost, high-throughput, broadly accessible, and aligned to their interests.
>
-> We'll be launching later this year.
+> We'll be launching later this year.
>
> Visit joinopenstandard[.]com to learn more and participate.
> — [@zcabrams](https://x.com/zcabrams/status/2071966580142125175)
## Open Threads
Private AI usage:
+
- The [National Design Studio](https://ndstudio.gov/)'s first model, Rampart, is an on-device PII-redactor. [Link](https://ndstudio.gov/posts/say-hello-to-rampart)
- Codex and Claude Desktop competitor that routes its model calls through [Venice](https://venice.ai/), a privacy-focused inference provider. [Link](https://github.com/open-software-network/os-june)
- Attempting to distribute model inference compute in a sort of [Folding@Home](https://en.wikipedia.org/wiki/Folding%40home) way. [Link](https://meshllm.cloud/)
Legislative branch pushes back on a few Executive branch moves (just a few):
+
- Lisa Cook retains Fed Governor job, for now. Notably, the Supreme Court really is saying the Executive needs to follow the right procedures. [Link](https://apnews.com/article/supreme-court-executive-power-trump-firing-cook-7b7676e5a066f8df41077a0920b9f334)
- Birthright citizenship executive order is struck down. Argument over the definition of "citizen," the 14th amendment author's intention, and England's version of the law from which this was borrowed, abounds. [Link](https://www.scotusblog.com/2026/06/supreme-court-strikes-down-trumps-order-ending-birthright-citizenship/)
---
-title: "Formal Verification: When Intention Is Not Enough"
+title: 'Formal Verification: When Intention Is Not Enough'
type: newsletter
date: 2026-08-11
modified: 2026-08-11
import YouTube from '@components/youtube.astro';
import gem from './formal-verification-when-intention-is-not-enough.png';
-<Picture
- src={gem}
- alt="Formal Verification in AI"
-/>
+<Picture src={gem} alt="Formal Verification in AI" />
I'll start today's article off with a definition of formal verification as provided by [Keagan McLelland on the Bitcoin Development Mailing List](https://groups.google.com/g/bitcoindev/c/OIml9stwbGQ/m/EWZUz9-7AQAJ):
> Formal Verification is a practice in software development that gives greater assurances than typical testing procedures or forum arguments on how various schemes, protocols, and implementations behave. When we argue about how Bitcoin is designed or how it ought to be designed, engineers will make statements of fact about how the system works, but depending on how much effort is put into the presentation of that argument, it can be difficult to tell if a statement is true, or if it is, why it is true. By formally specifying these statements of fact on how the Bitcoin protocol works we can be precise about what we are saying, what properties we have, what properties we want, and whether the system actually delivers those properties or not.
->
+>
> Normal programming languages do this to an extent, and indeed one of the best ways to settle a dispute is to look at how the code is implemented. However, the nature of C++ can make it difficult to get certainty over certain claims about the behavior of the protocol due to its type system being comparatively weak when judged against the tools typically tasked with this responsibility.
->
+>
> If you want to be able to make specific arbitrary logical statements about Bitcoin or any other software you typically need to specify it in a language with a Dependent Type Theory. The class of tools used for this is called an Interactive Theorem Prover (ITP).
Formal verification has existed for decades, but is not commonly used to verify software protocols for one simple reason: it's expensive. Designing a model that accurately represents a protocol, its rules, and allowed inputs requires specialized skill and a massive amount of work. For most projects until recently, the potential risks didn't justify the cost.
It is no longer enough to create tests and inspect a project's codebase. A motivated attacker with a large enough financial incentive will throw AI resources at relentlessly scanning for weaknesses to exploit. Look no further than recent attacks on the Lightning network:
> Statement
->
+>
> We regret to inform you that the day we never wanted to come has finally arrived: @lnp2pBot has come to an end.
->
+>
> As we announced a few days ago, @lnp2pBot has been under attack since day one. We have absorbed the losses we suffered for the simple reason that we wanted to continue providing our loyal users with the service they deserve. However, events have escalated, and although we have spent months working to improve the bot’s security, we simply do not have the resources required to fight an army of script kiddies armed with AI models.
