Google's next-gen AI agents won't arrive until 2025
Google just pumped the brakes on its advanced AI agents, with CEO Sundar Pichai confirming that Project Astra won't ship until 2025 at the earliest. Despite rumors of a December launch for automated web chores, Google is taking extra time to crack real-time multimodal reasoning.

Remember Project Astra, the sleek Google assistant that could spot broken bicycle parts and identify city neighborhoods in real time through a camera?
If you were hoping to use it on your phone before the holidays, Sundar Pichai has some bad news.
During Google’s Q3 earnings call on Tuesday, Pichai officially pushed the timeline back, stating that experiences based on Project Astra will ship as early as 2025 at the earliest.
A reality check for December rumors
First demoed at Google I/O in May, Project Astra was pitched as the foundation for Google’s next-generation multimodal agents. It promised software that could see what you see, reason about the physical world, and take actions on your behalf.
Pichai's statement throws cold water on recent reporting from The Information, which claimed Google was preparing to drop a consumer-facing agent capable of booking flights and buying products as soon as this December.
Unless that shopping agent is built on a completely separate tech stack, consumer-ready action models from Mountain View are officially shelved until next year.
The delay also highlights just how brutal the agentic frontier really is. Anthropic recently took the lead by rolling out computer-use capabilities for Claude, but early trials show that letting AI drive software often breaks down on even simple tasks.
Why it matters
For everyday users, this means your phone will stay in the prompt-and-response era for a bit longer. Don't expect Gemini to handle your holiday travel bookings or run errands inside your apps anytime soon.
For the AI industry, Pichai’s comments confirm that the transition from smart chatbots to reliable autonomous agents is proving to be a massive engineering bottleneck. Demo videos are easy to script, but real-world reliability across messy interfaces is another beast entirely.
We think Google is making the smarter long-term move by hitting pause rather than shipping a half-baked assistant. When a chatbot hallucinates, it is annoying; when an agent hallucinates your credit card details or flight times, it is catastrophic. That said, after blowing minds at I/O, Google now faces immense pressure to actually deliver the goods in 2025 without cutting corners.
Sources (1)
- techcrunch.com(opens in a new tab)Originaltechcrunch.com