52 minutes ago
19
Google just made it significantly easier for developers to build autonomous AI agents that can run in the background, connect to external servers, and handle complex multi-step tasks without human babysitting. The Gemini API’s Managed Agents platform, which launched less than two months ago, received a substantial feature upgrade on July 7 that reads like a wishlist from developers building production-grade AI systems.
The new capabilities include background execution for asynchronous tasks, integration with remote Model Context Protocol (MCP) servers, custom function calling, and streamlined credential refreshing during live interactions.
What Google actually built
The Managed Agents platform centers on what Google calls the Antigravity agent, an AI that runs inside a secure, Google-hosted Linux sandbox. Powered by models including Gemini 3.6 Flash, the Antigravity agent lets developers spin up autonomous agents with a single API call. Those agents can reason through problems, execute code, manage files, and perform web searches.
The original launch on May 19 was already notable. But these July additions address the gaps that separate a cool demo from something you’d actually deploy in production.
Background execution is perhaps the most consequential addition. Previously, agent tasks needed to maintain an active connection. Now agents can kick off long-running tasks and report back when they’re done, which is how most real-world automation actually needs to work.
Remote MCP server integration is the other headline feature. The Model Context Protocol has emerged as a standard for connecting AI models to external tools and data sources. By supporting remote MCP servers natively, Google is making it possible for agents to tap into third-party services, databases, and APIs without custom middleware.
Custom function calling rounds out the toolkit. Developers can now define their own functions that agents can invoke during execution. Combined with improved credential refreshing, which prevents agents from losing access to authenticated services mid-task, the platform is starting to look like genuine infrastructure rather than a research preview.
Why crypto builders should pay attention
Google’s announcement makes zero mention of cryptocurrency, blockchain, or digital assets. Not a single reference.
A developer building a crypto portfolio management agent, for instance, could use custom function calling to connect to exchange APIs, background execution to monitor positions around the clock, and MCP integration to pull in on-chain data from third-party providers.
The credential refreshing feature alone solves a pain point that anyone who’s built trading bots knows intimately. API tokens expire at the worst possible moments, usually during periods of high volatility when you need them most.
The competitive landscape is getting crowded
The MCP integration feature deserves particular scrutiny from crypto investors. As MCP becomes a standard protocol for agent-to-tool communication, the projects that establish themselves as essential MCP servers for blockchain data could capture outsized value.
Developers who are currently building AI agents for crypto applications now face a build-versus-buy decision that didn’t exist two months ago. Using Google’s managed infrastructure means faster time to market and lower operational overhead. It also means dependency on a centralized provider, which runs counter to the ethos of many crypto projects but aligns perfectly with the pragmatism of teams that just want their agents to work.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·