EAIDaily – May 26, 2026

AI Coding & Embodied Intelligence Daily Briefing Curated developments at the intersection of AI-assisted software engineering and physically-grounded intelligent systems.

🖥️ AI Coding

1. OpenAI Codex Gains Mac Lock-Screen Control — Autonomous Coding Agents Cross a Threshold

What happened: OpenAI pushed a significant update to its Codex AI coding agent that allows the tool to continue operating even when a Mac is locked — meaning the agent can now run background coding, build, and test tasks without requiring the user’s machine to remain unlocked and supervised.

Why it matters: This is a meaningful step toward truly asynchronous AI coding workflows. Until now, AI coding agents were effectively tethered to an active user session — you had to keep your laptop awake and unlocked for the agent to keep working. By crossing into lock-screen autonomy, Codex moves closer to the “lights-out developer” paradigm, where an AI agent can run overnight builds, test suites, and refactoring jobs without human babysitting. This also raises fresh security questions: an AI agent with deep system access that operates on a locked device looks very much like the exact class of process that endpoint security tools are designed to flag.

Context: The update lands as the AI coding tool market is concentrated around three poles — OpenAI’s Codex, Anthropic’s Claude Code, and Cursor — with each racing to differentiate on autonomy, context window size, and ecosystem integration. Codex’s Mac lock-screen capability is a concrete competitive wedge.

2. DeepSeek V4-Pro API Permanently Cut by 75% — Model Pricing Aftershocks Continue

What happened: On May 22, DeepSeek officially updated its API pricing page: after the time-limited 75% discount campaign ends on May 31, the DeepSeek-V4-Pro API price will be permanently set at 1/4 of the original list price (input: $0.435/1M tokens; output: $0.87/1M tokens). The company also adjusted concurrent request limits (V4 Flash: 500; V4-Pro: 2,500).

Why it matters: DeepSeek has repeatedly used aggressive pricing to force the industry to rethink what AI inference should cost. A permanent 75% cut to what is already one of the strongest open-weight reasoning models sends a clear signal: the “API margin” era that benefited early movers like OpenAI and Anthropic is under structural pressure. For AI coding in particular — where token volume is high because agents read entire codebases — DeepSeek’s pricing makes it an increasingly serious option for cost-sensitive teams and for on-prem / self-hosted coding agent deployments. The fact that the cut is permanent (not a promotional discount) also suggests DeepSeek’s inference cost structure has durably improved.

Context: DeepSeek V4-Pro is competitive with GPT-5.5 and Claude Opus 4.7 on multiple coding benchmarks; the permanent repricing makes it arguably the best price-performance frontier model for code-focused applications.

3. xAI Enters the AI Coding Agent Race with “Grok Build”

What happened: xAI (Elon Musk’s AI company) released an early public beta of Grok Build, a terminal-based AI coding agent. The tool can read and edit code, propose plans for user approval before executing them, and run in interactive sessions. It enters a market already occupied by Claude Code, OpenAI Codex CLI, Cursor, and others.

Why it matters: The AI coding agent space is crowded, but xAI’s entry is notable for two reasons. First, Grok Build benefits from xAI’s fast iteration culture and from the large codebase already inside X (Twitter) and related properties where xAI models are deployed — giving it a potential dogfooding advantage. Second, xAI has been aggressive on model context window size and on real-time information access via X data; if Grok Build can inherit those strengths, it could differentiate on “code + live web context” workflows that other coding agents struggle with. The timing — landing weeks after Google I/O and amid the Codex/Claude Code arms race — suggests xAI is keen not to be left out of what is arguably the most commercially explosive AI application category right now.

Context: Early reactions note that Grok Build’s plan-approval-before-execution flow is similar to Claude Code’s cautious approach, in contrast to Codex’s more autonomous defaults. This positions it as a “safer” coding agent — a potentially smart product wedge.

4. Anthropic Closes $30B Round at $900B+ Valuation — AI Coding Economics Crystallize

What happened: Anthropic is finalizing a $30 billion financing round led by Sequoia, Dragoneer, Greenoaks, and Altimeter, at a post-money valuation exceeding $900 billion — surpassing OpenAI’s $852 billion March 2026 valuation. The round is expected to close within days. Anthropic also disclosed that Q2 2026 revenue is tracking at $10.9 billion (2× Q1), with its first-ever quarterly operating profit, driven primarily by Claude Code enterprise adoption (PwC: 100,000+ seats; multiple Global 2000 contracts).

Why it matters: The valuation and profitability news together confirm that AI coding is not just a product feature — it is a standalone, durable business large enough to support a $1 trillion company. Anthropic’s ability to reach an operating profit this early in the generative-AI era is, in part, a testament to how quickly enterprises are willing to pay for AI coding tools that demonstrably reduce delivery times. For the broader AI coding ecosystem, Anthropic’s capital scale also means Claude Code will continue to be aggressively subsidized and improved — raising the bar for every competitor.

Context: Andrej Karpathy’s recent move to Anthropic (announced May 19) further strengthens the signal that top AI talent sees Anthropic as the leading lab for the next phase of model development, including coding agents.

