01-18-Daily AI News Daily
Today’s Summary
Brex leverages AI to get 500 people's worth of work from 300 employees, breaks $500M in annual revenue—the secret? Let employees freely choose their AI tools instead of forcing one on them.
Claude Code is now auto-posting to X, Codex writes more stable code, developers are starting to "pick sides" by task type, and the Agent skills marketplace is about to explode.
Volcano Engine's 8.9 yuan programming package just flipped the table—China's AI tool price war is entering white-hot territory.⚡ Quick Navigation
- 📰 Today’s AI News - Latest updates at a glance
💡 Tip: Want to experience the latest AI models mentioned in this article (Claude 4.5, GPT, Gemini 3 Pro) right away? No account? Head over to Aivora to grab one—ready in a minute, hassle-free support.
Today’s AI News
👀 One-Liner
Claude Code and Codex are locked in a silent battle over “who gets more work done,” while Brex is squeezing 500 people’s worth of output from a 300-person team using AI.
🔑 3 Key Hashtags
#AgentBoom #AIToolWars #EfficiencyRevolution
🔥 Top 10 Headlines
1. Brex’s AI Transformation: From 20% Layoffs to $500M Annual Revenue
Used to take 3 days to close the books monthly—now it’s done in half a day. Fintech company Brex was laying off 20% of staff in early 2024, but a year later pulled off Silicon Valley’s most stunning comeback powered by AI: same 300 engineers, but customers jumped from 30K to 40K, and annual revenue broke through $500M. Their secret sauce? Don’t force employees to use one specific AI tool—let them pick (ChatGPT, Claude, Gemini, whatever). They even built an Agent network so ops teams can tweak prompts themselves, and let AI supervise AI for audits. Core insight: AI isn’t a layoff tool, it’s a platform that makes every person more powerful. Domestic companies should study this “scale with AI, not headcount” playbook.

2. Claude Code Auto-Posts to X: Agent Skills Level Up
Write an article then manually post it to X? That’s so last season. Developer @JimLiu, building on @wshuyi’s Playwright MCP, created a Claude Code skill that lets Agents automatically post Markdown articles to X (Twitter), with support for long-form content and auto image uploads, plus preview-before-posting. The key move: uses a real Chrome browser to bypass anti-scraping detection, so no account bans. Installation is one command: npx add-skill jimliu/baoyu-skills, then just tell Claude “post this article to X.” This is how Agents should work: handling the repetitive grunt work so humans don’t have to.
3. Higgsfield AI: Raised $130M on AI Effects, Valued at $1.3B
A wrapper app just raised that much money? Higgsfield AI is an AI effects app that drops viral effects almost every month (remember the one that makes static photos dance?). They just closed an additional $80M Series A, bringing total funding to over $130M with a valuation hitting $1.3B. Growth is absolutely bonkers—pure viral momentum on TikTok and Instagram. Sure, the tech might not have much moat, but traffic is king, and this move is textbook “lock in users first, worry about defensibility later.”
4. Zhipu Open-Sources SCAIL: Bring Static Illustrations to Life with 3D Skeleton Rigging
Used to need multiple generations to get a usable animation—now just control it with 3D bones. A Tsinghua team partnered with Zhipu GLM to open-source SCAIL, which transforms static illustrations or characters into cinematic-quality videos, supporting complex motion transfer and multi-character interactions. Core tech: 3D cylindrical skeleton representation + full-context injection mechanism, boosting limb structure integrity by 47% in complex occlusion scenarios. 2.67M plays on international platforms, French artist @NebSH used it to make doodles dance, and B站 creators are pulling off “breaking the dimensional barrier” virtual fight scenes. Redefining 3D-to-2D and 2D-to-3D—anime creators are losing their minds.
5. Black Forest Labs Releases FLUX.2 [klein]: Fastest Image Generation Model
Speed that’ll make your head spin, and you barely need to think about prompts. Black Forest Labs just dropped FLUX.2 [klein] (German for “small”), claiming it’s their fastest image model yet, with preset prompt combinations so you just click a few times and get high-quality images. The demo video’s UX is insanely smooth—zero need to write complex prompts yourself. Perfect for people with tons of ideas but zero prompt-writing skills. Speed and ease-of-use are maxed out.
6. Codex vs Claude Code: Developers’ “Pick Your Side” Moment
Use Codex for code, Claude Code for everything else. Baoyu’s real-world take: Codex CLI is clunky, but the model writes rock-solid code with huge token limits; Claude Code is smooth, but code quality isn’t as good, though it crushes general tasks. OpenAI’s Codex team needs to reflect—they’ve wasted such a good model on flashy but useless updates. This comparison hits hard. Developers now have to context-switch based on task type. There’s no “one tool to rule them all.”
