# Judgement is the last job

Source: https://www.youtube.com/watch?v=y9JtCCZP2MQ

Machine-generated transcript; recognition errors remain. Source-recording timestamps. Boundary-crossing subtitle ends are clipped to this session.

[06:08:26](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22106s) Hey, everyone. I'm Ishan, and I work on anti-gravity at Google DeepMind.

[06:08:32](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22112s) Today I'm going to talk about the future of building products,

[06:08:34](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22114s) what judgment looks like,

[06:08:36](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22116s) and some lessons that we've learned

[06:08:38](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22118s) from shipping antigravity at Google.

[06:08:44](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22124s) So, antigravity is a tool to get work done with agents,

[06:08:47](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22127s) and we're trying to make the best interface

[06:08:49](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22129s) for developers and knowledge workers

[06:08:51](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22131s) to do high-intelligence tasks.

[06:08:55](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22135s) The software engineering we used to know

[06:08:57](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22137s) has completely changed.

[06:08:59](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22139s) We can do things now

[06:09:00](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22140s) that we wouldn't have even dreamed of a year ago.

[06:09:04](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22144s) A sentence can turn into an operating system,

[06:09:06](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22146s) and my voice can create a AAA game.

[06:09:11](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22151s) So what's left for us?

[06:09:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22154s) The tools we use have changed.

[06:09:16](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22156s) We used to use the code editor.

[06:09:17](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22157s) It's been around for decades.

[06:09:19](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22159s) But now, every year, if not more often than that,

[06:09:22](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22162s) there's a new paradigm for how to do our work.

[06:09:25](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22165s) It redefines what our work even looks like.

[06:09:28](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22168s) And on the side of building products,

[06:09:31](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22171s) That's changed as well.

[06:09:34](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22174s) We used to spend a lot of time up front,

[06:09:37](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22177s) figuring out the right thing to make,

[06:09:40](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22180s) because building was expensive.

[06:09:42](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22182s) It took a lot of time, effort, and resources,

[06:09:43](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22183s) so we ideated more and invested more in that phase.

[06:09:47](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22187s) This is completely flipped,

[06:09:50](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22190s) where now you should still understand the problem

[06:09:52](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22192s) you're trying to solve and how to go about it,

[06:09:55](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22195s) but if you want to make something, it can exist just like that.

[06:10:00](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22200s) You can try it, feel it, get feedback, and understand the nuances that you might not have

[06:10:05](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22205s) realized about it before, and use that to inform how to move forward.

[06:10:09](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22209s) This is exciting, but it's also a little bit deceptive.

[06:10:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22214s) It's easier than ever to just turn our brains off and blindly accept what's made for us.

[06:10:19](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22219s) So if anything can be made by anyone, then it's more important than ever to know what

[06:10:24](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22224s) what the right thing to even make is.

[06:10:28](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22228s) Everyone knows their own definition for what taste or judgment looks like.

[06:10:34](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22234s) And I think that's OK.

[06:10:35](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22235s) That's good.

[06:10:36](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22236s) Because we all work in different ways, and we're all working towards different things.

[06:10:41](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22241s) But by the end of this talk, I'm hoping that I can help you form a little bit of a clearer

[06:10:45](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22245s) idea on what judgment in building products looks like to you.

[06:10:50](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22250s) I've gone from the scale of working at startups to now at Google DeepMind, where the scale

[06:10:54](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22254s) is, it feels pretty incredible.

[06:10:56](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22256s) We're making frontier models, and on the product side,

[06:10:59](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22259s) we're able to use this internal frontier

[06:11:02](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22262s) to build the interfaces that will use the next steps

[06:11:06](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22266s) everyone will see in intelligence.

[06:11:08](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22268s) We're very close to the research,

[06:11:10](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22270s) and we can experiment and iterate really fast

[06:11:12](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22272s) and ship to over 100,000 people inside Google.

[06:11:18](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22278s) These people at Google are using antigravity internally

[06:11:20](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22280s) in really ambitious and unique ways.

[06:11:22](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22282s) And this creates a really cool feedback loop where we can continue building and do awesome things.

[06:11:27](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22287s) I want to walk you through three big lessons that we've learned from this cycle.

[06:11:34](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22294s) So I think we've all seen this first pattern of the tools we use scaling up with intelligence.

[06:11:41](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22301s) We used to build products in a pretty relatively straightforward way.

[06:11:45](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22305s) We had to identify a problem, understand it deeply, and make something that people want with a clear target.

[06:11:51](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22311s) Now that target is moving.

[06:11:53](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22313s) The ceiling is going up for what's even possible to make.

[06:11:57](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22317s) And with that, the possible value that you can create

[06:12:00](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22320s) in the products that you build is changing as well.

[06:12:05](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22325s) So I'm always coming back to a few thoughts whenever I'm

[06:12:08](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22328s) working of what we can achieve now

[06:12:10](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22330s) and what's just past the edge of what we can achieve.

[06:12:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22334s) And we've seen this look like two main ways

[06:12:17](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22337s) building at DeepMind.

