When Gemini was launched as Bard, a lot of people criticized it as being rushed and slow, a sign that Google couldn't compete. After a messy couple of years, Google's story finally holds together: multimodal first, deeply embedded in Workspace, and now agentic infrastructure at scale. They were figuring out where AI fits inside a 25-year-old product empire (Search, Workspace, Android, Cloud) all at the same time. Google was doing something most companies can't afford to do: take public criticism while repositioning.
That pattern is playing out again. Google is getting a lot of early criticism for changing Search and its approach to Agentic Development, but I think some of that criticism may be premature. What looks like disruption may actually be Google case-testing how search should evolve so it can stay competitive and future-proof the company. With that said, here are a few thoughts from my favorite key points from the Google I/O keynote, which I got the chance to see live.
✅ Gemini + Gemma: Pros
Google's AI model strategy feels increasingly accessible and flexible. Gemini for managed AI, agentic workflows, and cloud orchestration, and Gemma for openness, portability, and privacy. Google is approaching both sides of the market. It's cool to see how they're building an exclusive ecosystem while also modernizing the agentic workflow. With Gemini Sparks, an autonomous 24/7 background agent available through the AI Ultra tier, you can now have Gemini-powered agents doing tasks and automations even while offline. The release of model weights under Apache 2.0, including Gemma 4, means developers and organizations have a more open path to run AI locally or inside their own secure infrastructure when privacy, compliance, or regulatory requirements matter. And with the Agent platform (formerly Vertex AI), developers can work across multiple models and providers with the option to run AI completely outside of Google's cloud ecosystem.
❌ Gemini + Gemma: Cons
Google and the industry as a whole will have to continue making the case for premium AI subscriptions, as lower-cost APIs and open ecosystems, including many coming out of China, keep pushing capabilities at dramatically lower prices. That said, Gemma 4 is an open-source model you can use safely at the enterprise level locally, which is Google's own answer to that pressure.
⚡ Antigravity Platform: Pros
Google isn't positioning Antigravity as just another AI coding IDE. Beyond the desktop experience, it extends into a CLI, SDK, API, and managed agents, while also including verification tools to help developers review output. Instead of competing strictly on specialized code editing like Cursor, or rapid terminal-based reasoning like Claude Code, Antigravity 2.0 tries to solve the whole workflow, plan, execute, test, and report back. It's essentially a multi-agent orchestration platform that supports parallel work through dynamic subagents, which can split tasks and run them simultaneously. I had actually built my own app based on this concept, not just assigning multiple agents to tasks, but having them work and communicate together on a task. Honestly, it's pretty dope to use a platform that integrates so much of the workflow I already wanted in one place.
⚠️ Antigravity Platform: Cons
It may be too agent-first for people who still want a normal IDE workflow. Direct code editing has been entirely removed from the primary interface, so you're forced into automated tasks with a real sense of lost control. I like having access to the terminal and being able to fix minor things manually, and what I hate the most is that visual indicators for inline warnings or errors on files are completely gone. There's also a learning curve, managing the system feels entirely different from traditional programming. You're not just writing code, you're managing a fleet of AI agents working at the same time. And because those agents use a lot of compute and tokens, it can get expensive, which is partly why Google introduced that $100/month Ultra tier for heavy users.
🛍️ Agentic Search: Pros
My personal favorite from the keynote. Google is now leveraging the same Antigravity-powered infrastructure that powers its development platform to dynamically build experiences around user intent rather than simply returning static results. The search engine uses Antigravity background subagents to fetch real-time web APIs, design an interactive frontend, spin up a secure sandboxed execution environment, and serve you a fully functioning web app in seconds, right inside your search results. It's both exciting and a little scary. This is the one that should keep every SaaS founder up at night. It collapses entire categories of simple-use-case tools. Why navigate to a unit converter, currency calculator, or lightweight data viz tool if the search result is the tool?
🔍 Agentic Search: Cons
The rise of agentic search is quietly reshaping the SEO landscape in a way that goes beyond just rankings. Google can now answer queries directly, build interactive tools inline, and keep users inside its ecosystem entirely. Your site may be sourcing the answer without ever getting the visit, fewer clicks, fewer page views, fewer ad impressions. Smaller publishers feel this the hardest since they depend on traffic volume more than brand authority. The rules of SEO are changing. It's no longer just about ranking, it's about being visible inside the AI answer itself.
📱 Google AI Studio: Pros
What stood out to me is how much friction Google is removing from prototyping. AI Studio can now generate Android apps, preview and test builds directly in the browser, connect with Workspace tools like Docs and Sheets, and deploy projects to Google Cloud and Google Play without ever leaving the platform. It even runs an Android emulator directly in your browser to test your build. For anyone who's dealt with the usual environment setup headaches, this is a significant shift.
🚧 Google AI Studio: Cons
It still has backend and scaling gaps. It's strong at helping you build the frontend and prototype quickly, but more complex apps still need extra work for databases, auth, permissions, and production-grade infrastructure. There's also a privacy tradeoff on the free tier, since your usage may be handled in ways that aren't ideal for sensitive or commercial work.
Conclusion
Ultimately, the same way people wrote Google off during the Bard era and were proven wrong, I think the current wave of criticism around Search and Agentic Development may age just as poorly. Google's bet is that the future of AI is a race to build the core operating system for an agentic world. By rapidly improving Gemini's latest models' multimodal capabilities, and leveraging their open model ecosystem with things like Gemma 4, they are securing both sides of the developer market.
Capturing that ecosystem early means they don't need a viral consumer breakout. They can inject their AI infrastructure directly into the tools developers already use to build the world's software. Google knows it doesn't need to build every consumer application to win the AI race. If they can make Gemini's latest models and Antigravity 2.0 the fastest, cheapest, and most robust orchestration layer for other people to build on, they can win by default.
Top comments (0)