D3 Daily — Tuesday, September 15, 2026

Dario Amodei wants to pace the frontier while Beijing says no thanks, and two SpaceXAI designers are running Figma from the gym. Here is what actually matters from today's feed.

Today’s Highlights

We must pace the frontier

Sidebar.io
Inspiration

Amodei lays out a three-step framework for AI companies to deliberately slow capability progress so safety systems can catch up. Read it alongside the news that the White House rejected the slowdown push and China offered no matching promise. The safety argument just got a geopolitical counterargument, and nobody has a clean answer for it.

How Grok Bot designers use AI agents to build personal sites and product prototypes

Lenny’s Newsletter
Inspiration

John Bai and Peng Zheng walk through running Figma from the gym, updating a portfolio with one photo, and turning shower thoughts into working prototypes. Forty-two minutes, and the useful part is not the tooling. It is watching two working designers describe where they still intervene by hand.

Planning with agents

Sidebar.io
Inspiration

Appleton argues that collaborative planning with agents needs thicker interfaces and better boundary objects, not just a chat box. This is the design problem hiding inside every agent workflow you have shipped this year. Worth reading before your next planning tool decision.

Optical vs. mathematical alignment

Sidebar.io

Software places objects with mathematical precision. Your eyes disagree constantly. A short, sharp reminder that the grid is a starting point, not a verdict, and that optical correction is still a skill worth having on the team.

The death of the deliverable

UX Collective
Inspiration

AI did not kill the wireframe. It exposed that the wireframe was never the point. If your team still measures output in artifacts rather than decisions, this one will sting in the right way.

Polylane replaced up to 18 AI agents with one

The Neuron AI

Polylane collapsed an autofix workflow that used as many as 18 agents down to a single one after handoffs kept losing evidence. The lesson is not that fewer agents always wins. It is that delegation has a cost, and better models keep changing where that line sits.

Why OpenAI’s $300M-plus camera deal could be about much more than photos

The Neuron AI

OpenAI acquired computational-photography startup Glass Imaging at a valuation north of $300 million. The interesting question is not prettier photos. It is what happens when the same system constructs and interprets what a camera sees, and where the user sits in that loop.

Quick Hits