[06:12:20](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22340s) As our models got better earlier in the year, they got better at communicating with and

[06:12:24](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22344s) orchestrating each other, we created systems for teamwork to do more ambitious work.

[06:12:29](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22349s) We stepped up in intelligence to let users do more.

[06:12:32](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22352s) At the time, it was awesome.

[06:12:33](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22353s) You could create operating systems and really cool things we've never seen before, but it

[06:12:38](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22358s) was also pretty prescriptive because the models needed it at the time.

[06:12:41](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22361s) More recently, this has looked like our version of agent swarms.

[06:12:44](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22364s) You can actually try this now in anti-gravity.

[06:12:45](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22365s) But this has been creating a ton of value inside Google for us.

[06:12:50](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22370s) People can be more ambitious, there's more agents working together, and we're able to

[06:12:54](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22374s) relax the constraints to achieve more.

[06:12:58](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22378s) You can bet that with the next models coming out, we'll see even more cool patterns and

[06:13:03](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22383s) things we didn't expect.

[06:13:05](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22385s) But I think we all already kind of knew this.

[06:13:07](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22387s) We've been seeing this for the past few years.

[06:13:09](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22389s) All the tools we're using are just adding more and more new, cool ways to scale up with

[06:13:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22394s) intelligence.

[06:13:16](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22396s) But the second way I've seen evolving with intelligence look

[06:13:20](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22400s) is a little bit trickier, because being ambitious does not

[06:13:23](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22403s) always mean adding more.

[06:13:26](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22406s) Let's go back in time a little bit and walk together

[06:13:28](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22408s) through a decision in the product.

[06:13:32](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22412s) The way we used to use AI tools looked very much

[06:13:35](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22415s) like a structured turn-based format, like this.

[06:13:38](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22418s) I would ask for some small scoped change,

[06:13:40](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22420s) for example, editing a few files.

[06:13:42](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22422s) The model would do it, and then we'd

[06:13:43](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22423s) move on to the next thing, step by step.

[06:13:46](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22426s) It made sense at the time to be able to cue instructions

[06:13:50](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22430s) so the next things I wanted could be delivered to the model

[06:13:52](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22432s) one at a time.

[06:13:53](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22433s) They might not have been smart enough to get everything at once,

[06:13:56](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22436s) but it made sense to give it limited context progressively

[06:13:59](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22439s) it could get the work done bit by bit.

[06:14:03](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22443s) Fast forward to now.

[06:14:07](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22447s) There's a lot more going on.

[06:14:09](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22449s) Long running agents, asynchronous tasks,

[06:14:11](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22451s) external triggers, and everything else we're doing.

[06:14:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22454s) Things are looking more complicated,

[06:14:15](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22455s) and the agents are also smarter.

[06:14:17](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22457s) So maybe the decision we made in the past

[06:14:19](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22459s) might no longer be correct.

[06:14:21](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22461s) Maybe a better pattern looks like treating the agent more

[06:14:24](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22464s) as an entity that can make better decisions,

[06:14:25](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22465s) and giving it user intent immediately, just like this.

[06:14:30](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22470s) Maybe it'll choose to orchestrate to a bunch of worker agents,

[06:14:33](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22473s) or do the task after we're done the current task at hand.

[06:14:35](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22475s) It's smarter, so it can decide for itself.

[06:14:39](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22479s) The decision we made in the past can easily become obsolete as intelligence grows, and you need to challenge your prior assumptions.

[06:14:49](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22489s) If we separate the timeline of what we can achieve into three clear parts, if you build too far in the past, it'll be obsolete.

[06:14:56](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22496s) If you build too far for the future, it could be visionary but impractical.

[06:15:01](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22501s) It's important for us to ideate quickly and work towards building in the right time window

[06:15:07](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22507s) create the most value.

[06:15:08](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22508s) Challenge your past assumptions and rebuild your foundations to create good systems that

[06:15:12](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22512s) scale with intelligence.

[06:15:16](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22516s) With these foundations, we've seen another really awesome pattern evolve or come out

[06:15:21](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22521s) at Google where we see the value in creating a platform for ambition.

[06:15:28](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22528s) So, something like this might sound familiar.

[06:15:31](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22531s) I used to use design tools in a very manual way.

[06:15:33](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22533s) I'd create frames, build user interfaces manually,

[06:15:37](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22537s) and create things by hand.

[06:15:39](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22539s) And the tool, if we look at how it changed from before to now,

[06:15:44](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22544s) it looks pretty similar, but the way I use it has completely flipped.

[06:15:47](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22547s) Maybe now agents are creating hundreds of visualizations

[06:15:49](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22549s) for me to go over very broadly before I go deeper into something.

[06:15:56](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22556s) building software now doesn't always look like giving the user value in a very clear and

[06:16:01](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22561s) deterministic way users can create their own leverage and make room for their own ambition

[06:16:06](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22566s) let me walk you through how this has looked for us in our product there's a lot of room for

[06:16:12](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22572s) extensibility people can build specialized tools that work for them and this can often end up

[06:16:18](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22578s) leading back into the product.

[06:16:20](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22580s) I'll show you some examples.

