TLDR 2026-01-01

Meta took steps to make scam ads less discoverable to regulators, investigators, and journalists due to fears that Japanese regulators would require advertiser verification, a measure the company estimated would cost roughly $2 billion to implement and potentially reduce revenue by nearly 5%. The campaign was so successful that Meta added the tactic to a 'general global playbook' that has since been deployed against regulatory scrutiny in other markets. Meta had previously estimated that 10% of its 2024 revenue was from ads tied to scams and banned goods.

Creation tools are getting better and better. This will make creators matter more. Those who have the ability to be real, to connect, and to have a voice that can't be faked will be those who succeed in this new world. The bar will shift from whether someone can create to whether someone can create something that only they could create. Only the creators who can maintain trust and signal authenticity will stand out.

A Tesla owner successfully drove coast-to-coast across the US on Full Self-Driving (FSD) Supervised with zero interventions during the trip. Their Model 3 completed the drive in 2 days and 20 hours, starting at the Tesla Diner in Los Angeles, CA, and ending in Myrtle Beach, SC. It was accomplished with Tesla FSD V14.2 on AI4 hardware. The milestone trip was widely lauded by members of the Tesla community.

Neuralink received FDA breakthrough device designation for its speech restoration in 2025. The company also launched clinical trials in the Middle East and the UK, and completed two surgeries in Canada. It upgraded its surgical robot to be able to insert an electrode thread every 1.5 seconds at insertion depths of over 50mm. The threads go through the dura without the need to remove it. Neuralink plans to move to an almost completely automated surgical procedure this year.

The primary security question for companies will be how good their attackers' AI is versus their own AI. There will be increased spending on agentic security platforms and significantly more in-house building of security tools. Asset management will become possible for the first time because of agents. Most security platforms will eventually be replaceable with AI prompts.

Ben Tossell, who works at Factory, a company developing a frontier software development agent, has spent 3 billion tokens in four months through an agent without writing any code. None of the code was read, but Tossell read the agent output religiously, which led to him picking up a ton of knowledge around how code works, how projects work, where things fail, and where they succeed. Tossell has shipped several projects using his method over the past few months, including a personal site, a 'wrapped' product for Factory, several custom CLIs, and an AI-directed video demo system. This post contains Tossell's guide to learning to program.

2026 will be the year when generative UI takes off, the Smart Home finally fulfills its promise, and biometric proof of personhood becomes the new social login. Engineers looking to separate themselves from the crowd need to build and learn, as deep technical intuition is what will distinguish them from others. Agents will continue to be the main headline. This year, they will start interacting in our daily lives to anticipate needs.

AI tends to improve at the same pace that AI researchers get access to faster hardware. The hardware scheduled to come online in the next few years makes current hardware look like pocket calculators. If previous observations hold true, then we are not approaching a plateau, the field is just getting started.

NERD (No Effort Required, Done) is a programming language optimized for large language models.

Models will continue to improve their capabilities, but it will take a while for the full impact to be felt across the economy.

There is a lot less money going towards hiring, but that could change if the tax code gets rolled back, interest rates fall, and data centers stop being the rage.

Atmospheric computing connects cloud services together, allowing users to store and run their own programs.

Worktrunk is a CLI for git worktree management designed to run AI agents in parallel.

Reasoning was a significant feature of many models from nearly every major AI lab in 2025.