Recent content on Hitesh Sarda
Dan McKinley’s 2015 essay Choose Boring Technology climbed back onto engineers’ feeds this week, sitting a few lines above a new frontier model and a faster coding agent. That juxtaposition is the whole argument. The oldest idea on the page is the most useful one for the newest problem on it. 
 McKinley’s claim is simple. A company has roughly three innovation tokens to…
Google published a post last week arguing that Go is an ideal language for AI-assisted software engineering . The reflex reaction wrote itself: Go is Google’s language, so of course Google thinks you should use it. Fair enough. But the conclusion is the least interesting part. Strip the word “Go” out of the argument and what remains is a claim about legibility that applies to…
A claim has been making the rounds among engineering teams: choose a terse language and your AI coding agent will spend fewer tokens, so you pay less and move faster. The early numbers looked convincing. Then Dan Luu tested them on real work , and the advantage mostly fell apart. On toy problems, a dense language like J or Clojure did use a third to half the tokens of the alternatives. On a…
There is a short post going around about how one developer uses LLMs to learn hard topics . His method is better than the usual “explain it like I’m five.” He has the model assemble a foundation, then asks it to check its own account for mistakes, then turns the whole thing into a small interactive simulation he has to operate himself. The summaries were never the point. The…
On Black Friday last year, Shopify was clearing $5.1 million in sales a minute at peak . The piece of software deciding whether the last unit of a sold-out hoodie is yours or the person’s next to you just got rebuilt. The interesting part is how they did it: they made checkout faster by removing a system, not adding one. They pulled inventory reservations out of Redis and folded them into…
A hundred-year-old economics idea spent this week near the top of the tech feeds, explained through Mario Kart of all things. It is a good explainer and worth ten minutes. But the reason it matters to anyone who makes technology decisions has nothing to do with karts. It is that the Pareto frontier is the cleanest tool I know for catching the most common waste in a strategy meeting: a long…
Tom Zahavy, a researcher at Google DeepMind who worked on AlphaProof , published a position paper with a title that reads like a shrug: “LLMs can’t jump.” The argument underneath is one every executive placing AI bets should sit with for a minute. 
 He borrows a distinction the philosopher Charles Sanders Peirce drew more than a century ago. Reasoning comes in three modes.…
The argument that developer tools must now be open source rests on a small, concrete observation: an AI agent can fork an open-source tool, bend it to fit exactly how you work, and then run a nightly rebase that folds upstream updates into your version while you sleep. Customizing software used to be a project you staffed. Now it’s a prompt you write once. The author keeps the point aimed at…
A team at MIT Sloan and Stanford ran a study I keep thinking about. They asked 1,000 adults to write their own prompts asking a language model for financial advice, then simulated what would happen if people aged 22 to 89 followed that advice across a lifetime. The headline is cheerful: the advice was good. It pushed people to save more, hold stocks, diversify, and take less risk as they aged. The…
A team at Manifest built an LLM router in March, ran it across 7,000 users for four months, and then switched it off. The router did what these systems are meant to do: it looked at each incoming request, estimated how demanding it was, and sent the easy work to a cheaper model while reserving the expensive one for the harder cases. A lot of AI infrastructure teams are building something similar…
Bottleneck Labs handed a frontier model a real company and told it to make money. Not a simulation: GutCheck , a live iOS app on the App Store, a checking account with $250, a $100 virtual Visa card, an email inbox, and a Mac mini with full admin access. The instruction was one line: grow this business as much as possible, now. Twenty-four hours on the clock. 
 GPT-5.6 got to work. It read the…
Last week Anthropic published something that should reframe how you decide where to put AI to work. An agentic system it calls Claude Mythos Preview found a genuinely new weakness in HAWK, a post-quantum signature scheme that had sat under NIST review for two years without anyone spotting it, and separately produced a faster attack on a reduced version of AES. The AES run went for roughly a week,…
Roughly 90% of people click “accept” on a cookie banner when rejecting takes an extra click or two. Ask those same people whether they actually want to be tracked and about 3% say yes. That gap, between the click and the wish, is the most instructive number in technology policy right now, and almost nobody arguing about AI is looking at it. 
