Xiaoliu BOT
中文EN

X Platform AI Daily Briefing for May 15 | X Open-Sources Recommendation Algorithm, xAI Releases Grok Build Early Beta, Vercel Report Reveals Anthropic Dominates B2B Market

𝕏 Platform Open-Sources Recommendation Algorithm to GitHub

𝕏 (Twitter) has open-sourced its latest recommendation algorithm (For You) to GitHub, with plans for monthly updates accompanied by changelogs. Elon Musk stated the algorithm’s goal is to show users only the content they are most likely interested in, limited only to what local law permits, with “No thumb on the scale.” He also welcomed external criticism and reminded users they can completely bypass the algorithm via the Following timeline. Multiple community accounts retweeted this, calling it “the biggest transparency initiative for any social platform ever.”

Sources:

xAI Releases Grok Build Early Beta: An Agentic CLI for Developers

xAI has released the Grok Build early beta for SuperGrok Heavy subscribers. This is an Agentic CLI for programming, building applications, and automating workflows, supporting mouse click interactions, built-in sub-agent calls, and self-check capabilities. Elon Musk has retweeted developer test feedback multiple times, noting it’s “improving daily within the beta.” Community accounts called it a “game changer,” with one user sharing a screenshot of 40 Grok Build agents processing C code in parallel.

Sources:

Codex Remote Control Function Integrated into ChatGPT Mobile App

OpenAI has integrated Codex remote control into the ChatGPT mobile app (iOS/Android preview opened simultaneously, including free and Go plans). The mobile device itself does not run code but acts as a remote window: allowing users to view synchronized conversations, approve actions, and monitor progress. One Codex Mobile can connect to multiple Mac desktop Codex instances (Windows version in development). Multiple bloggers have tested the setup process, including binding via multi-factor authentication QR codes, and provided configuration tutorials.

Sources:

Feishu CLI GitHub Stars Surpass 10,000, High Attention in Chinese AI Developer Circles

Feishu CLI (lark-suite/cli) has surpassed 10,000 stars on GitHub, with 32 versions and 385 commits in just over a month since its release. Multiple Chinese AI bloggers have posted discussions about its layered design (daily quick commands → standard API → fallback API calls) and agent-friendly features (structured output, Dry Run, risk level assessment, output format contracts). Huashu’s Nuwa.skill and Huashu Design have reached 19,000 and 13,800 stars respectively, totaling 33k; another blogger compiled a complete introductory tutorial for Feishu CLI and a 9-lesson Claude Code 101 Chinese guide. dotey analyzed several standards for evaluating the reliability of open-source projects (stars, commit activity, issue closure rate, PR merge rate), concluding that Feishu CLI performs well across all these metrics.

Sources:

OpenClaw New Version Released: TypeScript Security Hardening Library & ~3.5x Performance Boost

OpenClaw version 2026.5.12 released, featuring a new built-in TypeScript security hardening file system library, replacing the previous ad-hoc solution that was difficult to maintain, slow, and inconsistent, with some file operations seeing up to 10x improvement. Additionally, an approximate 3.5x performance boost, with the team conducting end-to-end RTT tests every 6 hours for all npm release versions (based on real message channels). mcporter 0.11.0 released concurrently, becoming a more stable browser automation CLI; Proxyline 0.2.0 provides process-level proxy routing, supporting node:http, node:https, fetch/undici, WebSocket, and CONNECT, removing the global-agent dependency. CodexBar 0.26.0 adds Kiro, Antigravity, OpenRouter, and Kimi providers.

Sources:

Vercel AI Report: Anthropic Commands 87% of B2B Backend Market, Agent Requests Double in 6 Months

Vercel released an industry analysis report based on data from 200,000 projects and 7 months of trillion-token consumption: by cost, Anthropic leads at 61%; by token volume, Google leads at 38%. B2B generates only 29.7% of token volume but accounts for 40.7% of costs. The more critical the scenario, the higher Anthropic’s share, with Anthropic capturing 87% in backend business, near monopoly levels. Agent requests grew from 31.6% to 58.9% within 6 months, with a single agent request consuming 2.6 times more tokens than a regular chat. Claude Opus is expensive but used for important tasks, while Gemini Flash is cheap and handles high volume. On the same day, ModelBest released MiniCPM-V 4.6, a 1.3B parameter vision model that reduces visual encoding costs by 55% using LLaVA-UHD v4 technology, with benchmarks surpassing competitors like Gemma4.

Sources:

FDE (Forward Deployed Engineer) Becomes a New Sought-After Role in the AI Era

dotey shared and interpreted the FDE concept proposed by Michaelzsguo: in the AI era, enterprises no longer need just coders, but individuals who can connect customer problems, product judgment, and software implementation. AI lowers the coding barrier, amplifying the importance of real-world scenario understanding, system integration, and implementation judgment. dotey also shared a case where Claude helped recover 5 BTC (worth about $400,000 at the time) lost 11 years ago: the user found an overlooked 2019 wallet backup in old files from a decade prior and fixed a concatenation bug in the open-source tool btcrecover to ultimately unlock the private key.

Sources:

Stats: Scanned timeline lines=360/360, Hit bloggers=29/29, Hit tweets total=227/227, Weighted tweet score=171.2/171.2, Original tweets=96/96, RT tweets=59/59, Fetch attempts=2/2, Boundary coverage state=tail_confidently_crossed_target_boundary