Hitesh Sarda

Google’s search results have started doing things a trillion-dollar product should have grown out of years ago.

Two years ago Microsoft stood up a separate team, hired an outside AI veteran, Mustafa Suleyman, to run it, and gave the effort its own personality: a small animated character named Mico, a “Copilot Labs” for experiments, a feature for group chats with AI,

On October 1, a SpaceX Falcon 9 is due to carry four of Google’s TPUs into orbit under Project Suncatcher, the company’s bet that AI chips can run on satellites bathed in eight times more sunlight than any panel on Earth gets.

Michael Heap tells a story about an SVP who, five minutes into a detailed incident readout, stopped the meeting.

In the space of two days this week, three frontier labs updated their flagship line.

A 2023 essay is doing the rounds again, The LLMentalist Effect.

TypeSafe shipped something unusual this month: a model that never writes a sentence.

Two people who agree on almost nothing about AI just landed on the same rule for writing with it.

In May, Google’s Gemini broke into three real companies.

Last week OpenAI shipped Astra for Law, a legal-research layer on its GPT-6 model with a search index over 230 million pages of US case law, statutes, and court rules.

Happy Ganesh Chaturthi!

A developer clicked a fake “iPhone Storage is Full” ad served through Google, reported it, and got back a note saying the ad complied with Google’s policies.

Between May and June 2026, a swarm of AI agents uploaded more than 2,000 malicious packages to RubyGems, the registry the entire Ruby ecosystem installs from.

Shopify spent five years as one of React Native’s loudest champions.

Meta launched Muse this week, a personal AI agent that books your travel, pays bills through Stripe, orders online, and manages your calendar by reading your email, all from inside WhatsApp, an app with more than two billion users.

Bryan Cantrill has a good phrase for the person who pastes a language model’s output onto LinkedIn and hits publish: their intellectual fly is open.

Three websites published 215,128 pages titled “best software,” most of them machine-generated, and an AI search engine treated a good chunk of them as evidence.

Someone whose code I respect wrote this month that working with AI feels more like leadership than coding.

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.

Google published a post last week arguing that Go is an ideal language for AI-assisted software engineering.

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.

There is a short post going around about how one developer uses LLMs to learn hard topics.

On Black Friday last year, Shopify was clearing $5.1 million in sales a minute at peak.

A hundred-year-old economics idea spent this week near the top of the tech feeds, explained through Mario Kart of all things.

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 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 wh…

A team at MIT Sloan and Stanford ran a study I keep thinking about.

A team at Manifest built an LLM router in March, ran it across 7,000 users for four months, and then switched it off.

Bottleneck Labs handed a frontier model a real company and told it to make money.

Last week Anthropic published something that should reframe how you decide where to put AI to work.

Roughly 90% of people click “accept” on a cookie banner when rejecting takes an extra click or two.

Anthropic just removed more than 80% of the system prompt behind Claude Code and reported no drop in performance.

Most people I talk to hold two thoughts at once and don’t notice the tension.

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.

A lender will happily write a loan against a warehouse, a fleet of aircraft, or a bulk carrier.

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.

On July 14 someone deleted a country’s record of who owns what.

A group of researchers ran a small, clever experiment this month.

In December 2025, developers asked Stack Overflow 3,862 questions in the entire month.

Last week some AWS customers opened their billing console and found they owed trillions of dollars.

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.

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.

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

Microsoft ran the experiment most companies are only talking about.

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…

Last September, Nvidia said it would invest up to $100 billion in OpenAI, which plans to spend much of it deploying Nvidia chips.

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.

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.

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.

Yesterday a feature shipped in every new car sold in Europe, and not one buyer chose it.