->
+>
> For all these reasons, we have made the difficult and painful decision to indefinitely suspend all @lnp2pBot activity, effective immediately.
->
+>
> Once again, the losses are ours to bear. We will contact the few users who still have pending payments in order to resolve those individual cases. Not a single user will suffer any financial loss.
->
+>
> The bot was born as our legitimate form of defense against authoritarian governments that seek to force users to hand over all kinds of personal information under the excuse that it is for their own good. Many times, we thought they would be the ones who would eventually manage to shut the bot down. Ironically, it was not an authoritarian government that succeeded, but most likely some of our own users.
->
+>
> Thank you for everything!
> — [@negrunch](https://x.com/negrunch/status/2086899005355704703)
> **Update: Boltz will stay disabled until further notice.**
->
+>
> Our API remains available to process refunds cooperatively. In any case, unilateral refunds will work, as they do not depend on our infrastructure.
->
+>
> Our support team stays reachable.
->
+>
> To be clear: this is not a response to a single incident. Over the past months we have seen a steady rise in automated, AI-assisted probing of our infrastructure, and we have dealt with several exploits. Each was contained, but the pattern is clear: attackers now iterate faster than a team our size can find and patch. In the past few days alone we saw a drastic acceleration, and we do not believe this asymmetry will reverse. After reviewing the results of our own recent security scans, we cannot responsibly re-enable Boltz swaps, especially as we are being actively targeted by what appear to be multiple resourceful groups while we race to deploy fixes.
->
+>
> What we are seeing is a major paradigm shift for Bitcoin services operating on an open source stack, and it needs careful analysis. Do not expect swap services to resume shortly.
->
+>
> To be explicit: no user funds were ever at risk. Boltz is non-custodial by design. And as a fully bootstrapped company, the losses were ours alone.
->
+>
> We don't know yet how things will continue from here, but we'll keep you posted as soon as we have had the time to catch our breath and make a decision 🙏
> — [@Boltzhq](https://x.com/Boltzhq/status/2084311537502630319)
## Open Threads
Geopolitical maneuvers to test boundaries and create new trade routes:
+
- China bypasses shipping chokepoints via "Ice Silk Road". [Link](https://www.ft.com/content/4456f475-f2c8-4cb7-8312-0c47c5f781b8)
- Pentagon drafting new nuclear strike strategy. [Link](https://www.nbcnews.com/politics/national-security/pentagon-drafting-new-us-nuclear-strategy-case-regional-war-china-russ-rcna587370)
- Component in British naval drones sent signals back to China. [Link](https://www.wsj.com/world/uk/u-k-discovers-component-in-its-naval-drones-sent-signals-to-china-6ae60578)
- Syria recontrols its air fields, pushing Russia out of key bases for African missions. [Link](https://www.ft.com/content/df62a69e-aa53-4cde-8a02-4d2cd465a370)
Building monetary networks:
+
- Russia's A7 payment network. [Link](https://www.wsj.com/world/russia/russias-hottest-startup-is-a-state-backed-sanctions-evasion-network-7afc488c)
- UK's tokenized gold. [Link](https://www.ft.com/content/9973f186-5834-4b66-8faa-e8b2e8711466)
- Bank behind China's AI listings. [Link](https://www.ft.com/content/657249de-f5bd-4648-9a1c-77315cd5a1c9)
- Massive Nvidia financing. [Link](https://www.ft.com/content/98a8fd17-15b6-4f67-9cb4-825722b11348)
-- CLARITY stalls. [Link](https://www.ft.com/content/352a15ac-53c4-46eb-b75e-957f0388dc79)
\ No newline at end of file
+- CLARITY stalls. [Link](https://www.ft.com/content/352a15ac-53c4-46eb-b75e-957f0388dc79)
import YouTube from '@components/youtube.astro';
import ouroboros from './from-ai-tools-to-ai-loops.png';
-<Picture
- src={ouroboros}
- alt="Robot Ouroboros Artwork"
-/>
+<Picture src={ouroboros} alt="Robot Ouroboros Artwork" />
I wrote [last week](/articles/2026-06-02-tokenmaxxing-as-management-theory/) about tokenmaxxing as the awkward adoption phase of AI inside companies: first you get everyone using the tool, then you figure out whether any of that usage is actually enhancing output.