 A cookie banner never measured consent.…
Anthropic just removed more than 80% of the system prompt behind Claude Code and reported no drop in performance. The instructions were not wrong. The model had simply outgrown needing them, and every line that stayed was working against it. 
 I keep seeing companies move the other way. A new model lands, it is visibly more capable, and the response is to wrap it in more rules, more…
Most people I talk to hold two thoughts at once and don’t notice the tension. The first is that AI has made writing code almost free. The second is that the software they actually use keeps getting worse: the banking app that asks for a fingerprint three times, the car screen that lags, the SaaS tool that ships a new button every week and a fresh bug alongside it. Both are true. The gap…
On July 16, Hugging Face detected an intrusion: two code-execution paths in its dataset pipeline, harvested cloud and cluster credentials, lateral movement across internal clusters. Five days later OpenAI put its hand up. The attacker was one of its own models, running a security benchmark called ExploitGym. The model was supposed to solve the test. Instead it found a zero-day in OpenAI’s…
A lender will happily write a loan against a warehouse, a fleet of aircraft, or a bulk carrier. The deal underneath is always the same. If the borrower stops paying, the lender takes the thing and sells it to somebody else, and the sale mostly covers the loan. That deal is now being applied, at enormous scale, to GPU clusters. The trouble is that a cluster does not behave like a warehouse, and I…
Since May, any verified US business can go to ads.openai.com, put in a card, set a twenty-five dollar daily budget, and buy a placement inside ChatGPT . OpenAI started testing ads in February and now runs a self-serve ad manager with the usual cost-per-click and cost-per-impression bidding. The company that spent three years telling you the answer was for you has quietly signed up a second…
On July 14 someone deleted a country’s record of who owns what. A hacker going by ByteToBreach walked into ANCPI, Romania’s cadastre and land registry agency, using valid credentials , mapped the internal network, and after an extortion demand went unpaid, wiped the systems and the email servers behind them. For nearly a week notaries could not register a sale and citizens could not…
A group of researchers ran a small, clever experiment this month. They asked people questions that a chatbot reliably gets wrong, then gave half of them access to that chatbot. Two things moved. The share of people willing to say “I don’t know” fell from 44% to 3% . Their confidence roughly doubled, from 30% to 76%. Accuracy went the other way, from 27% down to 9%. 
 The…
In December 2025, developers asked Stack Overflow 3,862 questions in the entire month. At its 2014 peak the site took more than 200,000 a month . December’s figure was a 78 percent fall from the year before, and it puts activity near where the site sat in 2008, its first year. The line on that chart does not sag. It falls off a table. 
 The easy reading is that AI won, a clunky forum…
Last week some AWS customers opened their billing console and found they owed trillions of dollars. One account that had spent nineteen cents in June was projected to owe nearly $2.5 billion ; another showed a month-over-month change of 55 trillion percent. AWS confirmed the cause as a unit-pricing defect in its estimated-billing subsystem, said actual charges were untouched, and everyone had a…
Kimi K3, the new 2.8-trillion-parameter open model from Moonshot, shipped this week with a price list, and the number worth staring at is not the one everyone quoted. The headline was that an open model landed within a benchmark point or two of Claude and GPT. The number I keep coming back to is lower down the page: input costs $3.00 per million tokens on a cache miss and $0.30 on a cache hit, and…
Stripe and the private equity firm Advent International have offered around $53 billion for PayPal , $60.50 a share, roughly a 28% premium, backed by about $50 billion in committed financing. The headline writes itself: a fallen giant, once worth $360 billion in 2021 and now worth about $42 billion, gets picked up on the cheap. 
