Articles

A New Vision for Human Work After Automation

Frontier models are trained to satisfy the average user, which is why their outputs look impressive in a demo and fall apart the moment you apply them to a project you care about. Dan Shipper closes Thesis: 2027 with a vision for human work after automation—showing how we can build and work with AI that extends our perception and agency rather than replacing them.

Compound Writing: How to Make Your AI Know What You Know

After automation, decisions you once made yourself become decisions you have to equip an AI to make for you. That means deciding what AI needs to know, how to organize that knowledge, and when to bring it into the workflow. In this talk, Katie Parrott will show you how she works through those choices in Compound Writing and how you can apply the same approach to your own system.

From Vibe Checks to Evals

Turn your instincts about AI quality into repeatable evaluations. Mike Taylor explains how to choose human, model, and code-based checks so you can improve an agent, catch regressions, and compare models without losing your judgment.

Keep Going: Autoresearch and Learning Environments

Join Alex Duffy for a practical session on using AI to achieve ambitious goals—even those without a precedent. Explore the data industry, learn how reinforcement learning (RL) environments work, and discover how to train your own models beyond language models to solve problems and succeed at specific tasks.

Welcome to Thesis

Dan Shipper opens Thesis with a look at what becomes possible when AI takes on more of our work—and what that asks of the people building and using it.

Google is sunsetting its AI coding agent, Jules.

The Google graveyard: all the products Google has shut down

Amazon's 3Q results miss analysts' estimates, but posts solid sales gains in cloud computing unit

NEW YORK — Amazon offered a cautious sales outlook for the current fourth quarter that encompasses the critical holiday shopping period after delivering fiscal third-quarter results that came in below Wall Street analysts’ expectations.

You can generate and edit Google Docs files directly in Gemini chats now

Posts from this topic will be added to your daily email digest and your homepage feed.

My First Week With an AI Coworker

Plus: Why we’re skipping GPT-6.1 Terra for Gemini 4 Argon, and a prompt that keeps track of what you owe people

Granola crunches up a dictation feature to compete with Wispr

Uno reverse! After Wispr added meeting notes to its dictation app, Granola is adding dictation to its meeting notetaker. The voice AI wars are getting interesting.

Apple and OpenAI head to trial in July.

The judge in Apple’s trade secrets lawsuit against OpenAI has set a trial start date of July 19th, 2027, as reported by 9to5Mac.

How We Detect Frontier Model Misalignment

As we train increasingly capable AI models, we need reliable ways to detect when their behavior departs from what we intended. Agents can misbehave and take unauthorized actions during training, evaluation, or deployment, with potentially serious consequences. In order to detect these failures—and, where necessary, to intervene in time—we monitor how agents reason, what actions they take, and how those actions affect their environment.