7. SWE-rebench December Leaderboard: Claude Opus 4.5 Tops, Flash Wins on Value
Flash model outperforms Pro? Latest SWE-rebench rankings show Claude Opus 4.5 leading with 63.3% task resolution, OpenAI’s gpt-5.2-2025-12-11-xhigh close behind at 61.5%, but the real winner is Gemini 3 Flash Preview in third place: 60.0% resolution rate + just $0.29 per task. On the open-source side, Zhipu GLM-4.7 jumped from 40% to 51.3%, with DeepSeek-V3.2 at 48.5% and only $0.25 per task. The era of value is here—it’s not about the strongest model, it’s about the best bang for your buck.
8. Volcano Engine Programming Package: 8.9 Yuan for Doubao/GLM/DeepSeek/Kimi
A programming package even cheaper than GLM just dropped. Volcano Engine’s Lite Plan slashed from 40 yuan to 9.9 yuan monthly, Pro Plan from 200 yuan to 49.9 yuan, supporting model switching between Doubao, GLM, DeepSeek, Kimi, and more, plus compatibility with Claude Code, Cursor, and other tools. Pro users get 5x usage limits, perfect for heavy-duty coding needs. This pricing is putting serious pressure on competitors.
9. Quarkdown: Give Markdown Typesetting and Programming Superpowers
The sweet spot between Markdown and LaTeX—write papers, make presentations, handle it all. Quarkdown keeps Markdown’s readability while adding programming-language-style logic control and typesetting features (variables, conditionals, loops, functions). Compiles to HTML, PDF, or slideshows, with support for page layouts, headers/footers, page numbers. Perfect for people who love Markdown’s simplicity but hit its typesetting ceiling. Write books, papers, and presentations the way you’d write code.
10. Slidev: Write Presentations in Markdown—The Programmer’s Secret Weapon
Still dragging components around to make slides? That’s so 2020. Slidev is the programmer’s presentation tool, built on Vue 3 and Vite—just write Markdown and get beautiful slides. Built-in Shiki code highlighting, live coding support (edit code during your talk and watchers see it run instantly), plus embedded interactive charts, real-time dashboards, full web apps. Presenter mode includes notes, timers, next-slide preview, even remote control via phone. The go-to tool for tech talks.
📌 Worth Watching
[Products]
- Eigent: Open-Source Collaboration Desktop, Unleash Productivity - An open-source collaboration tool built for team workflows
- Puck: React Visual Editor with AI Superpowers - Frontend devs rejoice—visual editing meets AI
[Open Source]
- Superpowers: Agent Skills Framework and Software Development Methodology - A solid Agent skills framework worth studying
- AionUi: Free Local Open-Source Collaboration Tool - Supports Gemini CLI, Claude Code, Codex, and more
- asyncwf: Parallel Agent Tool for Claude Code - Parallelize multi-Agent workflows via Node.js API with shared global knowledge base
[Research]
- LangExtract: Extract Structured Data from Unstructured Text with LLMs - Google’s offering with precise source attribution and interactive visualization
- Paper2Any: Convert Papers into Research Diagrams and Presentations - Transform papers/text/topics into editable research diagrams, tech roadmaps, and presentation slides
😄 AI Fun
User Boosts Finance Efficiency 10x with Claude Code
Used to spend 3 days monthly on accounting—now it’s half a day. User @Roland built a financial automation framework with Claude Code: invoice entry dropped from 2 hours to 2 minutes, bank reconciliation from half a day to 5 minutes. The secret: nail down 4 things—deep requirement communication with AI, let AI handle prompts and skills, test subagents, build security guardrails. Game-changer for small business accounting. No more all-nighters reconciling accounts.
🔮 AI Trend Predictions
Claude Code vs Codex “Tool War” Will Intensify
- Timeline: Q1 2025
- Confidence: 75%
- Rationale: Today’s Codex vs Claude Code comparison + developer consensus forming around “Codex for code, Claude Code for everything else”—both companies will optimize their weak spots
Agent Skills Marketplace Will Explode
- Timeline: Q1 2025
- Confidence: 80%
- Rationale: Today’s Claude Code auto-post-to-X skill + growing number of developers open-sourcing Agent skills, lowering the barrier to entry—expect a “Chrome Web Store for Agent skills” to emerge
Domestic AI Programming Tool Price War Enters White-Hot Phase
- Timeline: Q1 2025
- Confidence: 85%
- Rationale: Today’s Volcano Engine 8.9 yuan package + domestic models (Zhipu, DeepSeek, Kimi) catching up to international performance—price wars are inevitable
❓ Related Questions
How to Experience Claude Code and Codex?
Claude Code and Codex are the hottest AI coding tools right now, but both require paid accounts. Domestic users might face payment friction or registration restrictions.
Solution: Visit Aivora to grab ready-made accounts—instant delivery, worry-free support.