🚨Micron predicts memory shortages will persist "beyond calendar 2026" as CEO says industry can only meet "half to two-thirds" of demand from key customers, revealing that AI's infrastructure buildout has created structural supply constraints where securing memory capacity becomes a competitive moat worth more than the chips themselves.

AI keeps coming up at work, but you still don't get it?

That's exactly why 1M+ professionals working at Google, Meta, and OpenAI read Superhuman AI daily.

Here's what you get:

Join 1M+ pros

Developers have been playing prompt roulette for two years tweaking instructions, hoping for consistency, and copying long system prompts between projects like digital incantations.

But Anthropic's Claude Skills, launched in November 2025, are changing the game by doing what should have been obvious from the start: packaging specialized knowledge into loadable folders that Claude accesses only when needed.

Claude Skills are folders containing instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. When working on tasks, Claude scans available skills to find relevant matches, loading only the minimal information and files needed—keeping Claude fast while accessing specialized expertise.

Founded by former OpenAI researchers and launched by Anthropic with 1-million token context windows in Sonnet 4 and 4.5, Skills represent a fundamental shift in how AI capabilities scale. Instead of cramming everything into a monolithic system prompt or building separate custom agents for each use case, Skills let you teach Claude like you'd onboard a specialist: give it a folder of best practices, examples, and executable tools, and it'll load them only when the task demands it.

How it works:

Skills are simple to create—just a folder with a SKILL.md file containing YAML frontmatter and instructions. The markdown content contains the instructions, examples, and guidelines that Claude will follow.

Here's the basic structure:

---

name: my-skill-name

description: A clear description of what this skill does and when to use it

---

# Skill Name

[Instructions for Claude to follow]

## Examples

- Example usage

## Guidelines

- Guideline 1

Progressive disclosure is the core design principle: like a well-organized manual with a table of contents, then specific chapters, and finally a detailed appendix, skills let Claude load information only as needed. Agents with a filesystem and code execution tools don't need to read the entirety of a skill into their context window when working on a particular task.

You can build your own skills and use them across Claude apps, Claude Code, and the API. You'll even see skills in Claude's chain of thought as it works. Creating skills is simple, the "skill-creator" skill provides interactive guidance: Claude asks about your workflow, generates the folder structure, formats the SKILL.md file, and bundles the resources you need.

What makes this different:

The requirement for code execution environments is the biggest difference between skills and previous attempts at expanding LLM abilities like MCP and ChatGPT Plugins. It's a significant dependency, but it unlocks bewildering amounts of new capability. The fact that skills are so powerful and simple to create is yet another argument in favor of making safe coding environments available to LLMs.

Model Context Protocol (MCP) has attracted enormous buzz, but its limitations have emerged over time. The most significant is token usage: GitHub's official MCP consumes tens of thousands of tokens of context on its own, and once you've added a few more there's precious little space left for the LLM to do useful work.

Skills solve this by loading only what's needed, when it's needed. Claude operates in a virtual machine with filesystem access, allowing Skills to exist as directories containing instructions, executable code, and reference materials, organized like an onboarding guide you'd create for a new team member.

Skills are composable: they stack together, and Claude automatically identifies which skills are needed and coordinates their use. This means you can combine a brand-voice skill with a document-formatting skill with a legal-compliance skill, and Claude orchestrates all three without manual coordination.

Why this matters now:

The best AI companies are obsessed with prompt engineering. There's a big difference between personal use and products—the best AI products treat the system prompt as a crucial component.

Prompt engineering frameworks turn LLMs from experiments into reliable business partners. Research shows AI tools like ChatGPT can boost workforce productivity by 14%, with some companies reporting gains of up to 400%. Compare that to the invisible tax of "workslop," which HBR estimates can cost organizations millions of dollars every year.

But treating every specialized task as a massive prompt engineering challenge doesn't scale. Skills teach Claude how to complete specific tasks in a repeatable way, whether that's creating documents with your company's brand guidelines, analyzing data using your organization's specific workflows, or automating personal tasks.

Custom instructions apply broadly to all your conversations. Skills are task-specific and only load when relevant, making them better for specialized workflows.

The traction:

Skills are available to Pro, Max, Team and Enterprise users. Anthropic provides pre-built skills for professional Excel spreadsheets with formulas, PowerPoint presentations, Word documents, and fillable PDFs.

Many skills in Anthropic's GitHub repository are open source (Apache 2.0), including the document creation and editing skills that power Claude's document capabilities under the hood. These are source-available for developers as reference for complex skills actively used in production AI applications.

Skills extend Claude Code with team expertise and workflows. Install skills via plugins from the anthropics/skills marketplace, and Claude loads them automatically when relevant.

