- OpenTools' Newsletter
- Posts
- 📱ChatGPT Gets App Store
📱ChatGPT Gets App Store
PLUS: Anthropic Opens Agent Skills | ByteDance's 10x AI Growth
Reading time: 5 minutes
🗞️In this edition
OpenAI launches app directory for ChatGPT
Sponsored: LLM Portal - Test, compare, and build with every major LLM from one command center.
Anthropic opens agent skills to competitors
ByteDance's Doubao usage explodes 10x this year
In other AI news –
Trump media links up with Fusion Energy to fuel AI growth
Meta AI chief Yann Lecun eyes 3.5 billion valuation for new startup
4 must-try AI tools
Hey there,
OpenAI is turning ChatGPT into a platform, not just a chatbot, by launching an App Directory that lets you order food, manage music, pull files, and run research without ever leaving the interface. Anthropic is going in the opposite direction, open-sourcing its Agent Skills and quietly defining the standard for how AI agents actually work inside enterprises. And over in China, ByteDance is skipping the hype entirely and locking down infrastructure, capturing nearly half of the country’s AI cloud market while most Western companies are still arguing about monetization.
We're committed to keeping this the sharpest AI newsletter in your inbox. No fluff, no hype. Just the moves that'll matter when you look back six months from now.
Let's get into it.
What's happening:
OpenAI just launched an App Directory and opened its SDK for developers to build interactive experiences inside ChatGPT's interface.
The company's also renaming "connectors" to "apps." Chat connectors are now "apps with file search." Deep research connectors are "apps with deep research." Synced connectors are "apps with sync."
New apps include Apple Music (find songs, make playlists, manage libraries) and DoorDash (turn recipes and meal plans into shopping carts without leaving ChatGPT). Spotify in ChatGPT expanded to the UK, Switzerland, and EU.
Apps can access Memory if enabled. For Free, Plus, Go, and Pro users, OpenAI may use app interaction data for training if "improve the model for everyone" is turned on.
OpenAI CEO Sam Altman said last month they're "planning to build the obvious features you would expect for a robust platform." An app store is a massive step toward that.
The monetization strategy? Still unclear. OpenAI says they're "exploring additional monetization options over time, including digital goods."
Why this is important:
OpenAI's building an operating system, not just a chatbot.
The App Directory transforms ChatGPT from a conversational AI into a platform where you order food, manage music, pull files from Google Drive, and conduct deep research all without leaving the interface.
This is the same playbook Apple used with iOS and the App Store. Create the platform, open it to developers, take a cut of transactions. OpenAI's running the same play.
The connector-to-app rebrand matters more than it seems. "Connectors" sound like plumbing. "Apps" sound like products. It's positioning ChatGPT as a destination, not a tool you use alongside other services.
The training data clause is huge. Every interaction with apps, recipes you search, playlists you make, files you pull can feed OpenAI's models unless you opt out. That's behavioral data at scale.
Our personal take on it at OpenTools:
OpenAI's copying Apple's homework and hoping no one notices.
The App Directory is the App Store. The SDK is iOS for AI. They're building a walled garden where ChatGPT becomes the interface layer for the internet.
But here's the problem: they haven't figured out how to make money from it yet. The announcement literally says "we're exploring monetization options" and "will share more as we learn." That's corporate speak for "we don't know."
Compare this to Apple, who took 30% of app revenue from day one. OpenAI's giving away the platform hoping demand creates a business model later. That works when you're venture-backed and can burn cash. It doesn't work when you're already unprofitable at a $500B valuation.
The DoorDash and Apple Music integrations are smart high-frequency use cases that keep people inside ChatGPT. But unless OpenAI takes a transaction fee, they're just driving revenue to other companies while paying computer costs.
The training data play is the real business model. Every app interaction is free training data. Users build playlists, order food, search recipes OpenAI learns behavioral patterns that make future models better. That's worth more than digital goods revenue.
But if they're serious about being a platform, they need to monetize developers, not just users. Revenue share, API fees, premium placement in the directory something. Otherwise, they're just a chatbot with a fancy UI.
From Our Partner:
AI moves fast. If you’re not experimenting with the latest large language models, you’re already behind. LLM Portal gives you a single command center to test, compare, and build with every major LLM.
Access multiple models in one place
Prototype workflows instantly — no code required
Stay ahead as new models launch
Move from idea to implementation without friction
This isn’t just another tool - it’s the infrastructure for your next big breakthrough.
What's happening:
Anthropic just released Agent Skills as an open standard, a bet that sharing its approach to making AI assistants better will cement its position in enterprise software.
Skills are folders with instructions, scripts, and resources that teach AI how to do specific tasks consistently. Instead of writing elaborate prompts every time, you package expertise into reusable modules.
Microsoft already adopted Agent Skills in VS Code and GitHub. Coding agents like Cursor, Goose, and Amp use it too. OpenAI quietly replicated the same architecture in ChatGPT and Codex CLI.
Anthropic launched enterprise management tools and a partner directory with skills from Atlassian, Figma, Canva, Stripe, Notion, and Zapier. Their GitHub repo hit 20,000 stars with tens of thousands of community-created skills.
There's no extra charge. Skills work across Claude.ai, Claude Code, and the API at standard pricing.
Why this is important:
Anthropic's making their IP the industry standard instead of locking it down.
This is the Linux playbook: open-source the tech, own the ecosystem. By releasing Skills as an open spec, Anthropic's betting ecosystem growth beats proprietary control.
It's working. OpenAI copied the architecture with exactly the same file structure, same metadata format, same directory organization. When your competitor replicates your design, you've won the standards war.
