Issue 43 — 2026-06-24

Lenny's Newsletter just published a walkthrough of how to actually build AI agent loops — the kind where a model doesn't just spit out one answer, but keeps thinking, checking its work, and trying again until it's done.

The piece lays out the mechanics: prompt the agent, let it reason, review its output, loop back if something's off. It's the difference between asking a friend for a recommendation and having them research, cross-check, and come back with a real answer.

What's interesting is that these loops are becoming the standard way to get AI to do actual work — not just generate text, but handle decisions, corrections, and multi-step tasks. That shift is why so many teams are moving past simple chatbots toward agents that can run through a process on their own.

Why this matters for us: if you're using AI to get things done, you're probably already running these loops whether you know it or not.

Dave's research on "timescales" shows that AI integration usually fails not because the technology is bad, but because teams are measuring the wrong things. We keep asking whether AI works when what we should be asking is whether our product still makes sense with AI in it. The tools change. The product doesn't.

This hits our communities hard. When a restaurant uses AI for ordering, when a clinic uses it for scheduling, when a small business automates invoices — it only works if someone who understands the business actually steers the integration. Otherwise la migra app is just another tool nobody knows how to use.

The fix is simple and nobody follows it: put product people at the center of AI decisions, not the engineers who built it. Because a tool that solves the wrong problem is worse than no tool at all.

Why this matters for us: When product decisions ignore the people who actually use the tools, we end up paying more for tech that doesn't work — and that money comes out of our pockets.

The TLDR newsletter is back with a fresh batch of what's worth knowing this week. It's been running a while now — Dan Kohn has been putting out the digest for folks who want the week's tech news without the noise. The format is tight: no long essays, no forced takes, just the stuff that matters and why it matters.

What's been on deck lately: AI tools shifting from demo to actual use, payments and banking getting reworked for the folks who work for a living, and immigration tech that's finally moving past the hype phase. Not everything that gets called a "disruption" is one — some of it is just the same old game with a new coat of paint. The newsletter keeps it straight.

Why this matters for us: When the big tech news cycles through, it's usually the same companies telling the same stories. TLDR cuts through that so la gente can figure out what's real and what's just noise.

Meta is hitting pause on its employee tracking program after the whole company was able to see the sensitive data it had been collecting. The internal rollout revealed that what started as a targeted monitoring effort had quietly become visible to everyone.

The pause gives Meta time to figure out what to do next. The company's own workers are now on the inside of a surveillance system that many tech companies have been rolling out for years — the kind of thing that starts with a small pilot and ends up quietly watching everyone.

Why this matters for us: When big tech treats its own workers like lab rats, the rest of us get the memo that surveillance is just how things are supposed to work.

Instagram is testing a new TV app that surfaces longer-form episodic content and live shows. The goal is straightforward: give people something to sit through between scrolling. Instead of short clips and Reels, the app is building out series-style content and live broadcasts that compete with Netflix, Disney+, and the rest of the streaming pack.

This isn't Instagram trying to be a content studio. It's trying to be a destination — a place you can open and watch something without thinking too hard about it. Episodic formats encourage you to come back. Live shows create urgency. Together they pull you off the infinite scroll and into something with a beginning, middle, and end.

Why this matters for us: as more of our media migrates to apps that reward watching over scrolling, the shows, creators, and creators of color who land there get seen — and the algorithms that shape what gets promoted start working for us instead of against us.

Meta's employee tracking program is on pause. The company had been tracking location, keyboard strokes, and other sensitive data, then let the whole company see what was being collected. Once it became visible, the backlash followed.

The program, which was part of Meta's broader workplace monitoring push, was rolled out to employees across the company. But something shifted when the data itself became public. Workers started seeing what they'd been sending, and the scale of what was being tracked became harder to ignore.

This is the same pattern we're seeing across Big Tech: companies build surveillance tools for their own workers, then realize the data they've collected is just as valuable when exposed. Meta's pause suggests the program's design had blind spots — not just in what it collected, but in how it shared it.

Why this matters for us: Every tracking system built in Silicon Valley eventually gets repurposed for us — our data, our movements, our habits — so when Meta's own workers push back, it's a reminder that the tools built on our backs can be turned around.