|
AI Builders Digest
Thursday, October 22, 2026
|
|
Aaron Levie is back with another entry in his ongoing argument that the best time to build AI products was yesterday. Meanwhile, the mystery project we've been tracking for two weeks finally has a concrete feature worth discussing. And OpenAI just quietly hired someone to think about the thing that keeps Sam Altman up at night.
|
|
---
|
|
01
|
The mystery project reveals itself: Thibault Sottiaux ships a real caching overhaul
|
|
|
Two weeks ago Sottiaux bought a domain. Last week he reset on day 15. Today, day 16 delivers something concrete: prompt caching extended to 12 hours via API, and for logged-in users across Codex and Work, the cache lives forever. No penalty for returning to an old conversation hours later, no cold-start cost when you pick up where you left off.
|
Why it matters: If you've ever had an AI coding session interrupted and come back to find you're paying full price to re-establish context, this is the fix. Permanent cache for returning users is a pricing model change disguised as a technical feature. The teams building on top of this infrastructure just got a meaningful cost reduction for any workflow that isn't strictly one-shot.
|
|
Source →
|
|
---
|
|
02
|
Aaron Levie declares the "LLM wrapper" era officially over
|
|
|
Box CEO Aaron Levie argues we now have a real form factor for AI: autonomous agents dedicated to specific tasks or domains, with genuine tool use and computer use behind them. His point is that the expectation has shifted from "generate text" to "accomplish work," and that's a qualitatively different bar.
|
Why it matters: Yesterday we covered Levie's case that better models create more room for product differentiation, not less. Today he's extending that argument: the new baseline is task completion, not text generation. If your product's value proposition is still "it writes things for you," you're already a generation behind.
|
|
Source →
|
|
---
|
|
03
|
OpenAI hires Aleksander to lead safety and security ahead of superintelligence
|
|
|
Sam Altman posted that Aleksander has joined OpenAI to take on safety and security challenges, framing it in unusually direct terms: "the safety and security challenges in front of us as we confront superintelligence are immense."
|
Why it matters: OpenAI makes a lot of announcements. This one is notable for the language Altman chose. "Confront superintelligence" is not a phrase companies use in routine hiring posts. Whether you take that framing at face value or read it as positioning, the company is now publicly treating this as an urgent leadership-level problem, not a research footnote.
|
|
Source →
|
|
---
|
|
04
|
Google's Josh Woodward amplifies Sundar's frontier commitment pitch
|
|
|
Google Labs VP Josh Woodward shared Sundar Pichai's line: "We are so committed to being at the frontier. It's in our DNA. We won't do it any other way." The post links to what appears to be a recruiting or partnership page.
|
Why it matters: This is a recruiting message dressed up as a conviction statement, and it's worth reading as such. Google has been saying this for two years while OpenAI and Anthropic set the agenda. The question isn't whether Google is committed to the frontier. It's whether "it's in our DNA" translates into shipping things people use.
|
|
Source →
|
|
---
|
|
05
|
Peter Steinberger heads to Cloudflare Connect to talk agents
|
|
|
PSPDFKit founder Peter Steinberger flagged he'll be at Cloudflare Connect to discuss agents with Cloudflare's Dina Kozlov.
|
|
This is a conference appearance notice, nothing more. Worth knowing Cloudflare is running a Connect event with an agents track if you're following where infrastructure conversations are happening this week.
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|