Google Evolves Stitch into AI Design Canvas

If you want to master building enterprise-level AI products from scratch and future-proof your career with the most important AI skills, this #1 AI PM Certification is for you.

Run by real AI operators: Rohan Varma (ex-cofounder, first ever PM at Cursor, now at Codex) & Henry Shi (former co-founder of Super.com, a $200M+ ARR company, now technical staff at Anthropic Labs, working directly with the founders).

3,000+ AI PMs graduated  ·  1,000+ reviews on Maven — highest in any category

#1 𝕏

Google AI has evolved Stitch by Google from a Labs prototype into an AI design canvas that turns natural language, image or code prompts into production-ready front-end code.

#2 𝕏

Josh Woodward launched Stitch Live in today’s huge Stitch update—now you can click and *talk* to your designs for instant edits or use it as a real-time sounding board for design critiques.

#3 𝕏

Philipp Schmid shared that Gemini API now lets you combine built-in tools like browser and calculator with custom function calls, backed by a revamped Python client and interactive docs with live examples.

#4 𝕏

Philipp Schmid announced that the Gemini API now combines built-in tools (Google Search, Google Maps on Gemini 3, File Search, URL Context) with custom functions in a single orchestrated API call, complete with automated tool chaining and signature-based context.

#5 𝕏

LlamaIndex 🦙 launched LlamaParse Agentic Plus mode, adding bounding‐box visual grounding for complex LaTeX, handwriting, multi‐column layouts, and charts. Now you can extract text and data with precise spatial context for smarter document workflows.

#6 𝕏

There’s An AI For That: MedOS pairs AI reasoning, XR smart glasses and robotics into a unified clinical co-pilot. Built by Stanford-Princeton and showcased at NVIDIA GTC 2026, it delivers real-time diagnostics and AR-guided procedures.

#7 in

Peter Yang demos a new tutorial on using Gemini in Google Workspace with five real use cases—Sheets (camp research, family budget), Docs (vacation planning), Slides & Docs (script and slide creation), and Drive (file Q&A).

#8 ▶️

Build Memory-Aware Agents

Deeplearning.ai

#9 📝 Simon Willison

Autoresearching Apple’s “LLM in a Flash” to run Qwen 397B locally - A write-up of Dan Woods' experiments getting a large Mixture-of-Experts Qwen model to run efficiently on a 48GB MacBook Pro by streaming expert weights from SSD and heavily quantizing experts. The work uses techniques from Apple's "LLM in a flash" paper and an autoresearch workflow driven by Claude to run many experiments and produce Metal/Objective-C code.

#10 𝕏

Yann LeCun warns that AI risks stem not from agency itself but from agents lacking world models and safety guardrails, and proposes building objective-driven AI systems that predict action consequences and only execute steps meeting defined safety constraints.

#11 📝 Simon Willison

Snowflake Cortex AI Escapes Sandbox and Executes Malware - A summary of a PromptArmor report describing a prompt injection chain in Snowflake's Cortex Agent that allowed execution of malicious commands by abusing an unsafe 'cat' allowance and process substitution. The issue has been fixed, and the author argues for deterministic sandboxes rather than trusting allow-lists of safe commands.

#12 𝕏

LlamaIndex 🦙 calls context engineering—strategically feeding system prompts, chat history, retrievals and structured data—the evolution beyond prompt engineering for AI agents. It launches LlamaParse and LlamaExtract to turn complex documents into neatly structured context.

#13 𝕏

clem 🤗 compares relying on off-the-shelf AI to claiming Squarespace outdid software engineering in 2000 and argues PMs must learn to train custom models to achieve cheap, fast, and accurate domain-specific AI.

#14 𝕏

Teresa Torres says measuring success by shipped features turns teams into feature factories, whereas measuring by impact drives learning, adaptation, and genuine value—though putting outcomes over outputs into practice is deceptively challenging.

#15 𝕏

Santiago warns that, just as you wouldn’t remodel a kitchen or shoot a movie without a blueprint, you shouldn’t let AI agents change your code without first drafting a plan.

#16 𝕏

Teresa Torres argues that a 1M-token context window offers the best performance, enabling LLMs to maintain extensive context and deliver richer outputs.

#17 𝕏

Santiago says texting your AI agent (e.g., HeyNoahAI) is now the norm and praises Noah’s approach of building context incrementally from your interactions instead of demanding full data access upfront.

#18 𝕏

clem 🤗 warns that their largest open-source repos are swamped by low-quality AI-generated pull requests (~1 every 3 minutes), making GitHub nearly unusable and surfacing fresh agentic challenges.

#19 𝕏

Claude is bringing back its Code with Claude developer conference this spring in San Francisco, London, and Tokyo, featuring full-day workshops, demos, and 1:1 office hours—register to attend in person or watch online.

#20 𝕏

Anthropic found that 67% of people globally view AI positively, with stronger optimism in South America, Africa, and Asia than in Europe or the United States.

#21 𝕏

Anthropic used its new Interviewer tool to conduct in-depth interviews that reveal how people worldwide are already experiencing AI’s opportunities and risks, and will run these interviews regularly across topics to guide building AI that benefits everyone.

#22 in

Thomas Hendrickx recommends Claire Vo’s “How I AI” YouTube series for its real-world AI workflows—building products, setting up systems, and solving problems in practice.

#23 ▶️

Master Gemini in Google Docs, Sheets & Slides in 18 Min (5 Real Use Cases)

Peter Yang