Google CEO Sundar Pichai’s description of Search as a future “agent manager” made headlines this week after an hour-long interview with Stripe CEO Patrick Collison.
As SEJ’s Roger Montti reported, Pichai described a version of search that leaves users “running lots of threads” and completing tasks rather than searching through results.
But the interview encompassed more than that one quote. During the conversation, Pichai laid out a timeline, identified the obstacles slowing adoption, described how he is already using an internal agent tool, and acknowledged infrastructure limitations that limit the rapid execution of this vision.
Here you can find out what the rest of the interview for search professionals reveals.
How Pichai’s language has escalated
The term “Agent Manager” was no coincidence. Pichai’s language about the future of search has become more concrete over the last 18 months.
In December 2024, he told an interviewer that search would “change profoundly in 2025” and that Google would be able to “answer more complex questions than ever before.”
In October 2025, during Google’s third-quarter earnings call, he called it an “expansive moment for search” and reported that searches in AI mode had doubled compared to the previous quarter.
In February 2026, it reported that search revenue reached $63 billion in the fourth quarter of 2025, with growth accelerating from 10% in the first quarter to 17% in the fourth quarter, attributing the increase to AI capabilities.
Now, in April, he’s putting a label on it. Not “Search will change” or “Search will expand,” but “Search as an agent manager” where users complete tasks.
Each time, language has moved from the abstract to the concrete, from prediction to description.
The turning point in 2027
Collison asked Pichai when a fully agent-based business process, such as automated financial forecasting, that doesn’t involve humans could happen at Google. Pichai pointed to next year.
“I definitely expect 2027 to be a major turning point for certain things in some of these areas.”
He added that non-technical workflows would change “pretty profoundly” in 2027, noting that some groups within Google are already working this way.
“There are some groups within Google that are changing more. For me, the big task is to expand this to more and more groups, especially in 2026.”
He also acknowledged that younger, AI-native companies have an advantage in adopting these workflows, while larger organizations like Google face retraining and change management issues.
The intelligence surplus
One of the most useful parts of the interview didn’t come from Pichai. It was Collison’s description of what he called the “intelligence overhang,” the gap between what AI can do today and the extent to which companies are actually using it.
Collison identified four barriers that slow adoption, even when the models are capable. The first is the ability to prompt. Getting good results with AI takes practice, and most people in organizations haven’t acquired this skill yet.
The second is the company-specific context. Even an experienced prompter needs to know which internal tools, data sets, and conventions to reference. The third is data access. An agent cannot respond, “What is the status of this deal?” if it can’t reach the CRM or if permissions are blocking it. The fourth point is the role definition. Job descriptions, team structures, and approval workflows are designed for a world without AI employees.
Pichai agreed with that assessment and said Google faces the same challenges internally.
“Identity access controls are really difficult problems and so we’re working on those things, but those are the key things that limit the spread for us as well.”
He described how Google’s internal agent tool, which he called Antigravity, is already changing the way he works as CEO. He said he polls it to get quick information about product launches.
“Hey, we started this thing. What did people think about it? Tell me the five worst things people talk about, the best five things people talk about, and I’ll write that.”
This is a concrete example of the agent manager concept that is being put into practice at Google today. Pichai uses search as a tool for completing tasks, not as a tool for returning links. The gap between that internal experience and what’s available to external users is part of what Google wants to close.
For SEO teams and agencies, it’s worth thinking about the information overload on two levels. There is an overhang within your own organization where AI tools may be able to do more than they currently do. And there’s an overhang on Google’s side, where the models are already capable of agent-style search, but the product hasn’t fully shipped it yet.
What is blocking the timeline?
Pichai confirmed that Google’s capital expenditures in 2026 will be between $175 billion and $185 billion, correcting the $150 billion figure cited by Collison. That’s about six times the $30 billion that Google spent before its current AI expansion.
When asked about bottlenecks, Pichai named four restrictions in turn.
Wafer production capacity is the most fundamental limit. Storage supply is “definitely one of the most critical constraints at the moment.” Approval and regulatory timelines for building new data centers are a growing concern. And critical supply chain components beyond storage add further pressure.
“There’s no way the leading storage companies are going to dramatically improve their capacity. So in the short term, there are those constraints, but they get looser as you go out.”
He said these restrictions would also lead to efficiencies, predicting that Google would make its AI systems “30 times more efficient” even as spending scaled.
He also noted that he personally spends an hour each week reviewing machine allocation at a granular level across teams and projects at Google.
What this means for search professionals
Pichai’s description of search as an agent manager changes the question SEO professionals must ask about their work.
With a results-based search model, the goal is to achieve ranking. In an agent-based model, the goal is to be useful to a system that completes a task. These are different problems.
Consider what the search completed by the agent looks like in practice. They direct searchers to find a plumber, check reviews, confirm Saturday morning availability, and schedule an appointment. The agent doesn’t return ten blue links. To complete the task, it uses structured business data, rating platforms and booking systems. The selected companies are those whose information is correct, structured and accessible to the agent. Those with outdated opening hours, no booking integration, or thin review profiles will not be shown.
The same pattern applies to e-commerce. One shopper says, “Find me running shoes under $150 that are suitable for flat feet and can arrive by Friday.” An agent who can handle this task will need product data, inventory availability, shipping estimates, and compatibility information. Websites that provide this data in structured, machine-readable formats become part of the agent’s toolkit. Websites that bury it in JavaScript-rendered pages or behind login walls will be skipped.
If an agent can compile an answer from five sources without directing the user to any of those sources, what value is there in being one of those five sources? That all depends on whether the agent quotes you, links to you, or treats your content as raw material without attribution.
This is consistent with the changes we see in AI mode. Google reported During its fourth quarter 2025 earnings conference call, the company said that searches in AI mode are three times longer than traditional searches and are frequently prompted Follow-up questions.
The schedule for 2027 is also important. As non-technical business operations become agents next year, the companies that provide the information and services those agents draw from will need to be structured for machine use, not just human browsing. Structured data, clean APIs, and accurate business information become infrastructure rather than nice-to-haves.
The measurement gap
Pichai’s insistence that AI search is non-zero-sum deserves more scrutiny than usual.
He made this argument consistently. In October 2025 he spoke of an “expansive moment”. In February 2026, he said Google had seen no evidence of cannibalization. In this interview, he compared it to YouTube thriving despite TikTok.
But overall request growth and individual website traffic are different metrics. Google may be right that more people are searching more often, while individual publishers and companies are seeing less referral traffic for those searches. Both things can be true at the same time.
Google did not share outbound click data from AI Mode. Until Google provides this data, Pichai’s “expansive” claim is an assertion and not a verifiable fact. Search professionals should independently track their own referral traffic trends rather than relying on Google’s characterization of the overall market.
Looking ahead
Pichai’s language in this interview goes beyond what Google has previously said publicly. Previous statements described AI search as an evolution. Here, Google’s direction for search is more clearly highlighted. Searching as an agent manager is a product vision.
The timeline he laid out, with 2027 as the tipping point for non-technical agent workflows, gives you some insight. How Google monetizes tasks completed by agents, whether agents cite sources or simply use them, and what visibility even means in an agent-manager model are all open questions that need to be answered before the start of 2027.
Google I/O 2026 is scheduled for May 19-20 and is expected to provide more details on how these features will be delivered.
Additional resources:
Featured Image: PJ McDonnell/Shutterstock
Follow us on Facebook | Twitter | YouTube
WPAP (907)
