Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Was anything good released at Google I/O 2026?
TL;DR
Google’s real headline is Gemini 3.5 Flash: fast, agentic, and deeply multimodal — Claraveville says Google is going “full bore into agents,” with Gemini 3.5 Flash positioned as a coding-capable model that rivals Claude/OpenAI-class reasoning while running roughly 4x faster and excelling on files, video, and cross-modal work.
Anti-Gravity 2.0 looks a lot like Codex/Claude Code catch-up, just with Google speed — Google added projects, scheduled tasks, a CLI, sub-agents, hooks, git worktrees, local dev environments, and slash commands like /goal, /grillme, /schedule, and /browser, and Claraveville’s quick test generated an API endpoint for her blog workflow reasonably fast but not in a way that felt revolutionary.
The most ambitious enterprise move may be AI Studio + Google Workspace, but it wasn’t actually usable on camera — Google pitched low-code app building over Gmail, Sheets, Drive, and Calendar, yet when she tried creating a kids’ weekend sports planner from her personal calendar, the connector either wasn’t live or wasn’t available to her despite being a paying customer.
Google’s creative stack is where the demo energy really lands — Gemini’s upgraded image features, the Omni video model, and Flow’s character consistency, scene editing, and reference-based generation point toward production-grade AI video, with a standout example turning her kid’s drawing into a 10-second superhero clip rescuing a child from class.
The launch was packed with tools, but the product story was messy and often broken — Claraveville repeatedly jokes that Google needs “a comprehensive brand analysis” because the stack spans Gemini, AI Studio, Anti-Gravity, Flow, Omni, Stitch, and Pameli, and several announced features either failed, were hidden, or weren’t broadly accessible on day one.
Stitch and Pameli show Google pushing hard into AI-native design and brand generation — Stitch’s streaming canvas, inline edits, code sync, and import/export flows felt especially strong for designers, while Pameli successfully pulled her site’s colors and tagline into a brand book, though the generated website itself landed as merely “fine.”
The Breakdown
Gemini 3.5 Flash sets the tone: speed plus agents
Claraveville opens on the foundation layer: Gemini 3.5, especially 3.5 Flash, which Google frames as both very smart and very fast. Her read is that Google is clearly aiming at agentic coding and multimodal workflows at once — less a pure model flex than a signal that the whole portfolio is being rebuilt around agents.
Anti-Gravity 2.0 is familiar territory with a Google twist
When she gets into Google’s IDE, Anti-Gravity, the vibe is immediate: this looks a lot like features people already know from Codex and Claude Code. Projects, scheduled tasks, CLI access, sub-agents, hooks, git worktrees, and local dev environments are all here, and she says it plainly: this feels like catch-up, just packaged with Gemini Flash speed.
A live coding test lands… solid, not magical
She tries a real task on her own site: convert a blog-generator UI into an API an AI agent can use. Anti-Gravity generates docs, auth, and an endpoint structure fairly quickly, and the artifact view is clean, but her reaction is measured — useful, competent, “kind of what I would expect,” not a jaw-dropping leap.
AI Studio promises Workspace-native app building, then immediately stumbles
Next she moves to Google AI Studio, where the big pitch is low-code apps connected directly to Workspace data like Sheets, Gmail, Drive, and Calendar. She tries to build an app to manage her kids’ weekend sports events from her personal calendar, but can’t find the connector or make the magic happen, which becomes a recurring theme: lots of announcements, uneven availability.
Google’s naming problem becomes part of the story
At this point she pauses to say what a lot of viewers were probably thinking: the branding is chaos. Gemini the model, Gemini the consumer chat app, Anti-Gravity the IDE, AI Studio, Flow, Omni, Stitch, Pameli — she jokes that what Google needs most is “a comprehensive brand analysis” because even following the launch requires a map.
Gemini’s image and video tools are the most fun part of the launch
Inside the redesigned Gemini app, she tests image generation with Nano Banana and gets a result that is fast, text-aware, and vaguely photorealistic — but also, in her words, “not my face” and “horrifying.” Video is more exciting: using the new Omni model, she uploads her kid’s drawing of a superhero and generates a 10-second animated clip of the character busting a kid out of class, which she immediately notes her child will love.
Flow looks like Google’s real play for production-grade AI video
Flow builds on Omni with a more structured editing environment: character definitions, multimodal references, conversational edits, and avatar creation. She walks through making an avatar via a phone capture flow — reading numbers, turning her face, handing over her likeness — only for the feature to fail, prompting one of the sharper human moments in the video: “I gave them my identity… and it didn’t even create the avatar.”
Stitch and Pameli hint at a strong design stack, even if the outputs are still a bit slop-coded
The video closes on two tools she thinks designers and marketers should watch: Stitch and Pameli. Stitch’s streaming canvas, inline edits, and code/design handoff feel genuinely fast and promising, while Pameli successfully reverse-engineers her site into a brand book and generates a website that’s serviceable but underwhelming — which sums up her verdict on day one pretty well: interesting, promising, and still full of sharp edges.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