🤖 Embodied Intelligence

5. Figure AI 120-Hour Robot Sorting Livestream — Human vs. Robot Benchmark

What happened: Figure AI, the Brett Adcock-founded embodied AI company (valuation: ~$39.5B), hosted a continuous 120-hour livestream from its San Jose headquarters showing its humanoid robots sorting packages. Midway through, the company staged a 10-hour head-to-head competition between a human intern (Aimé Gérard) and the robot on the same sorting task. The human won — 12,924 packages vs. 12,732 — but the margin was slim (the robot averaged 2.83 s/item; the human 2.79 s/item).

Why it matters: This is one of the most transparent, quantitatively rigorous public comparisons between human and robot performance on a real-world manipulation task to date. The fact that a human still won — but by less than 1% — is a powerful data point for just how rapidly embodied AI manipulation speed is converging on human baselines. For the logistics and warehousing industry, the livestream also demonstrated multi-day autonomous operation with human-intervention-only-when-needed reliability, which is the actual deployment bar for commercial humanoid robotics. The event was also a viral marketing masterclass: Figure used a livestream to turn a would-be dry benchmark into a spectator event.

Context: Figure’s F.04 robot is in pre-production; the company is simultaneously running long-duration autonomous warehouse trials. The livestream complements the F.04 pre-production announcement as evidence that Figure is moving aggressively from R&D toward commercial scale.

6. Hangzhou Deploys World’s First Humanoid Robot Traffic Police Brigade

What happened: The city of Hangzhou (China) put 15 AI-equipped humanoid robots into active traffic management duty alongside human officers at key intersections, in what is described as the world’s first public-facing humanoid robot traffic brigade. The deployment was announced in mid-May and is part of a broader push by Hangzhou to position itself as China’s “first city of AI innovation.”

Why it matters: Embodied AI in public spaces has, until now, been mostly pilot-scale and choreographed. A 15-unit deployment managing real traffic is an order of magnitude more complex: it requires real-time multimodal perception, interaction with unpredictable humans (pedestrians, drivers), and reliable all-weather operation. If the brigade operates safely for months, it becomes a proof point for embodied AI in “high-friction” public environments — and a template for other municipalities. It also highlights China’s willingness to put embodied AI into uncontrolled public spaces faster than US regulators are likely to permit.

Context: Hangzhou is also the host city for the 2026 Global AI Technology Conference (May 23–24), where embodied AI was a central theme. The traffic brigade deployment doubles as a live demonstration for conference attendees and the global AI press.

7. Shanghai’s “Ge Wu” Embodied AI Simulation Platform Goes Live + China Pushes ISO Humanoid Robot Standards

What happened: The “Ge Wu” (格物) embodied AI simulation platform, developed under Shanghai’s AI initiative, officially launched in May 2026. The platform supports 100+ robot types from a single codebase and includes a live sim-to-real synchronization demo where a physical robot mirrors its simulated counterpart in real time. Simultaneously, China is advancing a bid to host an ISO/TC299 humanoid robot subcommittee, seeking to shape global safety and interoperability standards.

Why it matters: Simulation is the bottleneck for embodied AI scaling: you need to train and validate in simulation before risking a $100,000 robot in the real world. Nvidia’s Isaac Sim has been the dominant platform; “Ge Wu” is the first credible open alternative from China, and the claim that it supports 100+ robot types from one codebase is a significant technical claim if independently verified. The ISO standards push is strategically important: whoever writes the safety and interoperability standards for humanoid robots will shape the global supply chain. China is moving early to ensure its robot manufacturers are not locked out ofglobal markets by standards they did not help draft.

Context: The “Ge Wu” platform launch and the ISO push together represent China’s most assertive embodied AI global-standards play to date, and they bookend the Hangzhou traffic brigade news as evidence of a coordinated national-scale embodied AI strategy.

8. AGIBOT Unveils New Embodied AI Robots and Foundation Models at 2026 Partner Conference

What happened: AGIBOT — one of China’s leading humanoid robot companies — announced a new generation of embodied AI robots and associated foundation models at its 2026 Partner Conference. The announcement is built around a “One Robotic Body, Three Intelligences” full-stack architecture and includes four new robotic platforms and multiple AI models aimed at accelerating real-world Physical AI deployment.

Why it matters: AGIBOT is a central player in China’s humanoid robotics scale-up (Omdia’s 2026 Market Radar report names it alongside Figure AI, Agility Robotics, and Unitree as a leading general-purpose embodied robot vendor). The “One Body, Three Intelligences” framing suggests AGIBOT is building a modular stack where the physical platform, the perception intelligence, the reasoning intelligence, and the action intelligence can be upgraded independently — a design choice that matters for long-term deployability and for robotics fleet management. New foundation models purpose-built for embodied AI also signal that the field is moving beyond general-purpose LLMs adapted for robotics, toward models trained specifically for the constraints of physical action.

Context: AGIBOT’s announcement complements the “Ge Wu” platform news: simulation platforms and robot foundation models are the two halves of the embodied AI stack, and China is moving aggressively on both fronts in 2026.

🔍 Watchlist

Compiled May 26, 2026 by WorkBuddy AI Daily briefings.