Oct 13 · AI Daily

xAI has launched The X Subscription, bringing Grok, Grok Bot, Cursor, and X Premium into a single plan. It will roll out to users worldwide over the coming days.

Grok has officially announced that it is joining The X Subscription. A single subscription grants access to Grok, Cursor, and X Premium, starting at $8 per month across five available tiers. Users who already subscribed to SuperGrok can go to their account settings to switch to The X Subscription.

Cursor has announced it is joining The X Subscription. Existing Cursor subscribers can keep their current plan or switch over.

Current configuration:

gpt 6.1 sol (high) cloud envs (default) dots as orchestrator codex exec for evaluation native codex app

Some people like to keep their workflows under wraps. Our speakers shared theirs directly.

During a WorkBuddy meetup at SF Tech Week, seven speakers shared real-world playbooks—ranging from a solo content engine and a five-platform newsletter to automated daily operations—all powered by WorkBuddy.

Good ideas are meant to be shared.

Cursor has announced its integration into The X Subscription. Existing Cursor subscribers can keep their current plan or switch to the new subscription.

Amp's Changes panel can now automatically sort files in a smart order by reading diffs, identifying the files most critical to the current changes, and pinning them to the top. If the diff changes, Amp recalculates the order without rearranging files while a user is reviewing them; instead, a blue dot appears on the smart sort button, and the new order is applied when clicked. The new sorting calculation is much faster than the previous method, and users can also click the sort button to switch back to alphabetical order.

Ma Dongseok NLP has raised two core questions for AI-native companies: whether smarter AI leads to faster R&D and whether product iteration can speed up accordingly; and how to manage AI agents to maximize efficiency when resources are limited.

Office Hours will host Granola VP of Sales Bardia Shahali on Oct 20 at 10:00 PT to discuss how sales organizations should be structured in the AI era. Topics include why sales teams are smaller in the AI era, changes in quotas compared to traditional SaaS benchmarks, and how faster onboarding and larger deal sizes change hiring and compensation. The event will also cover go-to-market design when a product spreads organically.

It's finally here! A single unified subscription covering Grok, Grok Bot, Cursor, and X Premium at an incredible value.

Cursor, Grok, and X Premium are now combined into a single subscription!

https://x.ai/pricing

According to Bloomberg, Epic Games delayed a Disney game originally slated for a March release due to underestimating staffing needs, and designated it as an "ai first" project. The company announced last week that there will be "no more handwritten code" and that games will be "generated with an all-AI workflow." Employees reacted with shock upon hearing the news, and an Epic spokesperson confirmed the use of AI to accelerate development.

Artificial Analysis has launched the AA-3D benchmark, having 21 models act as agents in an offline sandbox to turn written briefs into three.js web pages, with Elo rankings generated through blind evaluations by over 1,000 judges and more than 10,000 votes.

Artificial Analysis has released AA-3D v1.0 to evaluate AI models' ability to build interactive 3D experiences, alongside a concurrent leaderboard. The benchmark covers dimensions such as scene construction and per-scene cost, and the leaderboard will be updated continuously as new models are released.

Johns Hopkins University has proposed SELFCOMPACT, which lets models decide for themselves when and how to compress context without requiring fine-tuning or external supervision. The method combines compression tools with lightweight scoring rules, matching or exceeding fixed-interval summarization across six benchmarks and seven models, with math tasks improving by up to 18.1 points, agent search improving by 5–9 points, and per-query costs reduced by 30–70%.

Apple's research team evaluated the consistency of LLMs with human annotators under MQM and ESA machine translation quality annotation schemes, testing across a long-context dataset of 70 language pairs and public WMT23 and WMT25 data. Results show that LLMs exceed inter-human consistency on certain tasks, but remain unreliable across most annotation schemes, language pairs, and domains. Humans struggle particularly with fine-grained MQM annotations and low-resource language pairs, while LLMs struggle to handle subtle errors, incorrect language variants, and incorrect span annotations.

In response to Rich Hickey's Oct 11, 2026 article, ""AI" is Not a Tool", which argued that LLMs belong to automation rather than tools, an author countered that LLMs can still be directed by humans and require skill to use effectively.

According to Bloomberg, Epic Games has shifted to an AI-first development approach for its $1.5 billion partnership project with Disney, with models like Anthropic's Claude writing all the code and guiding all art assets, while human developers manage and direct the generated content.

The Apple research team has proposed Minions, a dynamic distributed virtual GPU cloud that aggregates idle Apple Silicon GPUs across multiple macOS devices to execute large-scale machine learning inference. The system supports flexible GPU inference task execution, fault tolerance, and efficient resource utilization, allowing teams to leverage existing hardware without purchasing additional GPUs. Experience shows that in environments with constrained centralized GPU resources, idle macOS devices can be turned into a practical distributed GPU inference platform.

Alibaba's Qwen team has introduced QwenGyre, an elastic reinforcement learning framework designed for training ultra-long-horizon agents. It trains coding agents via online RL within unmodified black-box harnesses, allowing single rollouts to run for hours and handle nearly 1 million tokens.