Alibaba Launches First Open-Weight Qwen3.5 Model

#1 𝕏

Qwen launched Qwen3.5-397B-A17B, the first open-weight model in the Qwen3.5 series with native multimodal capabilities and real-world agent training. It uses hybrid linear attention + sparse MoE for 8.

Also covered by: @Qwen

#2 📝 Simon Willison

Qwen3.5: Towards Native Multimodal Agents - Simon reviews Alibaba's Qwen 3.5 releases — a large open-weight Mixture-of-Experts model (Qwen3.5-397B-A17B) and a proprietary hosted Qwen3.5 Plus — noting their multimodal vision support, efficiency traits, and availability on Hugging Face and OpenRouter. He shares sample pelican images generated from both models and comments on context-length and hosted features.

Also covered by: @Qwen

#3 𝕏

Philipp Schmid built a minimal TypeScript agent framework for the Gemini Interactions API, split into agents-core (~500 LOC for a clean loop, streaming events and tool calling) and agent (built-in tools, hooks, sessions, skills & subagents).

#4 ▶️

Claude Code built me a $273/Day online directory

Greg Isenberg

#5 𝕏

Anthropic opened its Bengaluru office as its second Asia-Pacific hub and launched new partnerships across India to deepen its commitment to Claude, for which India is its second-largest market.

#6 𝕏

LlamaIndex 🦙 demoed an AI-powered Invoice Reconciler that uses LlamaParse to turn contracts and invoices into LLM-readable Markdown, indexes them in LlamaCloud, and applies a LlamaAgent workflow to match unit prices, math, and line items against custom rules—automatically fla...

#7 ▶️

WebMCP - Why is awesome & How to use it

AI Jason

#8 𝕏

claire vo 🖤 unveiled Joe’s custom AI accessibility tools for Slack built on Claude and shared a link to copy the workflows and integrate them into your stack.

#9 in

Carl Vellotti released a Claude Code skill file that permanently learns your writing voice and auto-flags AI-isms after a single use, with simple paste-in setup instructions. Join his 24,000+ PM Builders community at fullstackpm.com/subscribe.

#10 📝 Simon Willison

Rodney and Claude Code - Simon describes using Anthropic's Claude Code via native desktop apps (iPhone and Mac) and highlights a useful feature: Claude Desktop can display images it's "viewing" using a Read tool, providing visual previews while the agent works. He shows a screenshot produced during testing of his Rodney tool and explains the prompt used.

#11 𝕏

Cognition released GLM-5 and Minimax M2.5 on Windsurf, both now live with limited-time launch discounts.

#12 𝕏

Andrej Karpathy observes that LLMs excel at translating existing codebases—seen in surging C-to-Rust ports and COBOL modernizations—because the original code doubles as a detailed prompt and test reference.

#13 in

Udi Menkes: Product managers can now turn end-to-end workflows into Claude skills using Anthropic’s Skill Creator and publish them in the emerging Cloud Marketplace App Store, opening a brand-new go-to-market channel.

#14 𝕏

Santiago showcases Microsoft’s new multi-model evaluation and routing feature in Azure AI, letting you compare your app’s performance side-by-side across different models and automatically route prompts to the most cost-effective or highest-quality model.

#15 𝕏

Peter Yang relays Eno Reyes’ FactoryAI advice: evolve from ad-hoc prompts to reusable AI skills and workflows that compound. Build shared PM/UI/data skill libraries, deploy agents across sales and engineering, and grant read access company-wide.

#16 𝕏

DeepLearning.AI Andrew Ng sharply critiques the EU AI Act for stifling innovation, arguing that heavy-handed regulation prevents Europe from leading the AI revolution. They’ll dig into these real-world policy impacts at AI Dev 26 in San Francisco on April 28–29.

#17 𝕏

Sebastian Raschka argues that OpenAI’s free R&D access to the GPT-5.3 API gives it an edge, while Meta AI still lacks a flagship model beyond Llama 4 and its viability in OpenClaw remains unclear.

#18 𝕏

Sebastian Raschka argues API fees hurt solo developers but are trivial at enterprise scale, so claims of deep unprofitability only hold if every user got free API access—which won’t happen.

#19 𝕏

Logan Kilpatrick has added a ~30-second friction step to safeguard AI compute from nefarious use, aiming to balance service protection with minimal impact on legitimate users.

#20 𝕏

Logan Kilpatrick turned an AI Studio prototype into production-ready code with mobile support in just 1.5 hours, showcasing AI’s ability to accelerate real-world deployments.

#21 𝕏

Guillermo Rauch is piloting a virality-smoothing pricing model with Jmail to absorb sudden, Wikipedia-like global demand, and is enhancing Analytics pricing with automatic tiering to better support massive, outlier-scale sites.

#22 𝕏

Santiago recommends picking one model to quickly build a working AI application without early swaps or optimizations, then later refine for cost, speed, or performance.

#23 in

Peter Yang observes that titles—Engineer, Designer, PM, PMM, TPM, UXR—are largely irrelevant on top companies and teams, which prioritize collaboration and impact over labels.

#24 𝕏

claire vo 🖤 announces ChatPRD now indexes all your live and planned product features alongside competitors’ offerings, so you can code-drive a real-time us-vs-them feature matrix.

#25 in

Claire Vo launched a ChatPRD update that automatically indexes all your company’s live and planned features alongside competitors’, letting PMs “vibe-code” a comprehensive us-vs-them feature matrix.