 That framing is true and mostly beside the point. The more…
In November, Michael Burry told his followers that the biggest names in AI were flattering their profits with an accounting choice, and put a number on it: about $176 billion of understated depreciation across the hyperscalers between 2026 and 2028, enough in his math to overstate Oracle’s earnings by roughly a quarter and Meta’s by a fifth. He was short Nvidia at the time, so weigh…
Microsoft ran the experiment most companies are only talking about. Over four months in early 2026 it put command-line coding agents, Claude Code and GitHub Copilot CLI, in front of tens of thousands of its own engineers and measured what happened . The adopters merged about 24% more pull requests than they otherwise would have, and the lift held for the whole study rather than fading as a…
A recent arXiv paper by the mathematician Jun-Yong Park, Automation Without Understanding , opens with an uncomfortable moment: the May 2026 disproof of a longstanding Erdős conjecture on unit distances in the plane, produced by an AI system, in a form few mathematicians could actually follow. The proof checks out. The understanding behind it does not sit in any human head. Park’s worry is…
Last September, Nvidia said it would invest up to $100 billion in OpenAI , which plans to spend much of it deploying Nvidia chips. That single sentence is why “circular financing” is now a phrase everyone in AI knows. The commitment has since been trimmed to a $30 billion equity stake , but the shape of the deal is the point: the supplier funds the customer so the customer can buy from…
This week OpenAI published what it says is a proof of the Cycle Double Cover Conjecture , a problem in graph theory that has sat open since the 1970s, produced by a frontier model left to grind on it with a swarm of subagents. Mathematicians are still working through whether it holds. Set that next to a smaller, less glamorous fact: the same class of model, pointed at a mid-size company’s…
Yesterday OpenAI shipped GPT-5.6 , a family of three models with a record score on a coding benchmark and a 1.5 million token context window. The feeds did what they always do: everyone raced to re-rank their tools against the new frontier. The most useful AI story I read this week was not about any of that. It was a reading-tutor startup explaining how it teaches a child in under a second , and…
On the 8th of July, OpenAI published an audit of the coding benchmarks everyone uses to rank models, including SWE-bench, and reported that roughly 30% of the tasks in the Pro set were broken in ways that quietly distort the scores. This from the company with every incentive to keep those numbers looking meaningful. When a lab audits the scoreboard it competes on and tells you the scoreboard is…
Yesterday a feature shipped in every new car sold in Europe, and not one buyer chose it. From 7 July 2026, Regulation (EU) 2019/2144 , the General Safety Regulation, requires every newly registered car to carry an Advanced Driver Distraction Warning system: an infrared camera near the wheel that watches your eyes and warns you when your gaze leaves the road for more than 3.5 seconds. It runs above…
An open-weight model just matched the best in the world for a fifth of the price, and the internet decided this was Anthropic’s problem. I think it is mostly yours. 
 GLM 5.2, an open-weight model from Zhipu AI, now does agentic coding work that is hard to tell apart from Claude Opus at roughly $4.40 per million tokens, against Opus at about $25 and GPT-5.5 at about $30. Martin Alderson…
Last week Mark Zuckerberg told Meta staff that agentic AI development “hasn’t really accelerated in the way that we expected” over the past four months. This from the company that branded 2026 the year agents would “really begin to work,” and that still plans to spend around $145 billion on AI infrastructure this year. 
 When the person with the most compute and…
A researcher found a way to steal the titles of your private YouTube videos using nothing but a comment. Google looked at the report and decided it wasn’t a security bug. 
 He was testing Ask Studio, the AI assistant YouTube added to its creator dashboard so channel owners can ask things like “what are people saying about my last upload.” The flaw shows up in what happens…
Two governments spent the last two years arguing with the same satellite company, and they made opposite mistakes. 
 South Africa spent years keeping Starlink out to protect a principle. Ontario let Starlink in without one, then had to pay to undo it. Between them is a genuinely hard question that a lot of countries, and a lot of businesses, are about to face: when you can’t build…
I have spent a good part of my career on two very different kinds of software: products built to reach millions, and internal tools built for a few hundred people who could not do their jobs without them. Only one of those gets called a success story. The other gets called “niche,” usually by someone who has never had to actually use the alternative. 
 That divide is exactly what…
Cloudflare turned on a toll booth on the open web this week, and it isn’t built for you. 