-Looping is the next phase. The question is no longer, "Can an employee use AI at each step of the workflow?" The question becomes, "Can the *entire workflow* turn into an automated loop?"
+Looping is the next phase. The question is no longer, "Can an employee use AI at each step of the workflow?" The question becomes, "Can the _entire workflow_ turn into an automated loop?"
A (naive) workflow in one of today's software engineering companies might look something like:
Software engineers are already using AI at each of these steps through a mix of autocompletions, chat interfaces, and coding agents. You may notice there are some natural places where automation already speeds this up: continuous integration/continuous deployment (CI/CD) already handles parts of the build-test-deploy process, and observability tools already notify teams of issues or roll back failed releases. But these are mostly point solutions.
-What if you could create an automated loop out of the *entire* software engineering workflow? Boris Cherny (Anthropic) [certainly thinks so](https://youtu.be/SlGRN8jh2RI), as does Peter Steinberger (OpenAI):
+What if you could create an automated loop out of the _entire_ software engineering workflow? Boris Cherny (Anthropic) [certainly thinks so](https://youtu.be/SlGRN8jh2RI), as does Peter Steinberger (OpenAI):
> Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.
->
+>
> You should be designing loops that prompt your agents.
->
+>
> — Peter Steinberger (@steipete) [June 7, 2026](https://x.com/steipete/status/2063697162748260627)
There are already several examples and functional versions of loops that can be implemented today. Claude Code and Codex have `/goal` functions that continue working on a task until a final stopping condition is reached. In other words: don't just answer one prompt; keep working until the condition is true. Andrej Karpathy's [autoresearch](https://github.com/karpathy/autoresearch) is the research version of the same idea: edit the code, train for five minutes, compare the result, keep or discard, repeat.
Governments continue controlling who gets compute, who gets chips, who can work where, and which firms become part of national industrial strategy.
- Bernie Sanders wants to seize the means of AI production. [Link](https://www.nytimes.com/2026/06/01/opinion/artificial-intelligence-bernie-sanders.html)
-- Trump wants to invest in the means of AI production. [Link](https://www.bloomberg.com/news/articles/2026-06-05/us-exploring-government-partnerships-with-ai-firms-trump-says)
+- Trump wants to invest in the means of AI production. [Link](https://www.bloomberg.com/news/articles/2026-06-05/us-exploring-government-partnerships-with-ai-firms-trump-says)
- Possibly an 18-month loophole China exploited to access export controlled chips? [Link](https://www.techtimes.com/articles/317905/20260606/bis-closed-china-ai-chip-loophole-trump-officials-dispute-whether-gap-ever-existed.htm)
- China won't let top AI talent travel. [Link](https://www.bloomberg.com/news/articles/2026-05-26/china-expands-travel-curbs-to-top-ai-talent-at-private-firms)
- Claude Mythos - Neutered Version (aka Fable 5) limits its capabilities opaquely to the user depending on the type of question asked. [Link](https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf)
- China starts a prefabricated power hub for data centers. [Link](https://www.bloomberg.com/news/articles/2026-06-07/china-starts-prefabricated-power-hub-for-data-centers-cctv-says)
- Google selects Intel as a TPU manufacturer [Link](https://qz.com/google-intel-tpu-order-tsmc-capacity-2028-060826)
-- SpaceX has solved the datacenters in space heat problem? Big if true. [Link](https://x.com/SpaceX/status/2064099405758906727)
\ No newline at end of file
+- SpaceX has solved the datacenters in space heat problem? Big if true. [Link](https://x.com/SpaceX/status/2064099405758906727)
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?
+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?
---
## 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 Niño 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
+- 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)
import YouTube from '@components/youtube.astro';
import cube from './why-ai-companies-are-hunting-for-data-nobody-else-has.png';
-<Picture
- src={cube}
- alt="Glowing Unique Data Cube"
-/>
+<Picture src={cube} alt="Glowing Unique Data Cube" />
The first question you would probably ask if I told you that [Google bought Spirit Airlines' data at auction](https://www.cnn.com/2026/08/18/business/google-spirit-airlines-data), including internal emails, documents, booking information, and other corporate data, is whether Google would use it to improve Google Flights (Google's flight booking tool) or start its own airliner.