Skills solve a core LLM problem: models know everything generally but lack procedural expertise for specialized work. A skill for PowerPoint might include formatting rules and slide structure. The AI loads it only when needed.
The "progressive disclosure" design is smart. Each skill uses dozens of tokens when summarized, full details load only when activated. Organizations can deploy massive skill libraries without crushing context windows.
Our personal take on it at OpenTools:
Anthropic's playing 4D chess while OpenAI's still figuring out how to monetize apps.
Releasing Skills as an open standard seems counterintuitive. Why give away your competitive advantage? But Anthropic understands something critical: in infrastructure, the standard-setter wins even if they don't own the tech.
Red Hat didn't invent Linux but built a $34B company around it. Google open-sourced Android and now controls mobile. Anthropic's running the same play with AI capabilities.
The Microsoft adoption is the proof point. VS Code has 20M+ developers. GitHub has 100M+ users. When Microsoft builds your spec into their tools, you've defined how an industry works.
OpenAI's response quietly copying the architecture validates the strategy. They could've built something different. Instead, they replicated Anthropic's design because Skills already became the de facto standard.
Here's what people miss: Skills + MCP (Model Context Protocol) is Anthropic building the plumbing layer for AI. MCP connects to external data. Skills provide procedural knowledge. Together, they're the API standard for agentic AI.
The Agentic AI Foundation timing isn't coincidental. Anthropic donated MCP to Linux Foundation, co-founded the foundation with OpenAI, and got Google, Microsoft, and AWS as members. They're standardizing infrastructure while competitors scramble.
Two months ago, Skills looked like a dev feature. Today, its infrastructure. That's how you win AI: not by building the best model, but by defining how everyone uses models.
What's happening:
ByteDance's Doubao AI models saw usage explode 10x this year. Daily token consumption hit 50 trillion in December 2025, up from 4 trillion in December 2024.
Tan Dai, president of ByteDance's Volcano Engine (their cloud unit), said they now hold 49.2% of China's AI model-focused public cloud market. That means "for each two tokens consumed in Chinese public cloud services, one was from Volcano Engine."
They have 100+ corporate clients using over 1 trillion tokens total. Major users include Mercedes-Benz, BYD, China Merchants Bank, and education platform Yuanfudao.
ByteDance also revealed that 90%+ of their engineers use Trae, their Cursor-like AI coding tool with 1.6M monthly active users in China. Over 40% of code behind Douyin's local services platform is AI-generated.
They just launched Doubao 1.8, optimized for multimodal agent use and complex instructions, plus new visual-generation systems. Volcano Engine cut prices again tiered plans offer up to 47% savings for enterprise clients.
Why this is important:
ByteDance is quietly dominating China's AI infrastructure market while US companies fight over consumer apps.
50 trillion tokens daily is massive. For context, that's roughly 50 million conversations per day just from Doubao. At that volume, ByteDance is processing more enterprise AI workloads in China than anyone else.
The 49.2% market share in AI-focused public cloud is stunning. AWS, Azure, and Google Cloud combined don't control half the US market. ByteDance captured that in under two years.
The pricing cuts matter. ByteDance's subsidizing AI adoption the same way they subsidized TikTok growth aggressive pricing to lock in market share, worry about profitability later.
Trae with 1.6M monthly activities means ByteDance is training a generation of Chinese developers on their AI tools. That's ecosystem lock-in at the developer layer.
Our personal take on it at OpenTools:
ByteDance is executing the WeChat playbook for AI infrastructure.
They're not chasing consumer chatbot downloads. They're embedding AI into enterprise workflows manufacturing, banking, education, automotive. Once Mercedes-Benz and BYD run production workloads on Doubao, switching costs become prohibitive.
The 10x growth in one year proves China's enterprise AI adoption is moving faster than the West. US companies are still debating ROI. Chinese enterprises are already running 40% AI-generated code in production.
Volcano Engine's 49% market share is ByteDance leveraging their own internal AI usage as social proof. When you tell customers "90% of our engineers use this," adoption becomes inevitable. It's the Stripe playbook: eat your own dog food publicly, then sell the infrastructure.
The Trae numbers are the real story. 1.6M monthly activities makes it China's largest AI coding tool. That's bigger than Cursor in the US. ByteDance is training developers on their stack while OpenAI's still figuring out how to monetize Codex.
The price cuts signal they're still in land-grab mode. When you're offering 47% discounts at scale, you're not optimizing for profit. You're buying market share. The question is what happens when they stop subsidizing and start extracting.
Trump Media to merge with nuclear fusion company that wants to power AI – The parent company of President Donald Trump’s Truth Social media platform announced Thursday it is merging with a fusion power company, an unusual pairing of the Trump name with a futuristic clean energy venture that aims to power the next wave of artificial intelligence.
Meta's Yann LeCun targets $3.5 billion valuation for a new AI startup – The startup plans to build AI systems using so-called world models that can understand the physical world. The systems could support applications including robotics and transport.
👩🏼‍🚒Discover mind-blowing AI tools
Altersnap - A mobile application that allows users to take selfies and create altered photos
LogoPicture AI - An online tool that allows users to create optical illusion art with their logos
FinWise - A financial management tool that helps users manage their profiles and money effectively
Mindie - Allows users to have conversations with anything they encounter on the web, such as YouTube videos, blog posts
We're here to help you navigate AI without the hype.
What are we missing? What do you want to see more (or less) of? Hit reply and let us know. We read every message and respond to all of them.
– The OpenTools Team
How did we like this version? |
Interested in featuring your services with us? Email us at [email protected] |