The biggest shift in AI right now isn't about smarter models. It's about cheaper ones.

For the past year, running serious AI felt like a luxury. Top-tier models cost $15–$20 an hour. That made autonomous agents fun to demo but hard to justify running 24/7. Now frontier-class models are showing up at $1/hour.

That pricing change might end up mattering more than any benchmark improvement.

Meanwhile, Spotify casually revealed their best engineers haven't manually written code in months. Google dropped a free marketing agent that replaces work you'd have paid an agency thousands for. And autonomous agent workflows are starting to look less like experiments and more like infrastructure.

In today's edition:

Let's get into it.

Running a frontier AI model for an hour of sustained work — Claude Opus, GPT-5-class — typically costs $15–$20.

MiniMax just dropped M2.5, claiming comparable real-world performance at roughly $1 per hour.

Pricing: $0.30 per million input tokens, $1.20–$2.40 per million output. Cheap enough to double-check the numbers.

What really got my attention wasn't the press release — it was OpenHands independently benchmarking M2.5 as the top "unlocked" model for coding agents, beating options that cost 10x more.

Will it dethrone Opus across every task? Probably not. That's not the point.

The point is what happens when capable-enough AI costs 90% less.

At $15–$20 an hour, always-on AI agents are a fun demo. At $1 an hour, they're a line item you don't think twice about. That's the difference between "let me try this" and "let me deploy fifty of these."

The bottleneck for autonomous agents was never just capability — it was unit economics. This kind of pricing shift quietly changes what's buildable.

Worth watching.

Since OpenClaw launched, everyone's asking the same thing: "Cool, but what do I actually do with it?"

Here are six workflows worth paying attention to.

1. Second Brain
Text it anything — ideas, quotes, reminders. It remembers everything and makes it searchable. No Notion rabbit holes. Just "remember this" and move on.

2. Morning Brief on Autopilot
Every morning at 8am, it sends you relevant news, ideas, your tasks, and tasks it already completed overnight. You wake up with work done.

3. Content Factory
One agent researches trends. One writes. One generates visuals. Runs daily without you touching it.

4. Business Builder
It scans Reddit and X for real pain points, then builds a product that solves them. Actual code, not just ideas.

5. Goal-Driven Task Engine
Brain dump your goals. It breaks them into daily tasks and starts executing — not just organizing.

6. Build Your Own Apps
Hate your to-do app? Tell it to build a better one. It ships it, connected to all your AI memory.

The real shift?

None of this is new in theory. What's new is the economics.

At $20/hour, these are experiments. At $1/hour with models like MiniMax M2.5, they're infrastructure. That's the difference between testing one agent and deploying fifty.

That's when it stops being a toy and starts being infrastructure.

This one stopped me mid-scroll.

During Spotify's Q4 earnings call, co-CEO Gustav Söderström dropped a wild claim: their best developers haven't manually written code since December.

Instead, they're using an internal tool called Honk — powered by Claude Code and other AI — to fix bugs from Slack on their phones, generate features remotely, and push to production before they even get to the office.

The results speak for themselves. Spotify shipped 50+ new features in 2025, including AI-powered playlists and audiobook tools, crediting AI for the speed.

They also flagged something interesting — their proprietary dataset on how people feel about music gives them a moat that generic LLMs can't easily replicate. Nobody's training on "what makes a good rainy Sunday playlist" at scale.

The bigger picture here isn't really about Spotify.

It's that we've quietly crossed a line. AI isn't pair-programming with developers anymore. At some companies, it's become the primary way code gets written.

And if Spotify's top engineers are already there — everyone else is on the clock.

Google just quietly dropped something worth paying attention to. It's called Pomelli, and it's basically a free AI marketing agent.

You give it your website. It analyzes your brand — colors, tone, messaging, style — and generates full campaign ideas that actually sound like you. Captions, images, headlines, ready-to-post content. Everything's editable if you want to tweak it.

A year ago, this kind of output would've cost you $5,000/month from an agency. Now it takes two minutes and costs nothing.

If you're running a startup, ecom brand, newsletter, or any kind of personal brand — honestly just go test it. There's no reason not to.

Data organizer

What it does: converts unstructured text into a structured JSON table by identifying key entities and organizing their related information into clearly defined fields.

Your task is to take the unstructured text provided and convert it into a well-organized table format using JSON. Identify the main entities, attributes, or categories mentioned in the text and use them as keys in the JSON object. Then, extract the relevant information from the text and populate the corresponding values in the JSON object. Ensure that the data is accurately represented and properly formatted within the JSON structure. The resulting JSON table should provide a clear, structured overview of the information presented in the original text.

What it's for: Turning messy, unstructured text into clean, structured JSON data for easier analysis, storage, or automation.

Works with: ChatGPT 4, Claude, Gemini

Use this ready‑to‑paste prompt with Google Nano Banana Pro
Tweak the style/lighting to match your brand.'

Perfect for:
Viral posts, meme edits, chaotic group shots, album-style covers.

Style:
35mm fisheye, heavy distortion, harsh flash, film grain, candid chaotic vibe.

Works with:
Portraits, friends, celebrities, fictional characters, cosplay.

Pro tip:
Add imperfections — off-center framing, blown highlights, slight blur — messy = more real.

Question of the Week:

Hit reply and let me know - I read every response!