Hi! Something happened this week that didn't make many headlines, but I think it's one of the more important things to happen in AI this year.
Perplexity quietly killed its advertising business. And the timing (right as OpenAI rolls ads out to millions of users) makes it feel less like a business decision and more like a statement.
Let’s discuss!
💸 The AI Ad War Got Its First Casualty
💀 Google VP: AI Wrappers Are Dead
🌍 World Labs Raised $1B to Fix AI's Blind Spot
📰 + Quick News That Matters
🛠️ + New Tools to Try
📚 + Weekend Picks
Subscribe to receive new posts and support my work.
What’s Happening: Perplexity executives confirmed to the FT that the company has permanently abandoned its advertising strategy. The startup was actually one of the first AI products to test ads — launching sponsored placements in November 2024 with partners like Indeed and Whole Foods. But it quietly stopped accepting new advertisers by late 2025, and now says it won’t revisit the model.
“A user needs to believe this is the best possible answer,” one executive told the FT, adding that ads risk making users “suspicious of everything.”
Instead, Perplexity is going all-in on subscriptions: the company currently has 100M+ users, $200M in ARR, and an $18B valuation driven by paid tiers at $20 and $200/mo.
The Context: This story is only half about Perplexity. The more interesting part is what it reveals about the industry split forming in real time.
The timeline is almost too clean. February 9: OpenAI starts showing ads to free and Go-tier ChatGPT users. February 18: Perplexity declares ads “misaligned with what users want.” And somewhere in between, Anthropic runs Super Bowl commercials with an unmistakable message: Claude is the place where your AI isn’t for sale.
The numbers make Perplexity’s decision look less risky than it sounds. According to Sacra, Perplexity’s advertising revenue totaled about $20,000 out of $34M in total received revenue in 2024. They weren’t giving up much.
But the signal matters more than the money. Two of the most trusted names in AI have now drawn a line, and the question is whether OpenAI’s scale advantage is strong enough to make users stop caring.
What’s Happening: Darren Mowry, who leads Google’s global startup organization across Cloud, DeepMind, and Alphabet, said on TechCrunch’s Equity podcast that LLM wrappers and AI aggregators have their “check engine light” on.
“If you’re really just counting on the back-end model to do all the work and you’re almost white-labeling that model, the industry doesn’t have a lot of patience for that anymore”
He pointed to Cursor and Harvey AI as examples of wrappers that actually work — but only because they built real vertical moats on top of the model.
The Context: There’s an obvious conflict of interest here — this is Google, a company that directly benefits when wrappers fail and startups come back to its infrastructure. But even setting that aside, the argument is hard to dispute. A startup whose only value is “we wrap GPT-4 better” inevitably loses the moment GPT-5 does the same thing natively. This isn’t a prediction. It’s already happening.
What’s Happening: World Labs, founded by AI pioneer Fei-Fei Li, raised $1B in new funding from Autodesk ($200M), Nvidia, AMD, a16z, Fidelity, and others. The company is building “world models”: AI systems that understand and generate 3D environments rather than text. Its first product, Marble, lets users create persistent, editable 3D worlds from images, video, or text prompts.
Target use cases: gaming, VFX, robotics, and scientific discovery.
The Context: Li put it cleanly: “If AI is to be truly useful, it must understand worlds, not just words.” That’s a harder problem than it sounds. Every major chatbot today reasons in text, which means it fundamentally doesn’t understand physics. World models are an attempt to fix that from the ground up.
What makes this round different from a typical hype raise is Autodesk’s $200M strategic check. Autodesk builds software for architects, engineers, and designers — people who live in 3D. When they put in that kind of money, it means someone with real commercial skin in the game believes this is more than a research project.
By the way, Li is not alone: Yann LeCun just left Meta to launch AMI Labs around the same bet, and Google DeepMind has its own world model program.
The Context Loop is a free newsletter.
If you like what I do, you can support me through a donation.
Sam Altman: “It Also Takes a Lot of Energy to Train a Human”: Pushing back on AI’s energy criticism, Altman noted that raising a human takes “20 years of life and all the food you eat.” The internet was not impressed.
His actual point — that per-query inference is more efficient than people think — isn’t wrong. It’s just that comparing a data center to human civilization is the kind of analogy that tends to bury whatever you were trying to say.
An AI Agent Deleted the Entire Inbox of Meta’s Head of AI Safety: Summer Yue asked the OpenClaw agent to suggest which emails to clean up, explicitly telling it not to act without her approval. It ignored her and started mass-deleting. She had to physically unplug her computer to make it stop, and later admitted it was a “rookie mistake.”
For the full breakdown check out my post here.
Anthropic Is at War With the Pentagon: The Pentagon is reportedly close to labeling Anthropic a “supply chain risk” after months of failed negotiations. The sticking point: Anthropic refuses to let Claude be used for mass surveillance or fully autonomous weapons — a line OpenAI, Google, and xAI have all agreed to cross. Awkward, given that Claude is currently the only AI deployed on fully classified U.S. military networks.
Claude Crashed the Cybersecurity Market: Claude Code Security launched in limited preview on February 20 — and CrowdStrike fell 8%, Okta dropped 9.2%, and the sector lost $15B in market cap in a single day. Unlike rule-based scanners, it reasons through code like a human security researcher. Nothing gets applied without human approval, but that didn’t stop Wall Street from panicking.
Share this post if you like it!
Brainstream — AI note-taking app that doesn’t just organize your thoughts, but acts on them: creates tasks, sends daily briefings, and drafts calendar events from your voice memos and quick notes.
Origami.chat — Find your ideal leads with a single plain-English prompt, without building lists or running manual searches.
Moda — Generate AI design mockups and actually edit them afterward, instead of starting over every time something looks wrong.
Toolspend — Tracks what your team spends across all AI tools in one dashboard, so the “wait, how much are we paying for Notion + ChatGPT + Midjourney?” conversation finally has an answer.
Google Pomelli — Turns a basic product photo into a professional studio-quality image instantly, without a photographer or editing skills.
SuperX — An all-in-one growth dashboard for X: analytics, scheduling, competitor tracking, and post ideas — for anyone building an audience on the platform.
Starnus — Finds and reaches your next potential customers on autopilot, designed for founders who don’t have a dedicated sales team yet.
Head of Claude Code: What happens after coding is solved | Boris Cherny — Lenny's Podcast
How AI agents could destroy the economy — TechCrunch
5 Interesting Startup Deals You May Have Missed — Crunchbase
AI Unicorn Founders Are Getting Younger — Fast — Business Standard
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.