|
AI Builders Digest
Monday, October 19, 2026
|
|
The big story this week isn't a model release or a funding round. It's two data points sitting next to each other: AI agents filing your taxes while coordinating with your CPA, and the guy who built Replit clarifying which AI company is now powering all of it. The infrastructure is quietly becoming real, and the allegiances underneath it are shifting.
|
|
---
|
|
01
|
An AI agent coordinated with another AI agent to file someone's taxes. It took 20 minutes.
|
|
|
Thibault Sottiaux posted something that got 4,300 likes over the weekend, and it deserves the attention. His work AI agent ("Burger") coordinated with his personal AI agent ("Dorito") over email, retrieved documents, then looped in his CPA to complete a tax filing. No human in the middle of the handoffs. Twenty minutes start to finish. Sottiaux's closer: "We're still so early."
|
Why it matters: Tax filing is one of those tasks that's annoying specifically because it requires gathering documents from multiple places, coordinating with multiple parties, and not making mistakes. If two agents can do that handoff reliably, the list of "this requires a human to orchestrate" tasks just got shorter. Your accountant's real competition isn't TurboTax anymore.
|
|
Source →
|
|
---
|
|
02
|
Replit CEO clarifies: OpenAI is the default, but Anthropic isn't going anywhere
|
|
|
Replit CEO Amjad Masad walked back some apparent confusion from his TechCrunch Disrupt appearance. OpenAI is now Replit's "preferred partner" powering Replit Agent by default, chosen on performance, cost, and partnership strength. But Masad was clear: Replit isn't exclusive. Users can still pick Anthropic models or others, and the platform intends to keep it that way.
|
Why it matters: When the default model on one of the most widely used AI coding platforms shifts, millions of developers start getting answers from a different engine without changing a setting. OpenAI winning "default" is a distribution win that doesn't show up in any benchmark. Anthropic stays on the menu, but being the non-default option on a platform this size is a different business.
|
|
Source →
|
|
---
|
|
03
|
You probably have 40+ apps with access to your Google account. Here's a prompt to fix that.
|
|
|
Peter Yang pointed out that he had 45 apps and agents connected to his Google account, and shared a prompt to audit them: open the Google permissions page, review every connected app, distinguish basic sign-in from apps that can read your Gmail or Drive, then decide what stays. The prompt asks Claude (or whichever AI you use) to ask you about each app rather than guessing.
|
Why it matters: Every AI agent you've connected to Google in the last two years probably has broader permissions than you remember granting. The audit takes 20 minutes and the downside of not doing it grows every time you add another agent to your stack.
|
|
Source →
|
|
---
|
|
04
|
"What did we pay attention to before AI?" "Front end JavaScript frameworks."
|
|
|
Nan Yu shared a line from a team offsite that landed: developers sitting around reminiscing about a time when the big debates were about React versus Vue. It got 56 likes but it's the most honest thing in today's feed.
|
|
Source →
|
|
---
|
|
05
|
Vercel's Guillermo Rauch on what actually matters when building with AI
|
|
|
Rauch shared a quote that's getting traction in builder circles: "What matters is the end product, whether it's a game or an engine." Short, but it's a direct pushback on the endless debate about which model or framework to use.
|
Why it matters: The builders spending 2026 arguing about AI infrastructure are making the same mistake as the ones who spent 2012 arguing about JavaScript build tools. The product ships or it doesn't.
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|