Agents move into the tools you already use
Notion 3.0, GPT-5-Codex and Claude in Xcode, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
Notion
Notion 3.0
DEPLOYif your team lives in Notion
WHAT SHIPPEDNotion 3.0 rebuilds Notion AI as Agents that can take on a whole project, break it into tasks, assign them, and draft docs. Agents can update or create hundreds of pages at once across databases.
THE CATCHThe Agent pulls context from your workspace, connected tools like Slack, Google Drive, and GitHub, and the web within your permissions. Database row permissions are available for Business and Enterprise plans.
IF YOU RUNa team that tracks work in Notion databases and wants agents to draft launch plans, update records, and file tasks without manual busywork
SOURCENotion, Releases / September 18, 2025 3.0 Notion 3.0: Agents
OpenAI
GPT-5-Codex
DEPLOYif you use Codex for agentic coding
WHAT SHIPPEDGPT-5-Codex is a version of GPT-5 optimized for agentic coding in Codex. It is the default for cloud tasks and code review, and developers can choose it for local tasks via Codex CLI and the IDE extension.
THE CATCHOn OpenAI employee traffic, for the bottom 10% of user turns sorted by model-generated tokens, GPT-5-Codex uses 93.7% fewer tokens than GPT-5. For the top 10%, it thinks more, spending twice as long reasoning, editing and testing code.
IF YOU RUNa development team that wants code review to catch critical bugs before they ship and an agent that works independently for more than 7 hours on large tasks
SOURCEOpenAI, Introducing upgrades to Codex
Anthropic
Claude in Xcode
DEPLOYif you build Apple platform apps
WHAT SHIPPEDDevelopers can connect their Claude account to Xcode 26 to power coding intelligence features with Claude Sonnet 4. The integration includes a coding assistant and Coding Tools for documentation, explanations, SwiftUI previews, and inline code changes.
THE CATCHYour Claude subscription usage limits are shared across platforms, with a portion available through Xcode. Claude in Xcode works with all plans that include Claude Code: Pro and Max, plus premium seats for Team and Enterprise plans.
IF YOU RUNan Apple platform development team that wants Claude's coding capabilities directly in Xcode without leaving the IDE
SOURCEAnthropic, Claude is now generally available in Xcode
The Deploy Log
Models and APIs
| DEPLOYif you need EU data sovereignty Scaleway Inference Provider. Scaleway is now a supported Inference Provider on the Hugging Face Hub, offering serverless access to models like gpt-oss, Qwen3, DeepSeek R1, and Gemma 3 with pay-per-token pricing starting at €0.20 per million tokens. Hugging Face |
| DEPLOYif you train robot learning models LeRobotDataset v3.0. LeRobotDataset v3.0 packs multiple episodes in a single file using relational metadata, and natively supports streaming mode to process large datasets on the fly without downloading prohibitively large collections onto disk. Hugging Face |
Tools and products
| WATCHif you deploy on Vercel Vercel build cache. Vercel improved its build cache performance by downloading parts of the cache in parallel using a worker pool, decreasing build initialization time by 30% on average and reducing build times by up to 7 seconds for all plans. Vercel |
| DEPLOYif you export observability data Vercel Drains. Vercel Log Drains are now Vercel Drains, exporting OpenTelemetry traces, Web Analytics events, and Speed Insights metrics alongside logs, with data billed at $0.50 per GB on Pro and Enterprise plans. Vercel |
| WATCHif you run many static paths Vercel Bloom filter routing. Vercel replaced a slow JSON parsing operation with a Bloom filter in its global routing service, reducing memory usage by 15% and improving time-to-first-byte from the 75th percentile and above by 10%. Vercel |
| DEPLOYif you run serverless on Vercel Vercel Fluid cold starts. Vercel delivers zero cold starts for 99.37% of all requests through scale to one, Fluid compute, predictive scaling, bytecode caching, and rolling releases, with single instances handling peaks above 250 concurrent requests. Vercel |
Left out
- GPT-5-Codex system card addendum: The page text is a safety publication summary with no new deployment figure beyond what the main Codex release already carries.
- Vercel deployment retention: the publisher's page does not show it available yet
- mcp-to-ai-sdk: it did not carry a number that moved
- Gradio visible watermarking: it did not carry a number that moved
- Waymo Nashville: it did not carry a number that moved
- Public AI Inference Provider: it did not carry a number that moved
- Observability Plus: it did not carry a number that moved
- Vercel Agent code reviews: it did not carry a number that moved
- Stability AI Image Services: it did not carry a number that moved
The pattern
The agentic tools are no longer separate destinations. Notion 3.0 puts Agents inside the workspace where teams already track work, OpenAI makes GPT-5-Codex the default inside Codex, and Anthropic connects Claude Sonnet 4 to Xcode 26. Each move embeds an agent where the user's context already lives, so the agent can act on hundreds of pages, run code review, or edit SwiftUI previews without a new workflow.
The infrastructure underneath is tightening at the same time. Vercel cuts build initialization time by 30%, reduces memory usage by 15%, improves time-to-first-byte by 10%, and delivers zero cold starts for 99.37% of requests, while Hugging Face adds Scaleway pay-per-token pricing starting at €0.20 per million tokens and LeRobotDataset v3.0 streaming for large datasets. Agents get faster, cheaper, and more local, and the platforms they run on get faster and cheaper in the same release cycle.