## Open Threads
China participates more deeply in the global order:
+
- China intends to expand the use of the RMB for international trade. [Link](https://www.gov.cn/lianbo/202608/content_7077845.htm)
- Z.ai's GLM-5.3 will have a staggered rollout like Mythos and GPT-5.6. [Link](https://www.wired.com/story/zai-open-weight-ai-models-release-cybersecurity-hacking/)
- Apple urged by the US against using China's CXMT for chips. [Link](https://www.wsj.com/tech/apple-china-memory-chip-plan-57773a83)
US tries to build infrastructure:
+
- Washington state tells Kalshi to stop operating within the state. Related to a similar push by New York. Both are being fought by the CFTC. [Link](https://www.seattletimes.com/business/kalshi-ordered-to-stop-most-operations-in-wa/)
-- OpenAI 10GW data center in Ohio with SoftBank's SB Energy, partially backed by Nvidia. [Link](https://www.wsj.com/articles/openai-locks-in-lease-for-huge-data-center-in-ohio-with-backing-from-nvidia-7474bb9c)
\ No newline at end of file
+- OpenAI 10GW data center in Ohio with SoftBank's SB Energy, partially backed by Nvidia. [Link](https://www.wsj.com/articles/openai-locks-in-lease-for-huge-data-center-in-ohio-with-backing-from-nvidia-7474bb9c)
import bag from './wisconsin-wont-write-the-ai-put.png';
import delta from './database-of-emerging-large-load-tariffs.png';
-<Picture
- src={bag}
- alt="Money bag with clenched grip"
-/>
+<Picture src={bag} alt="Money bag with clenched grip" />
Oracle wants to build four datacenters in Wisconsin, estimated to cost $15 billion, as part of its broader compute infrastructure deal with OpenAI. But before Oracle commits to the development, they need to ensure the energy infrastructure needed for the massive data center will be built. Wisconsin considers these, "Bespoke Resources," aka "power plants we built for a customer."
So We Energies builds the energy infrastructure, Oracle promises to buy the electricity for a long enough period of time that they can repay them and We Energies recoups the cost of building the infrastructure. If there continues to be demand for this Oracle data center, then Oracle can pay the bills on time and everything is great. If demand or technology changes, Oracle abandons the project, or some other unforeseen circumstance happens, We Energies (and Wisconsin) is left holding the bag for this now unnecessary energy infrastructure.
To prevent this, the Wisconsin Public Service Commission requires large customers to lock themselves into a long-term deal, pay for 100% of the generation costs, and those below an A-minus credit rating must provide additional collateral. From [Financial Times](https://www.ft.com/content/b37030b6-bda8-4ba9-8e08-e6b88687b8f5):
+
> The Public Service Commission of Wisconsin, which is responsible for scrutinising and setting the prices the state’s utilities can charge, has declined to reconsider rules it imposed on utility We Energies, which would require Oracle to provide a $7bn letter of security at a yearly cost of more than $100mn.
Wisconsin isn't alone in these terms, with[ at least 5 other states approving or proposing tariffs attempting to bear less of the data center energy infrastructure risk](https://sepapower.org/large-load-tariffs-database/). As more states update their rules, we may expect a repricing both in expected data center development costs and in the perceived risk (higher lending rates) associated with these builds.