What's next

Claude Sonnet 4.5 is the best model in the world for building agents. These features unlock new possibilities for long-running agents—processing entire codebases, analyzing hundreds of documents, or maintaining extensive tool interaction histories. Context management ensures agents can leverage expanded capacity efficiently while handling workflows that extend beyond any fixed limit.

Expect Google to keep undercutting on price while matching on capability. They don't need to be better. They just need to be good enough and everywhere.The Skills architecture enables what Anthropic calls "progressive disclosure at scale"—the amount of context bundled into a skill is effectively unbounded because Claude doesn't load everything at once.

Agent Skills can now be added to Messages API requests and the new /v1/skills endpoint gives developers programmatic control over custom skill versioning and management.

BTW: Skills operate fundamentally differently than normal tools. Instead of executing discrete actions and returning results, skills inject comprehensive instruction sets that modify how Claude reasons about and approaches tasks.

Have an ad concept ready but don't want to deal with expensive shoots or stock footage? ScriptKit lets you generate, curate, and edit AI ads in one platform.

What ScriptKit gives you

Give ScriptKit a shot — go from concept to finished AI ad without wrangling teams or gear.

The Recap: Trump Media & Technology Group announced a $6 billion all-stock merger with Google-backed fusion power company TAE Technologies, with shareholders of each company owning approximately 50% of the combined entity at close. The deal, expected to finalize mid-2026, will see Trump Media provide up to $200 million at signing and another $100 million upon regulatory filing, while the combined company plans to site and begin construction of the world's first utility-scale fusion power plant in 2026. This marks a dramatic pivot for Trump's money-losing social media company—down nearly 70% this year—into the energy infrastructure race driven by AI's insatiable power demands.

Unpacked:

Bottom line: This merger crystallizes the shift from AI being a software/compute story to being fundamentally an infrastructure and energy story. Trump Media's pivot from social media to fusion power isn't eccentric—it's recognition that whoever solves AI's energy problem captures more value than whoever builds better models.

Micron confirms DRAM shortages extending beyond 2026 despite new Idaho and New York fabs coming online in 2027-2030, as memory manufacturers prioritize HBM production over consumer RAM—proving that hardware supply chains have become winner-take-all battles where securing multi-year contracts with chip makers now determines who gets to build AI infrastructure at all.

YouTube withdrew music streaming data from Billboard charts effective January 16, 2026, after the publication refused to count ad-supported and paid streams equally (maintaining a 2.5:1 weighting), demonstrating that platform holders will abandon industry-standard metrics entirely when measurement methodologies don't favor their business model—algorithms over transparency.

YouTube secured exclusive global rights to stream the Oscars from 2029-2033, outbidding ABC's high eight-figure offer and ending the network's hosting streak since 1976, as the ceremony moves to free streaming for 2 billion global viewers—proving YouTube's dominance in live video is making it the default distribution layer for cultural tentpole events that once defined broadcast television's relevance.

Meta paused its Horizon OS licensing program for third-party headsets from Asus and Lenovo just 20 months after announcement, citing need to "focus on building world-class first-party hardware"—signaling that vertical integration won the VR platform war before Android XR and open ecosystems even got their chance, as Meta prioritizes control over ecosystem expansion while cutting metaverse budgets by up to 30%.

Apple opened its App Store to competition in Japan under regulatory pressure, demonstrating that gatekeeping remains economically defensible under existing terms and selective regional compliance creates a fragmenting global internet where companies maintain monopoly power everywhere except jurisdictions with enforcement teeth.

Instacart's AI-powered dynamic pricing tool attracted FTC scrutiny, exposing how algorithmic price discrimination has become a regulatory flashpoint that will force enterprises to choose between optimization and legal risk—making pricing audit trails and transparency tooling an emerging product category for companies deploying AI in consumer-facing applications.

Adobe faces proposed class-action lawsuit for allegedly misusing authors' work in AI training, as copyright litigation shifts from theoretical to existential for model makers—the outcome will determine whether training data becomes a massive licensing business or if the entire foundation model approach needs fundamental restructuring around provably licensed datasets.

AI-assisted development produces code 1.7x faster but ships 1.7x more bugs, revealing that velocity gains come at the cost of technical debt accumulation—companies trading short-term shipping speed for long-term system reliability are building a hidden crisis that will compound when critical production systems fail at scale from code nobody fully understands.

UK watchdog claims creating encrypted messaging apps like Signal could constitute "hostile activity," signaling that governments are redefining privacy tools as security threats rather than civil liberties—the next regulatory battleground won't be data localization, it will be whether end-to-end encryption itself is legal to build and distribute.

Best of AI™ Team

Was this email forwarded to you? Sign up here.