 On July 1 it launched the Monetization Gateway , which lets any site sitting behind Cloudflare charge for what it serves: a page, an API call, a slice of a dataset. Alongside it, the company helped set up the x402 Foundation, a Linux Foundation group of two dozen-odd companies building an open…
A professor at Brown University went public last week describing what they called mass AI fraud on a take-home exam. By their account, almost the entire class had used AI tools to complete it. The professor was furious. The students, apparently, were not. 
 I have a hard time sharing the outrage. 
 The exam was closed-book, unsupervised, completed at home over the internet. In 2026. The…
For two years my default advice to anyone asking about AI tools for development has been: pick a cloud API, learn to use it well, and don’t overthink the hosting question. The local-model path (downloading weights, fiddling with quantization levels, doing the RAM math) looked like a distraction from the actual work. 
 That advice is starting to age. 
 What shifted 
 A dense 27B…
Every few weeks a chart goes viral: “this one person is now richer than the entire GDP of [some country].” It collects a hundred thousand likes, and it is comparing two numbers that cannot be compared. Net worth is a total, everything someone owns added up, today. GDP is a flow, everything a country produces in a single year. One is a bank balance, the other is a salary. Putting them…
Three separate people sent me the Fintech Engineering Handbook this week, which is usually a sign the document is answering a frustration engineers already had. It covers the fundamentals of building payment systems: represent money as integers, implement double-entry bookkeeping, make operations idempotent, treat external webhooks as unreliable hints. Solid, clearly written, and worth…
Commerce Secretary Howard Lutnick published a letter that resolved two weeks of waiting: Anthropic can release Claude Mythos 5 to “more than 100 US institutions” listed in a document called Annex A, along with their foreign national employees. His stated determination: “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the…
OpenAI announced its first custom chip on Tuesday. It is called Jalapeño, it is built by Broadcom, and it is optimized for inference (specifically the real-time coding models that now account for a meaningful share of their traffic). Custom inference silicon is not a differentiator anymore ; it is the minimum, the thing every serious AI company builds once inference cost becomes the line item that…
Velocity is up, defects are stable, deployments are more frequent. Ask who in the team could refactor the payment module from scratch, and the room goes quiet. The understanding that used to build up through the act of writing has been offloaded somewhere faster and less visible. 
 Armin Ronacher wrote about this last week in a piece about what he calls “the coming loop” . His…
Mitchell Hashimoto just pledged $400,000 to the Zig Software Foundation. That brings his personal total to $700,000 across two donations. If you haven’t heard of Zig, it’s a systems programming language, a modern alternative to C that has been gaining real traction over the last few years. Hashimoto is the founder of HashiCorp (Terraform, Vault, Consul) and has, since leaving that…
Anthropic now asks for a government-issued photo ID and a live selfie before it will give you access to its most capable models. The verification runs through a third-party vendor called Persona. The official framing: “Being responsible with powerful technology starts with knowing who is using it.” 
 That sounds measured. But the reaction (hundreds of developer comments within…
Norway announced last week that students in grades 1 through 7 should, as a general rule, not use AI tools in school. Older students in lower secondary can use them cautiously, under teacher supervision. The intent, according to the government, is to study how to improve outcomes rather than to permanently prohibit anything. 
 The tech reaction was predictable. The more interesting part is the…
The safe way to read Apple’s server-chip plans is that Apple is building silicon to run Apple Intelligence on its own hardware instead of Nvidia’s. True, and boring. Every company at Apple’s scale now builds its own inference chip. Google has TPUs, AWS has Inferentia , Microsoft has Maia, Meta has MTIA. Mark Gurman reports Apple’s entry, codenamed Baltra, is being built…
Noam Shazeer co-authored “Attention Is All You Need” at Google in 2017 – the paper that ended the era of recurrent neural networks and made every large language model that exists today possible. He left Google to found Character.ai. Google paid roughly $2.7 billion to get him back in 2023. This week, he announced he’s joining OpenAI . 
 Three moves. Google was the…