Daily Brief

Parag Agrawal is making a bet that goes against two decades of web search: agents will query the web a thousand times more than humans ever have, and the infrastructure built around human clicks is wrong for them. The former Twitter CEO, now founder and CEO of Parallel Web Systems, explains why Parallel treats human click data as a bug and trains on agent feedback instead. He unpacks the counterintuitive choice to ship a search agent before a search engine, building an index incrementally, and how the new Turbo product cut agentic search to 200 milliseconds. But the problem Parag keeps returni

Parag Agrawal (former CEO of Twitter, now founder of Parallel Web Systems) on the single bet the company was built on: agents will search and browse the web 1,000x more than humans ever have — which means reinventing both the technology and the business model underneath it. #ai #internet

Is synthetic data a general method that scales with computation? Rich Sutton's answer is immediate. The reasoning is the Big World Hypothesis, which Khurram Javed wrote up as a paper: the world holds infinitely many things to learn, so you can generate datasets forever and there will still be more — and a human always decides what's worth generating. #ai #syntheticdata #machinelearning #llm

Their onboarding form asks how you heard about us. On April 13th the answers started spiking, and the single largest source of inbound for c15t is now an LLM telling someone to install it. Christopher Burns is not a researcher and says so twice. He founded Inth, built c15t, the open source consent banner library, and reckons he has been hacking on this only slightly longer than the room has. The Collison brothers used to install Stripe by taking your laptop off you; going through Y Combinator, Burns found himself handing people a prompt instead. Good developer experience primitives turned out

Over a week off, Jeffrey Wang built an AI clone of himself. He analyzed 760 of his own emails to derive his voice, down to averaging 18 words and signing off with best rather than sincerely, then turned hundreds of past decisions into evals to calibrate the agent's judgment against his own. Anyone at Exa can now ask Jeffbot to draft a Slack message. Wang cofounded Exa, a search engine for agents, and his framing is that go to market is now an AI engineering problem. The product versus distribution argument he treats as settled: you need both. Underneath, go to market is a data problem, which m

The clearest explanation of web search you'll hear. Parag Agrawal breaks down crawling, indexing, retrieval and ranking — and why the whole thing is really a billion-to-billion matching problem: hundreds of billions of pages, hundreds of billions of queries, and one job, matchmaking between them.

Cloudflare's weekly go to market summary is written by three agents in sequence: one drafts from the data, a second checks that draft against the data, and a third, the tone agent, rewrites it so risks and opportunities land with equal weight. Justin Joyce's team read every run for two to three months before trusting it. Joyce works in sales operations and strategy at Cloudflare, after seven years on the machine learning side, and his diagnosis is that traditional go to market does not scale. Operations rebuilds the same analysis in spreadsheets every week, or ships dashboards that meet most n

Ask an assistant to compare code intelligence tools and Sourcegraph comes up 65 percent of the time. Describe the actual pain instead, that you keep breaking downstream services when you change shared libraries and cannot see all the consumers, and it comes up zero percent. It suggests your developers write a wiki page. Stephanie Jarmak ran that experiment. The gap between shopping and hurting is invisible without measuring it. Jarmak, an astronomer a year ago with no commits, now has 12,000 and maintains an open source multi agent orchestration framework under the title agent advocate. The ta

Rich Sutton gets called a radical. He thinks he's the one thinking normally. Before the current AI moment, nobody would have said "continual learning" — learning that wasn't continual made no sense as a concept. We act, we learn, we keep learning. That the field needed a special term for it says more about the field. #ai #machinelearning #llm

Full episode: https://www.youtube.com/watch?v=aV26V1UvkJw Me on twitter: https://x.com/dwarkesh_sp

Voice agents are one of the hottest use cases in enterprise right now, but also one of the hardest to actually take live. Getting latency low enough to feel human without dumbing down the responses, making reliable tool calls to a CRM without dropping the customer mid-call, building fallback models for when your main provider goes down. None of it is as simple as the demos make it look. Basil Chatha hosted a fireside chat with five eng leaders who deal with this stuff every day: Basia Sudol (Head of Enterprise Solutions, Decagon), Varun Singh (CPTO, Daily), Steven Diaz (FDE Manager, Vapi), Ty

Most of AI is pointed at language. Anima Anandkumar points it at the physical world, weather, fusion, materials, the systems physics writes down as equations, and finds the usual playbook breaks. There isn't enough data, the resolution is brutal, and no transformer is big enough. Her way through is older than deep learning: build the structure of the world into the model. It already produces weather forecasts that rival the supercomputers on a single GPU. Co-founder of Accelerated Understanding, Anima has spent two decades in AI, from the theory that predates deep learning, through scaling it