Sponsored by
Hello from The AI Night,
Today in AI:
Anthropic Launches Claude Design for Visual Work
Codex Gets Computer Use, Memory, and Automations
Alibaba Open-Sources Qwen3.6-35B-A3B for Agentic Coding
Here's the deal: Anthropic Launches Claude Design in research preview, a new Anthropic Labs product for creating designs, prototypes, slides and one-pagers through conversation. It runs on Claude Opus 4.7, Anthropic's most capable vision model.
The Breakdown:
Available to Pro, Max, Team and Enterprise subscribers. For Enterprise, admins must enable it in Organization settings.
During onboarding, Claude reads a team's codebase and design files to build a reusable design system that auto-applies to future projects.
Inputs include text prompts, images, DOCX, PPTX, XLSX, codebase pointers and a web capture tool that pulls elements from live sites.
Refinement works through inline comments, direct edits and custom sliders Claude generates for spacing, color and layout.
Exports to Canva, PDF, PPTX and standalone HTML, plus a handoff bundle for Claude Code. Usage draws from existing subscription limits.
The bigger picture: Canva sells to people who can't design. Figma sells to people who can. Claude Design sells to people who don't even want to think about design. Reading a team's codebase and auto-building a design system means the first draft already looks like your brand. That solves the blank canvas problem entirely.
Here's the deal: OpenAI released a major Codex update that expands the tool beyond the editor into full desktop and browser workflows. The company says 3 million developers use Codex weekly.
The Breakdown:
Background computer use lets Codex see, click and type with its own cursor, running multiple agents in parallel on macOS without blocking other work.
A new in-app browser allows developers to comment on pages to give the agent precise instructions for frontend and game work.
Codex can generate images using gpt-image-1.5 inside the same workflow.
90+ new plugins add integrations including Atlassian Rovo, CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon, Remotion and Render.
Automations can now reuse threads and schedule future work across days or weeks; a memory preview stores preferences and corrections.
Support added for GitHub review comments, multiple terminal tabs, SSH to remote devboxes in alpha and rich file previews.
The bigger picture: Three million weekly developers is a bigger installed base than most dev tools ever reach. OpenAI is now using that base to push Codex past the editor and into territory owned by project management, CI/CD, and browser testing tools. This is not a coding assistant anymore. It is a platform play.
Here's the deal: Alibaba released Qwen3.6-35B-A3B, a mixture of experts model with 35B total parameters and 3B active parameters, published as open weights alongside API access under the name qwen3.6-flash on Alibaba Cloud Model Studio.
The Breakdown:
Scores 73.4 on SWE-bench Verified and 51.5 on Terminal-Bench 2.0, beating predecessor Qwen3.5-35B-A3B (70.0 and 40.5) and topping dense Qwen3.5-27B on Terminal-Bench (41.6).
Hits 1397 on QwenWebBench for front-end code generation, ahead of Gemma4-31B at 1197 and Qwen3.5-27B at 1068.
Matches Claude Sonnet 4.5 across most vision-language benchmarks, reaching 92.0 on RefCOCO and 50.8 on ODInW13 for spatial tasks.
Supports thinking and non-thinking modes plus a preserve_thinking option for multi-turn agent runs.
Compatible with OpenClaw, Qwen Code and Claude Code via the Anthropic API protocol; weights are on Hugging Face and ModelScope.
The bigger picture: Alibaba just proved a sparse model activating only 3B parameters can match what teams pay frontier API prices for. For builders burning thousands monthly on hosted coding agents, Qwen3.6 running on your own hardware changes the math overnight. Open-weight agentic coding stopped being a compromise. Alibaba made it a legitimate alternative.
The World's Biggest Dev Event Hits Silicon Valley
WeAreDevelopers World Congress comes to San José, CA — September 23–25, 2026. 10,000+ developers, 500+ speakers, and the full software development lifecycle under one roof, in the heart of Silicon Valley.
Kelsey Hightower. Thomas Dohmke (fmr. CEO, GitHub). Christine Yen (CEO, Honeycomb). Mathias Biilmann (CEO, Netlify). Olivier Pomel (CEO, Datadog). The people actually building the tools you use every day — all on one stage.
AI, cloud, DevOps, security, architecture, and everything real builders ship with. Workshops, masterclasses, and the official congress party.
What else you need to know:
Google Labs launched Flow Music, a standalone site rebranded from ProducerAI, for creating, sharing, and remixing AI tracks through natural language prompts, extending its Flow family beyond images and videos.
xAI launched standalone Grok Speech to Text and Text to Speech APIs, pricing STT at $0.10 per hour batch and TTS at $4.20 per million characters, undercutting Deepgram, ElevenLabs, and AssemblyAI.
Ollama released version 0.21 with built-in support for Hermes Agent, Nous Research's self-improving AI agent that features automatic skill creation, cross-session memory and 70+ default skills.
Canva integrated with Anthropic Labs' Claude Design, letting users import Claude drafts and AI-generated HTML artifacts into Canva's drag and drop editor to adjust layouts, colors and elements without regenerating code.
Nvidia CEO Jensen Huang told Stanford that AI is not eliminating jobs, arguing tasks get automated while job purpose remains, and warning that people using AI will replace those who do not.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
3 ways to support us:
Forward this to your AI-curious friend → https://www.theainight.com
Sponsor The AI Night and reach 500+ AI builders daily → passionfroot.me/theainight
Reply to this email — I read every response