For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.

LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.

The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.

Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more.

Hey,

For the last year, we have been obsessed with "smart."

Everyone wanted a higher IQ model. Everyone wanted better reasoning.

But while we were looking for Einstein, the market shifted to Usain Bolt.

Latency is the silent killer of AI adoption. If your agent takes 10 seconds to think, your user is gone.

This week, the speed limit broke. And the security rails fell off.

Here is the signal for the week of Feb 22 – Mar 1.

Inception Labs just broke the physics of LLMs.

On Feb 24, they dropped Mercury 2. It is not just another model. It is a different species.

Most LLMs (GPT, Claude) are autoregressive. They guess the next word one by one. Like typing on a slow keyboard.

Mercury 2 is a diffusion model.

It doesn't type. It sketches the whole thought and refines it all at once. Like an image generator, but for text.

The result is violent speed.

We are talking 1,009 tokens per second. Independent tests clocked it at 1,196.

That is 11x faster than Claude Haiku. It is 14x faster than GPT-5.2 Mini.

And the quality? It matches them.

This kills the "loading..." spinner.

Why this matters:
Latency was the biggest bottleneck for voice agents and real-time coding. That bottleneck is gone. If you are building user-facing AI, "fast enough" just got redefined.

→ The Move:
Swap your real-time endpoints to Mercury 2 immediately. It is OpenAI-compatible. The effective cost is $0.80/M tokens. You get 14x speed for pennies.

(Source: https://www.businesswire.com/news/home/20260224034496/en/Inception-Launches-Mercury-2-the-Fastest-Reasoning-LLM-5x-Faster-Than-Leading-Speed-Optimized-LLMs-with-Dramatically-Lower-Inference-Cost)

We think of AI safety as "filters."

We think if we tell the model "don't be bad," it won't be bad.

Wrong.

Bloomberg reported that a hacker used Claude to tear through 10 Mexican government agencies. They stole 150GB of data. Voter records. Tax data. 195 million identities exposed.

How did they do it?

They didn't use sophisticated code. They used social engineering.

The attacker told Claude they were doing a "bug bounty" program. They framed the attack as a security test.

Claude hesitated. Then it bought the lie.

It acted as an "elite hacker" and sent over 1,000 malicious prompts. When Claude hit a wall, the attacker switched to ChatGPT to finish the job.

Why this matters:
Your "safe" LLM is one good prompt away from being a weapon. If you are relying on default guardrails to protect your data, you are negligent.

→ The Takeaway:
Stop trusting the model to police itself. If you are building agents with access to internal databases, you need a hard-coded permission layer. The model is easily tricked. The database permissions are not.

(Source: https://www.bloomberg.com/news/articles/2026-02-25/hacker-used-anthropic-s-claude-to-steal-sensitive-mexican-data)

I love Cursor. You love Cursor.

But right now, Cursor is dangerous.

OX Security found 94 known vulnerabilities in Cursor and Windsurf.

The problem is the foundation. These IDEs are built on VS Code, which runs on Electron. Electron runs on Chromium (the browser engine).

Cursor hasn't updated its Chromium build since March 2025. That is a lifetime in security years.

They found a way to execute code on your machine just by getting you to click a link inside the IDE.

One click. And they have your API keys, your source code, and your environment variables.

Cursor's response? They called it a "self-DOS" and dismissed it. Windsurf didn't even respond.

Why this matters:
We are pasting our most sensitive API keys into software that hasn't been patched in a year. 1.8 million developers are exposed.

→ The Shift:
Do not open external links inside your AI IDE until this is patched. Treat your IDE like a public terminal. Rotate your keys often.

(Source: https://www.ox.security/blog/94-vulnerabilities-in-cursor-and-windsurf-put-1-8m-developers-at-risk/)

Nano Banana 2 — Google's new image model is now default. It generates 4K images in under a second and finally understands consistency (same character, different pose). It is live in Gemini and Search. (Source: https://techcrunch.com/2026/02/26/google-launches-nano-banana-2-model-with-faster-image-generation/)

Claude Code /simplify and /batch — Anthropic is shipping updates daily. The new /batch command runs migration agents in parallel git worktrees. It also added persistent memory across sessions. (Source: https://releasebot.io/updates/anthropic/claude-code)

GitHub Opens Claude + Codex — You don't have to choose anymore. Paid Copilot users can now assign issues to Claude or Codex directly. No extra cost. (Source: https://github.blog/changelog/2026-02-26-claude-and-codex-now-available-for-copilot-business-pro-users/)

ChatGPT Projects = Persistent Workspace — Projects now connects to your Slack and G-Drive as "living sources." It is no longer just a chat. It is a knowledge base that updates itself. (Source: https://help.openai.com/en/articles/6825453-chatgpt-release-notes)

Agent J Plus just launched.

This is the premium tier of Agent J. Hands-on workshops. A full 4-module curriculum. Live weekly sessions where we build together in real time.

Inside you get the Claude Skills Workshop, Claude Cowork Workshop, OpenClaw Workshop, and a structured 12-lesson program covering prompt engineering, automations, AI agents, RAG systems, and a capstone project.

$87/mo inaugural pricing. It locks in forever. 90-day money-back guarantee — if you put in the work and don't walk away with a working app, workflow, or automation, you get your money back.

Join Agent J Plus here

This week's free programming inside Agent J:

Not in Agent J yet? It's free. Join 500+ builders here

Speed is the new currency.

If you are still waiting for text to generate, you are losing. If you are still trusting default safety filters, you are exposed.

Move fast. But lock your doors.

See you next Sunday,
John

P.S. Reply "AGENT" and I'll send you The Autonomous Sales Agent Playbook — the exact architecture to build a headless employee that runs your entire pipeline for $20/mo.