I actually loved this interview with Sundar Pichai by John Collison and Elad Gil of Stripe.
Listed below are the 5 most fascinating issues I discovered.
1. Search Will Nonetheless Exist In The Future, However A lot Of It Will Be Agentic
Sundar was requested if brokers would substitute Search. He mentioned:
“If I quick ahead, a number of what are simply information-seeking queries will probably be agentic in Search. You’ll be finishing duties. You’ll have many threads working.”
And in addition, Search will change in order that we consider it like an agent supervisor.
“It retains evolving. Search will probably be an agent supervisor by which you’re doing a number of issues. I feel, in some methods, you recognize, I take advantage of Antigravity at present, and you recognize, you’ve gotten a bunch of brokers doing stuff, and I can see search doing variations of these issues, and also you’re getting a bunch of stuff carried out.”
He mentioned that folks do deep analysis in AI Mode, and it’ll quickly be the norm to do long-running duties. He additionally mentioned that the shape issue of gadgets will change.
2. Google Makes use of Antigravity Internally
Boy, do I really like Google’s IDE and agent supervisor, Antigravity. I’ve constructed so many issues with it, together with my very own RSS feed reader, a screenshot and annotation device, workflows to publish issues I write in a Google Doc to my WordPress website, and a bunch of instruments to do agentic issues with Google Search Console and Google Analytics 4 knowledge. Whereas I feel Claude Cowork and Claude Code are unimaginable, I really do choose utilizing Antigravity.
It seems that Google makes good use of Antigravity internally. Besides they don’t name it Antigravity. They name it “Jet Ski.”
Sundar mentioned that the Google DeepMind and the Google Software program Engineers use it:
“I can see teams, and particularly I’d say GDM and among the SWE teams actually change their workflows. They’re utilizing, we name this for some unusual purpose, now we have a special title internally than externally of the identical product, but it surely’s Jet Ski internally which is Antigravity. You’re residing on it, you’re residing in an agent supervisor world. You may have workflows, and also you’re working on this new means.”
He additionally makes use of it himself.
“I’d question in Antigravity, in our inner model of Antigravity. “Hey, we launched this factor. What did individuals take into consideration this? Inform me the worst 5 issues persons are speaking about?” and I sort that. Now that brings it again. Has my life gotten simpler? Sure. Previously I’d have to spend so much extra time making an attempt to get a way for it. Now an AI agent helps me in that journey.”
Additionally, simply final week, the Google Search staff began utilizing Antigravity.
“Simply final week we rolled it [Antigravity] out to the Search staff. We’re continually pushing that. In a big group, I feel change administration is a tough facet of this expertise diffusing, which can be straightforward for a small firm. You possibly can rapidly swap over.”
If you wish to learn to use Antigravity, I’ve created a full information educating you the way it works, and the way I take advantage of it to not solely code, however create full agentic workflows that I truly use in my day-to-day work. It’s obtainable within the paid a part of my neighborhood, The Search Bar. And subsequent Thursday, the Search Bar Professional crew is having an occasion the place we’re going to separate into two groups, Crew Claude Code and Crew Antigravity, and see who can construct the higher search engine optimisation device.
I do know it’s a little bit of a ache to attempt to use one thing new in your workflows. I completely imagine that those that learn to use Antigravity at present can have a giant benefit as issues actually begin to take off as AI improves.
3. Robotics Is Rising Quick
Sundar admitted that Google was beforehand too early to robotics. AI has change into the lacking ingredient for concepts conceived 10 to fifteen years in the past. The Gemini Robotics fashions have reached state-of-the-art standing for spatial reasoning. Google has partnered again with Boston Dynamics and Agile and some different firms.
Most fascinating to me was the dialogue on Wing for drone supply.
“I feel we’re scaling up Wing the place in some cheap time interval, 40 million People can have entry to a Wing supply service. I’m not speaking years out or one thing like that.”
When requested if Google was going to do extra to construct {hardware}, Sundar mentioned having first-party {hardware} for robotics and AI can be vital.
“I feel we’d hold a really open thoughts. My lesson from Waymo and on the AI facet with TPUs, et cetera, I would like to actually push the curve properly, notably in areas the place you’ve gotten security, regulatory, all the pieces. You need the primary hand expertise of the product suggestions cycle. I feel having first celebration {hardware} will find yourself being essential.”
4. Agentic OpenClaw-Like Techniques Are The Future
There’s a purpose why OpenClaw (initially Clawdbot) went loopy viral a couple of weeks in the past. I nonetheless haven’t arrange an OpenClaw system as a result of I don’t really feel I do know sufficient about safety to make this method protected.
When Sundar was requested if one thing OpenClaw-like was coming from Google, he mentioned he thought it was the long run.
“I feel you wish to give customers functionality the place you’ve gotten persistent long-running duties in a dependable, safe means. It’s important to suppose via issues like identification, entry, et cetera. However I feel that’s the long run. That’s the agentic future. And bringing that for shoppers is a little bit of an thrilling frontier we’re . That is one in every of mine too.
I feel successfully the patron interfaces are going to have full coding fashions beneath, and the fitting harnesses and the fitting expertise and the flexibility to persist and run someplace safety within the cloud, regionally and within the cloud. All these primitives are coming collectively.
Right this moment I really feel like there’s 1% of the world, possibly not 1%, 0.1% of the world who’s residing this future. They’re constructing stuff for themselves, however bringing that to mass adoption. Sure. It’s a very thrilling frontier I feel.”
As I’m penning this, Google DeepMind has simply tweeted out directions for utilizing their new native open mannequin Gemma 4 with OpenClaw. A brand new means of speaking with our machines is beginning to unfold!
5. AI And AI Brokers Are Going To Enhance Dramatically In 2027
Sundar was requested when he thought it could occur that agentic programs would have the ability to work absolutely with no human within the loop. He mentioned twice that 2027 was more likely to be a giant 12 months.
“I positively count on in a few of these areas ’27 to be an vital inflection level for sure issues. Even the individuals doing it, that’s the workflow via which they’d produce it. Perhaps for a when you would verify it within the typical means, however you turn over, a crossover. However I count on ’27 to be a giant 12 months by which a few of these shifts occur fairly profoundly.”
The interview completed with Sundar speaking about what he was most enthusiastic about. He did point out that placing knowledge facilities in area was very thrilling, however this final bit was tremendous fascinating.
“I actually hung out yesterday with somebody who was explaining some enchancment in post-training, which is one individual speaking via the development they’re doing. Listening to it, I’m like, “Oh, it’s going to actually present up as a pleasant bounce.” That’s the fixed energy of this second. All of that, I don’t wish to be particular about the second, however we’ll publish it in the future I’m positive.”
It sounds to me like he’s speaking about agentic self-improvement.
We’re at the moment studying the way to have AI construct and do issues for us. I recall first studying to code with ChatGPT as a companion. It will give me code to stick into VS Code. Then I’d run it and paste the errors again into ChatGPT. We went backwards and forwards till one thing truly labored. I felt like I used to be pointless on this course of – the copying and pasting robotic, and positive sufficient, at present’s programs like Antigravity, Claude Code, and ChatGPT Codex run the code, verify the errors, and make things better up with out a lot want for human involvement.
It is smart to me that the subsequent step on this course of is to have AI programs study to enhance their usefulness with out us having to immediate them particularly. I count on that when this occurs, we are going to see even quicker development of AI capabilities and usefulness!
Extra Assets:
Learn Marie’s e-newsletter, AI Information You Can Use. Subscribe now.
Featured Picture: isasoulart/Shutterstock
