Here’s what’s shaping AI right now - without the noise:

HeyGen's Video Podcast converts documents, links, or raw ideas into a produced video show hosted by two AI avatars, with edits and multiple camera angles already baked in.

The workflow is four steps: upload photos of the hosts, add a topic file, preview the auto-generated script, publish. Watch how it works.

For content creators, marketers, and educators who want video output without a production team, this closes a gap that previously required either budget or technical skill. The script generation is the piece worth watching. Auto-generating a two-person dialogue from a raw document is a harder problem than it sounds, and getting it right at this level is new.

It sits inside HeyGen's growing library of AI content tools, which keeps expanding.

The bigger signal: 👉 Video content production just got a lot more accessible. The barrier was never the camera. It was the script, the edit, and the hosting. HeyGen is removing all three.

ChatGPT is insanely powerful.

But most people waste 90% of its potential by using it like Google.

These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.

Sign up for Superhuman AI and get:

Claim your free prompts

OpenAI published new pricing for the GPT-5.6 family this week. Luna, already the cheapest tier, dropped 80%.

The new numbers:

The efficiency gains came from Sol rewriting its own GPU code, cutting serving costs 20% and making the 5.6 models 15% more efficient overall. OpenAI published the technical research here.

Sam Altman framed it around competing with Chinese and open-source models on price. That's the real story. Luna at $0.20 per million input tokens is a serious number. Frontier-level intelligence at those prices changes what's viable to build.

The bigger signal: 👉 The cost of running capable AI just dropped hard again. Anything you ruled out as too expensive to automate six months ago is worth reconsidering now.

Workers with AI skills earn roughly 50% more than peers in comparable roles. But "AI skills" covers everything from writing a decent prompt to training models on your own hardware. Those are completely different things.

Here's a practical ranking from easiest to hardest, and why each one matters:

1. Prompt engineering

Writing instructions that get reliably good output. The gap between "write me a marketing email" and a prompt that specifies audience, tone, structure, and examples is the gap between a tool that occasionally helps and one that actually works. Start here.

2. AI tool fluency

Knowing which tool to reach for and why. Claude for long-context reasoning, ChatGPT for fast structured output, Midjourney for images. Fluency means you're not re-learning the same interface every time.

3. Workflow automation

Connecting AI to the rest of your tools. Zapier, Make, n8n. The ability to chain a trigger to an AI step to an output is where most of the practical time savings live.

4. Working with APIs

Calling AI models programmatically. You don't need to be a developer to do this. Knowing how to send a request and handle a response unlocks everything that can't be done through a chat interface.

5. RAG (Retrieval-Augmented Generation)

Making AI work with your own data. Building a system that retrieves relevant chunks from a knowledge base before generating a response. This is how companies build internal AI tools that actually know their domain.

6. Fine-tuning

Adapting a model to a specific task using your own examples. More controlled than prompting, more accessible than training from scratch. Valuable for anything where generic model behavior isn't precise enough.

7. AI agent design

Building systems where AI can take actions, not just generate text. Tool use, memory, multi-step reasoning. This is where Claude Code, Cowork, and agentic frameworks live.

8. Evaluations

Measuring whether your AI system actually works. Most teams skip this and wonder why their outputs are inconsistent. Knowing how to design evals is what separates people who ship reliable AI products from people who demo well.

9. Vector databases

Storing and retrieving information by meaning rather than keywords. The infrastructure layer under most RAG systems. Pinecone, Weaviate, pgvector.

10. LLM ops

Running AI systems in production. Monitoring, logging, latency, cost tracking, model versioning. The difference between a demo and something that works at 3am without you watching it.

11. Multimodal AI

Working with models that handle text, images, audio, and video together. GPT-5.6 Sol, Claude Fable 5, Gemini. Understanding what each modality is good for and how to combine them.

12. Model training fundamentals

Understanding how models learn. You probably won't train a frontier model. But knowing what loss functions, data quality, and compute requirements actually mean makes you a much better user of the ones that already exist.

Where to start:

Pick the skill one level above where you are now. If you're already good at prompting, learn the API. If you know the API, build something with RAG. Each skill compounds. The people earning the wage premium aren't necessarily the most technical. They're the ones who kept moving up the stack.

You're spending 40 hours a week writing code that AI could do in 10.

While you're grinding through pull requests, 200k+ engineers at OpenAI, Google & Meta are using AI to ship faster.

How?

The Code newsletter teaches them exactly which AI tools to use and how to use them.

Here's what you get:

Sign up and get access to the Ultimate Claude code guide to ship 5X faster.

Join 200K+ engineers

A few tools quietly worth exploring:

Luna is now $0.20 per million tokens.
HeyGen just automated the video production workflow.
And the AI skill gap is only widening for people who aren't moving up the stack.

The tools keep getting cheaper and more capable at the same time.

See you next issue.