Google CEO Sundar Pichai Says 2027 May Be a Turning Point for Agent-Based Workflows
Google CEO Sundar Pichai stated in an interview that 2027 could become a significant turning point for agent-based workflows, with notable changes expected in some non-engineering processes.
He pointed out that information retrieval searches will increasingly shift towards agent-based systems, allowing users to complete tasks while running multiple threads simultaneously. Search may evolve into an "agent manager," coordinating multiple tasks for users.
Pichai mentioned that he is already using Antigravity, and some internal engineering teams have adopted this tool to manage agent workflows. Google I/O 2026 will launch Antigravity 2.0 as a standalone desktop application to support the orchestration of multiple agents, alongside the Gemini 3.5 Flash model.
He also acknowledged that Google is currently relatively behind in agent-based coding concerning tool usage, instruction following, and long-cycle tasks. The company is increasing usage internally through Antigravity to gather feedback and improve the model.
Approximately 75% of new code at Google is generated by AI and reviewed by engineers. The engineering team is shifting towards agent-based workflows, completing complex migrations through the division of roles among planners, orchestrators, and coders, achieving speeds several times faster than before.
Agent orchestration will shift developers from line-by-line coding to task decomposition and coordination among multiple agents. Capital and computing power continue to be invested in models and platforms, while developers will spend more time reviewing agent outputs and system architecture rather than pure syntax implementation.
Source: Public Information
ABAB AI Insight
Sundar Pichai has long advocated for Google to deeply embed AI into products and internal processes. The company will launch the Antigravity 2.0 desktop application and SDK at I/O 2026, positioning it as a platform for orchestrating multiple agents, while Gemini 3.5 Flash is optimized for agents and long-cycle tasks. Internal engineering teams have already utilized agents for code generation, migration, and daily tasks.
Pichai described Search as the future agent manager in the interview, noting that 2027 may be a turning point for non-engineering workflows to shift towards agents. About 75% of new code at Google is generated by AI and then reviewed by engineers, with complex migrations accelerated through a multi-role agent system, indicating a shift of resources from pure manual coding to agent orchestration and review.
Similar trends can be seen among other model providers pushing for coding agents and multi-agent orchestration tools, with some companies productizing internal agent capabilities into developer platforms. Google is currently in a phase of acknowledging its relative lag in agent-based coding while expanding usage feedback through Antigravity, differing from earlier tools primarily focused on single generation or IDE plugins.
This represents a technological replacement: agents will detach line-by-line coding and some execution tasks from direct developer operation, while elevating review, task decomposition, and multi-agent coordination as primary work content. The developer role is shifting from direct implementation to managing agent outputs and system logic, changing the traditional skill structure centered on syntax and API memorization.
ABAB News · Cognitive Laws
Agent orchestration equals delegating execution to models
Code generation does not equal the disappearance of architecture review
2027 turning point equals the year of workflow transition