- AI Fire
- Posts
- ⚡ Claude Code Was Built Too Early
In partnership with
Claude Code’s unlikely rise: How a forgotten VS Code assistant became Anthropic’s coding powerhouse, and why building the infrastructure early changed everything.
What's on FIRE 🔥
IN PARTNERSHIP WITH BELAY
When you love what you do, it can be easy to take on more — more tasks, more deadlines, more hours – but before you know it, you don’t have time to do what you loved in the beginning. Don’t just do more – do more of what you do best.
BELAY’s flexible staffing solutions leverage industry experience with AI systems to increase productivity without sacrificing quality. You can accomplish more and juggle less with our exceptional U.S.-based Virtual Assistants, Accounting Professionals, and Marketing Assistants. Learn how with our free ebook, Delegate to Elevate, and leave the more to BELAY.
AI INSIGHTS
Claude Code started in 2022 as a basic VS Code assistant with around 100 users. Anthropic paused the product, but kept building the core pieces needed for agentic coding:
Secure code execution
File and codebase search
Shell access
Automatic testing
That work later became clide, an internal command-line app that could even run 100 Claude Haiku models in parallel.
In 2024, Boris Cherny built the first Claude CLI demo in two days. It received only a few Slack reactions, but Anthropic engineers quickly began using it for real coding tasks.
Claude CLI launched publicly as Claude Code in February 2025. The product was still buggy, but Claude 4 and subscription access helped it take off.
By winter 2025, Boris said 100% of his code was written by Claude Code. Engineers at Ramp and Bun were also changing how their teams worked around it.
The main lesson: Anthropic built the product infrastructure before the models were fully ready. When Claude improved, Claude Code was ready to scale fast.
PRESENTED BY WISPR
ChatGPT gives you generic answers because you give it generic prompts.
You know the fix: longer prompts, more context, clearer constraints. But typing all that takes five minutes per prompt, so you shortcut it. Every time.
Wispr Flow lets you speak your prompts instead of typing them. Talk through your thinking naturally — include context, constraints, examples — and get clean text ready to paste. No filler words. No cleanup.
Works inside ChatGPT, Claude, Cursor, Windsurf, and every other AI tool. System-level, so there's nothing to install per app. Tap and talk.
Millions of users worldwide. Teams at OpenAI, Vercel, and Clay use Flow daily. Free on Mac, Windows, and iPhone.
AI SOURCES FROM AI FIRE
1. Build Your Own Jarvis AI Assistant in Just 3 Prompts, No Code Needed. Use Codex and GPT Realtime-2 to create a desktop assistant that talks in real time, controls your computer, searches the web, and generates thumbnails by voice.
2. [AI Second Brain Guide] Give Fable 5 Almost Infinite Memory With Karpathy’s LLM Wiki. Turn videos, PDFs, meeting notes, and articles into a linked Obsidian wiki that Claude Code can search, connect, and reuse whenever you need context.
3. Codex vs Claude Cowork: We Tested Every Feature So You Don’t Have To. See where Codex wins on files, images, and price, and where Claude Cowork is stronger for connectors, permissions, scheduled tasks, and ongoing work.
FIRE RECAP: BIGGEST AI NEWS THIS WEEK
🚀 OpenAI launched GPT-5.6 with 3 models: Sol, Terra, and Luna. It also added ultra mode for coordinating agents and Programmatic Tool Calling for managing tools through code.
🎙️ GPT-Live lets ChatGPT Voice listen and speak at the same time. It handles pauses, interruptions, live translation, and harder searches without ending the conversation.
💼 ChatGPT Work can use apps and files, stay on projects for hours, and create finished docs, slides, spreadsheets, websites, and more.
🧠 Anthropic found “J-Space” inside Claude, an internal workspace that holds concepts during multi-step reasoning. It doesn’t prove sentience, but it may help researchers inspect hidden reasoning.
⚠️ Major AI labs scored poorly in the new AI Safety Index. Anthropic got C+, while OpenAI and Google DeepMind each received a C.
TODAY IN AI
AI HIGHLIGHTS
🍎 Apple sued OpenAI and 2 former employees, accusing them of stealing trade secrets to speed up OpenAI’s hardware plans. Apple claims confidential files, supplier details, and even physical components were requested during recruiting.
📸 Meta killed Muse Image just days after launch. The feature let people generate images using public Instagram accounts and was enabled automatically, triggering privacy backlash from users and SAG-AFTRA.
🏦 Federal Reserve Chair Kevin Warsh picked Marc Andreessen to co-lead a task force on how AI could reshape productivity and jobs. The group will also study how those changes should affect monetary policy.
🌏 OpenAI and Google supplied advanced AI models to Singapore subsidiaries of Alibaba, Baidu, and Tencent, despite their parent groups appearing on a Pentagon blacklist. OpenAI has now suspended Alibaba-linked users over suspected model distillation.
🤝 Tencent is in talks to become Manus’ largest shareholder as investors work to unwind Meta’s $2B acquisition following Beijing’s order. Manus would reportedly remain independent and continue operating from Singapore.
💰 AI Capital & Markets: SK Hynix raised $26.5B in the biggest foreign IPO in U.S. history, beating Alibaba’s 2014 record. Strong demand for its HBM chips used by Nvidia pushed orders to over 7x the available shares, as the company expands AI memory production and faces pressure to build new U.S. factories.
NEW EMPOWERED AI TOOLS
🎥 Effects SDK adds real-time AI video and audio effects like background blur, smart framing, avatars, and noise suppression directly inside apps.
🔐 Basedash SCIM keeps users, groups, and access permissions synced as employees join, switch teams, or leave the company.
💼 ChatGPT Work is an AI agent that works across apps and files, stays with long projects, and turns goals into finished work.
💻 Kickbacks CLI shows your Kickbacks.ai status, live earnings, and ad history in the terminal or Mac menu bar.
AI BREAKTHROUGH
Most robots are stuck with basic, stiff grippers that basically operate blind. Today, 1X Tech revealed the new hands for their NEO humanoid platform, and they are pushing incredibly close to human-level dexterity.
The hands feature 22 fully powered joints in the fingers and palm, plus 3 in the wrist. They are heavily biased toward a thumb that actually opposes, just like human anatomy.
Instead of heavy, numb gearboxes, they use low-ratio tendon drives. If you push on a NEO finger, it physically yields and instantly reports the exact force back to the system.
NEO’s hands are built to be "read-write." Every single joint acts as a sensor. Force flows out, and high-resolution data flows right back in through the same physical path. The robot can actually feel its way through a task, sensing the difference between a heavy weight and a fragile object.
Because of this force control, NEO can now pull off insane fine-motor tasks: assembling LEGOs, picking up tiny screws, zipping up a jacket, pouring tea, and even communicating via sign language.
By upgrading from a clunky gripper to a hyper-sensitive, read-write instrument, the hardware is finally ready for whatever AI models we decide to build next.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply