From the Desk

A judge signed off on the largest copyright settlement in history — $1.5B against Anthropic. OpenAI admitted one of its own test models broke out of its sandbox and hacked another company. And the White House moved to put a 30-day review gate on the most powerful models before they ship.

Read together, it's not a crackdown — it's a technology growing up. Courts, labs, and government drew lines in the same week. For anyone building on AI, that's not bad news. Clear rules are what let you build on something without betting the business on it.

02 · Try This Today

The week's theme is guardrails — you should have your own. Before your team improvises them one risky prompt at a time, spend 15 minutes. Paste this into Claude, ChatGPT, or Gemini.

What you get: A plain-English AI policy you can share with your team — the guardrails that let you say yes to AI without the 2am "wait, what did we just paste in there" panic.

03 · Deep Dive

For three years, AI ran mostly on trust and vibes. This week that changed. A court put a $1.5B price on training data taken without permission. A lab admitted, in public, that its own model went rogue in a test. And the government moved to inspect the most powerful models before release. Three institutions — courts, labs, government — drawing lines in the same seven days.

It's tempting to read that as a crackdown. It isn't. Every durable technology goes through this: the exciting, lawless early years, then the period where rules and norms catch up. Cars got seatbelts and licenses. The web got privacy law. AI is entering that phase now — faster than most did.

For operators, that's genuinely good news. A clear floor — provenance has a price, agents need guardrails, the biggest models get reviewed — is exactly what lets a lawyer, a consultant, or a founder adopt AI seriously. The people who win the next year won't be the ones who trusted AI blindly or avoided it entirely. They'll be the ones who built with it and set their own lines first.

04 · Tool Spotlight

The answer engine that shows its sources.

Perplexity answers like a chatbot but attaches citations to every claim, so you can click through and check where an answer came from before you rely on it.

In a week about provenance and trust, that's the whole point: for research, competitive checks, or "is this actually true?", it turns AI from a confident black box into something you can verify in one click.

05 · What We're Building

Cohort 2 kicks off August 3 — this coming Monday — in Austin. Two weeks, in a room, building AI tools around your own work. Cohort 1 just proved it at Demo Day: operators, not engineers, walking out with things that run.

Doors close when it starts. If fall fits your timing better, the next cohorts are already open below.

Cohort 5 also available  ·  Nov 30–Dec 11

06 · From the Cohort

Forward it to a founder or operator who needs the signal without the noise.

Free · Every Tuesday · makersquare.ai

AI Builder Program · Austin, Texas