- Chinese firms engineering solutions around the disadvantage they are in, given EUV controls. [Link](https://newsletter.semianalysis.com/p/steel-smic-n3-teardown)
- Kimi K3 makes a significant performance jump towards the frontier. I discuss this in more detail our recent Mine Print Hash episode. [Link](https://apnews.com/article/kimi-k3-china-ai-0d8a5e268deb11a673f4d444fc597cc5)
- - US may take steps to ban or disincentivize the use of foreign models. [Link](https://www.semafor.com/article/07/15/2026/white-house-not-ruling-out-action-on-open-source-ai-models)
+ - US may take steps to ban or disincentivize the use of foreign models. [Link](https://www.semafor.com/article/07/15/2026/white-house-not-ruling-out-action-on-open-source-ai-models)
Market power competition with stablecoin issuers:
---
type: video
-title: "Hawkish Weakness: ECB Hikes, Dollar Flows, and Kalshi/Polymarket as Stablecoin Demand Markets"
+title: 'Hawkish Weakness: ECB Hikes, Dollar Flows, and Kalshi/Polymarket as Stablecoin Demand Markets'
date: 2026-06-04
modified: 2026-06-04
publication:
---
type: video
-title: "Tokenized Assets: The Market Plumbing Contest"
+title: 'Tokenized Assets: The Market Plumbing Contest'
date: 2026-06-11
modified: 2026-06-11
publication:
---
type: video
-title: "AI Export Controls and the Warsh Era: Fed Regime Change Amidst AI Sovereignty Challenges"
+title: 'AI Export Controls and the Warsh Era: Fed Regime Change Amidst AI Sovereignty Challenges'
date: 2026-06-18
modified: 2026-06-18
publication:
---
type: video
-title: "SpaceX, FHLB Lending, and the Dollar Liquidity Scramble"
+title: 'SpaceX, FHLB Lending, and the Dollar Liquidity Scramble'
date: 2026-06-25
modified: 2026-06-25
publication:
---
type: video
-title: "SCOTUS on the Fed, Open USD, and Brady Bonds: Precursors of Dollar Stablecoin Headlines to Come"
+title: 'SCOTUS on the Fed, Open USD, and Brady Bonds: Precursors of Dollar Stablecoin Headlines to Come'
date: 2026-07-02
modified: 2026-07-02
publication:
import YouTube from '@components/youtube.astro';
Precursors of future dollar stablecoin headlines have hit the tape:
+
- SCOTUS rules on Fed independence, but Matt Dines sees breadcrumbs for the next legal fight
- I outline Open USD's challenge to issuer-led stablecoins
- Mexico's Brady Bond debt comes to roost
---
type: video
-title: "GENIUS Act Deadlines, Kimi K3, and the China Competition"
+title: 'GENIUS Act Deadlines, Kimi K3, and the China Competition'
date: 2026-07-17
modified: 2026-07-17
publication:
---
type: video
-title: "Houthis, the Red Sea, and Sovereign Debt Repricing"
+title: 'Houthis, the Red Sea, and Sovereign Debt Repricing'
date: 2026-07-23
modified: 2026-07-23
publication:
---
type: video
-title: "Reading the Fed Vote Tea Leaves, China Creates a SOFR Bridge, and America’s Defense Cycle"
+title: 'Reading the Fed Vote Tea Leaves, China Creates a SOFR Bridge, and America’s Defense Cycle'
date: 2026-07-30
modified: 2026-07-30
publication:
---
type: video
-title: "Yentervention: Japan as a Critical Nexus in US Monetary and Resource Policy"
+title: 'Yentervention: Japan as a Critical Nexus in US Monetary and Resource Policy'
date: 2026-08-07
modified: 2026-08-07
publication:
---
type: video
-title: "Asheville: Bretton Woods for the Stablecoin Dollar Era"
+title: 'Asheville: Bretton Woods for the Stablecoin Dollar Era'
date: 2026-08-13
modified: 2026-08-13
publication:
---
type: video
-title: "Canada, Brookfield, and the Geopolitics Surfacing in Markets"
+title: 'Canada, Brookfield, and the Geopolitics Surfacing in Markets'
date: 2026-08-20
modified: 2026-08-20
publication:
- really toed the line between too much and okay
- if you don't have a connection to the original material this is probably a tough watch, but felt they did a solid job updating for a modern audience while keeping the ridiculousness
-<Picture src={Skeletor} alt="Skeletor running away meme. Until we meet again." />
+<Picture
+ src={Skeletor}
+ alt="Skeletor running away meme. Until we meet again."
+/>
"format": "bun x --bun prettier --check .",
"format:fix": "bun x --bun prettier --write .",
"typecheck": "bun x --bun astro check",
- "test": "bun test"
+ "test": "bun test",
+ "prepare": "simple-git-hooks"
},
"simple-git-hooks": {
"pre-commit": "bun x --bun lint-staged"
},
"lint-staged": {
- "*.{ts,tsx,astro,js}": [
+ "*.{ts,tsx,astro,js,mjs,cjs}": [
"eslint --fix",
"prettier --write"
],
projects / cotsuka.github.io / commitdiff