Google turns Search into an agent manager: Sundar Pichai's vision

In a wide-ranging interview with Stripe's CEO, Sundar Pichai sketched far more than an abstract vision. He set a deadline, listed technical obstacles, and described how he personally uses these tools internally. Here’s what web professionals should take away.

Search as an agent manager

This is not the first time Sundar Pichai has spoken about the evolution of Google Search. But this time, the language clearly shifted register.

In December 2024, he promised that search would " In April 2026, Sundar Pichai now puts a precise label on the project: Search as an "agent manager," a tool where users have "multiple execution threads running" and accomplish concrete tasks rather than browsing results. ". In October 2025, during Google's quarterly results, he mentioned "" and announced that AI-mode queries had doubled quarter after quarter. In February 2026, he attributed the growth in Search revenue (USD 63 billion in Q4 2025, with an acceleration from 10% to 17% year over year) to AI features. to change profoundly in 2025 an expansionary moment for Search

In short, each of these statements moved the discourse from abstract to concrete, from prediction to description. This semantic shift is not trivial: it signals that the product vision is now defined enough to be named.

When asked by Patrick Collison about the date when an entirely agentic business process could appear at Google, Pichai pointed to 2027.

2027: the inflection point Pichai himself set

He specified that non-technical workflows—that is, outside engineering—would undergo "quite deep" transformations as early as that year. Some internal Google teams are already working this way. His mission for 2026:

spread these practices to as many groups as possible He also acknowledged a structural advantage for "AI-native" younger companies, able to adopt these new workflows without the burden of training and change management that weighs on organizations like Google..

This timeline is operational intelligence for SEO and marketing teams: this is not a distant transformation, but a shift to prepare for now.

This timeline is an operational tool for SEO and marketing teams: it’s not a distant transformation, but a shift that needs to be prepared now.

"Intelligence overhang": the gap between capability and real-world use

One of the most instructive exchanges in the interview did not come from Pichai, but from Collison himself. The CEO of Stripe described what he calls the "intelligence overhang": the gap between what AI can do today and what organizations actually use it for.

He identifies four barriers to adoption:

• The first is mastery of promptingGetting good results requires practice, and the majority of company employees have not yet developed this skill.
• LThe second is the context specific to each organization : even a good prompter must know which internal tools, which datasets, and which conventions to mobilize.
• The third is access to data : an agent cannot answer "what's the status of this file?" if it doesn't have access to the CRM or if permissions prevent it.
• The fourth is the definition of roles : job descriptions, team structures and validation workflows were designed in a world devoid of AI colleagues.

Pichai validated this diagnosis and admitted that Google faces the same problems internally. He specifically cited identity access controls as a difficult issue that also limits internal rollout.

For SEO teams and agencies, this concept of overhang applies on two levels: first within their own organizations, where AI tools could be exploited far more than they are. Second, on Google's side, where models are already capable of agentic search, but the product has not yet fully caught up to that capability.

Antigravity: how Google's CEO is already using agentic search

Beyond the declarative statements, Pichai gave a concrete example of what agentic search produces in practice, describing his personal use of an internal tool called Antigravity.

As CEO, he queries it after every product launch to quickly get an overview of reactions. He asks for the five most criticized points and the five most praised. It's a use of search as a task-completion tool, not as a link-return engine.

The gap between this internal experience and what is accessible to external users is precisely what Google is trying to close. It provides a concrete measure of the product's direction.

Constraints slowing deployment

Sundar Pichai confirmed that Google's investment budget for 2026 will be between $175 billion and $185 billion. It's about six times what Google was spending before the start of its AI ramp-up.

Asked about bottlenecks, he listed four constraints in order:

• The wafer (silicon slice) production capacity is the most fundamental limit.
• Memory supply is " certainly one of the most critical constraints today ".
• Permit and regulatory delays for building new data centers are an increasing concern.
• Finally, some critical components of the supply chain beyond memory add further pressure.

Sundar Pichai nevertheless indicated that these constraints drive efficiency gains: he predicts Google will make its AI systems "30 times more efficient" despite increased spending. He personally spends one hour per week reviewing in detail the allocation of compute capacity across teams and projects.

What this means concretely for SEO

The concept of an agent manager changes the questions SEO professionals need to ask themselves. In a result-based search model, the goal is to rank. In an agentic model, the goal is toto be useful to a system that performs a task. These are two different problems.

Concrete example: a user asks the search system to find a plumber, check their reviews, confirm availability on Saturday morning, and make an appointment. The agent does not return ten links. It draws on structured data, review platforms, and booking systems to complete the task. The selected businesses are those whose information is accurate, structured, and accessibleThose with outdated hours, no booking integration, or few reviews do not appear.

The same pattern applies to e-commerce: if a buyer asks " running shoes under 150 euros, suitable for flat feet, deliverable on Friday ", the agent needs product data, stock availability, delivery estimates, and compatibility information. Sites that provide this data in structured, machine-readable formats become resources the agent uses. Others are bypassed.

There is also the question of visibility in an agentic world: if an agent can synthesize an answer from five sources without sending the user to any of them, what is the value of being one of those sources? It entirely depends on whether the agent cites you, links to you, or simply treats your content as raw material without attribution.

The assertion Sundar Pichai still needs to demonstrate

Pichai has repeatedly stated that AI search is a non-zero-sum game: in October 2025 he spoke of an expansionary moment; in February 2026 he said he saw no sign of cannibalization; in this interview he compares the situation to YouTube, which prospered despite TikTok.

But total growth in queries and individual site traffic are two different metrics. Google can be right that more people are searching more often while leaving publishers and e-commerce sites with less referral traffic from those searches. Both can be true at the same time.

Google has not published data on outbound clicks since AI Mode. Until those figures are available, Pichai's claim of expansionism remains an assertion, not a verifiable factnot a verifiable fact. Search professionals therefore have every reason to track their own referral traffic trends independently, rather than relying solely on Google's reading of the overall market.

Open questions before 2027

How will Google monetize tasks completed by agents? Will agents cite their sources or just use them? What does visibility mean in an agent-manager model?

These questions have no answer yet. Companies that structure their data, APIs, and product information now for machine consumption will be ready. Those that do not will have to catch up in an already reconfigured environment.

Scheduled for May 19–20, Google I/O 2026 should provide details on how these capabilities will be deployed in practice.

The article “Google turns Search into an agent manager: Sundar Pichai's vision” was published on the site Abondance.