The channel's been quiet lately — this edition covers the best from recent weeks.
Maury caught an interview with Anthropic's Felix Rieseberg — the engineer behind Claude Cowork — and pulled out a key line: the product roadmap is "one month, at most." For a team already building inside the product, that's not a horizon. It's a sprint.
The full interview covers local-first agents, why skills will matter more than most people expect, and a remarkable detail: Cowork itself was built in a week and a half using Claude. Felix's team described it to or built with the model, and steered as they went. The whole thing was a proof of concept for the very workflow it now enables.
Worth watching in full if you haven't already. The part about autonomy, safety, and the changing shape of knowledge work is where it gets interesting for us specifically.
GitHub shipped gh-stack on April 13, putting stacked pull requests into private preview. The idea: break a big feature into a chain of small, reviewable PRs that build on each other. The AI angle that matters — you can run npx skills add github/gh-stack and your coding agents learn to work with stacks natively. Achal flagged this as especially interesting for large features built with AI, where chunking the work is both safer and easier to review.
Shopify announced Agentic Storefronts, letting merchants get discovered and purchased through conversations on ChatGPT, Perplexity, and Microsoft Copilot. Commerce showing up wherever AI conversations happen is one of those pivots that sounds incremental until it isn't. The channel to retail is shifting underneath everyone.
via Achal — Apr 10Xenia put Google's new AI UI tool Stitch through its paces with the design team. The verdict: it's genuinely faster than Lovable and Cursor for visual output, and clicking a single element can generate a full screen of design. But it's not consistent — you'll get a scroll bar in the middle of a screen one pass and a solid layout the next. Design system imports work, but colors and styles drift. Multi-screen changes don't hold. Good for exploration and getting unstuck, not for production UI.
via Xenia — Mar 19Adobe launched Acrobat Student Spaces, a free AI study tool that converts PDFs into flashcards, quizzes, mind maps, and audio summaries. Positioning is direct: this is Adobe's answer to Google NotebookLM. The "no login required" entry point is a strong move. The underlying pattern — drop in a document, get back structured, consumable learning formats — applies well beyond the student market.
via Kunal — Apr 7Achal put a distinction on record that's worth repeating:
"Vibe coding = Prototypes / not for production. Agentic Engineering = reviewed production ready code which works and is safe."
This isn't just semantic. It's the frame we should be using when clients ask whether AI-built software can ship. The answer depends entirely on which of these two modes you're operating in.
via Achal — Apr 2Kunal flagged a Business Insider piece on vibe-coded apps clogging Apple's review pipeline. Apple's current pace is 200K+ submissions a week at a 1.5-day average review time — numbers they're managing, for now. The subtext: as the volume of AI-generated apps keeps rising, the bar for getting noticed in the store quietly gets higher. Shipping is easier. Standing out is not.
via Kunal — Mar 29Xenia's been running a Claude Code + Figma Console MCP workflow that deserves more attention: Claude crawls Figma frames directly, reads structure, naming, layout logic, and annotations, then compiles fully formatted design spec pages. Two projects in, output quality is strong. She also built a custom Claude skill to restyle the generated docs in Born West brand guidelines. Still WIP, but it's already cutting formatting time.
via Xenia — Mar 18Siddhaarth raised something on the slower end of the news cycle but harder to ignore: we're now at a point where social media's long-term damage is visible, and AI is moving faster and at greater scale than social media ever did. The question isn't whether AI is useful — it clearly is — but whether anyone is slowing down to think seriously about what we're building toward in five or ten years.
No hot take here. Just a good thing to keep in the peripheral vision.
via Siddhaarth — Mar 18Rauchg told TechCrunch that Vercel is "ready and getting more ready" for a public offering. The number he mentioned: 30% of apps running on Vercel's platform already come from agents. That's not a future projection — that's the current base. The infrastructure market, per Rauch, now has "no ceiling." Worth watching as a signal of how the market is pricing the AI infrastructure layer.
via Achal — Apr 15