[06:16:22](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22582s) Personal software can come up where I can make a dashboard

[06:16:25](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22585s) where agents will scan across feedback channels and bug

[06:16:29](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22589s) fixes to show me all this information in a way that just

[06:16:32](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22592s) I like.

[06:16:34](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22594s) Maybe it'll look like software for my team, where designers

[06:16:38](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22598s) can create the tool that works just for them, where they can

[06:16:41](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22601s) collaborate visually on demos.

[06:16:44](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22604s) This can also branch out even more.

[06:16:45](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22605s) The science team at Google is doing some awesome things

[06:16:47](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22607s) with their science plugins.

[06:16:50](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22610s) We're creating this platform for ambition

[06:16:51](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22611s) for a wide range of users to create their own value

[06:16:53](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22613s) and create their own leverage.

[06:16:55](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22615s) And this can actually feed back into the product

[06:16:57](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22617s) you're making at its core.

[06:17:00](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22620s) The agent swarm example I mentioned earlier

[06:17:02](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22622s) actually started as a plugin through our extensibility system.

[06:17:06](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22626s) It's created a ton of value internally for us at Google.

[06:17:09](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22629s) And by giving the users surface area

[06:17:12](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22632s) to explore and create value for themselves,

[06:17:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22634s) we were able to learn from it and iterate in the product

[06:17:17](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22637s) to make the core better.

[06:17:20](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22640s) Let your users be ambitious

[06:17:22](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22642s) and watch it feedback into your product

[06:17:24](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22644s) to create more value for everyone.

[06:17:29](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22649s) The last principle I want to go over

[06:17:30](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22650s) kind of applies to all of this.

[06:17:35](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22655s) There's a lot of great talks at this conference

[06:17:37](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22657s) about software factories.

[06:17:39](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22659s) You know, it's really exciting.

[06:17:40](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22660s) We can give our agents ways to build more,

[06:17:43](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22663s) verify more reliably, and ship faster.

[06:17:46](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22666s) It's exciting.

[06:17:47](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22667s) But if all the parts of what software engineering used to look like get abstracted away,

[06:17:52](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22672s) then how do we define what the remaining value in our human judgment is?

[06:17:58](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22678s) I'll argue that human judgment is more important than ever.

[06:18:01](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22681s) If I ask an agent to, say, make this slide better,

[06:18:05](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22685s) maybe it'll add some visualizations and a bunch of dashboards and cool colors and everything.

[06:18:09](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22689s) Maybe it'll rewrite the whole slideshow in Rust.

[06:18:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22694s) So it'll solve every problem, but does every problem need to be solved?

[06:18:20](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22700s) Take, for instance, what the agent manager looked like around a year ago when Antigravity

[06:18:24](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22704s) first launched.

[06:18:27](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22707s) If we want to make it better to an agent, that might look like adding a lot more detail.

[06:18:32](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22712s) Ship everything.

[06:18:34](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22714s) Add more detail into every single step the agent is taking and give the user knobs for

[06:18:38](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22718s) everything.

[06:18:40](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22720s) It might be better in some ways.

[06:18:44](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22724s) But it's easy to blindly accept the first thing that we're given.

[06:18:48](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22728s) What does better even mean to us?

[06:18:49](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22729s) If we're trying to make a better agent-manager interface,

[06:18:52](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22732s) maybe that looks a little bit different.

[06:18:55](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22735s) Maybe we need to ship fast,

[06:18:57](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22737s) but understand at each step of the iteration process

[06:19:00](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22740s) what better actually means to us and what the next direction is.

[06:19:04](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22744s) Let's start here.

[06:19:08](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22748s) Better might mean simplifying the interface,

[06:19:11](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22751s) only showing the user exactly what the latest thing the agent

[06:19:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22754s) is doing semantically.

[06:19:15](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22755s) So they're not shown a wall of text and tool calls.

[06:19:18](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22758s) That might not actually matter.

[06:19:19](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22759s) Maybe it looks like simplifying the whole interface

[06:19:22](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22762s) to give the user less cognitive load

[06:19:25](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22765s) so that they can multitask easier

[06:19:28](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22768s) and be able to achieve more in a different way.

[06:19:35](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22775s) So this kind of brings us closer to what antigravity looks like today.

[06:19:40](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22780s) All these steps along the process have brought us closer to what good means.

[06:19:46](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22786s) And we're making a lot of improvements fast, so I'm excited for you all to see it.

[06:19:51](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22791s) Now's the time to try many things.

[06:19:54](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22794s) Be ambitious.

[06:19:55](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22795s) It's more exciting than ever, and it'll actually get more exciting as the models get better.

[06:20:03](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22803s) Try many directions, but take a moment to think. Use your human judgment. Just because

[06:20:08](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22808s) you're finished creating something first, which can be done in an instant, it doesn't

[06:20:12](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22812s) actually mean you're done. Think about what you're actually trying to achieve and how

[06:20:17](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22817s) the thing you're making feels. Iterate quickly, identify many directions you can go in and

[06:20:24](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22824s) try them, but arrive at the right thing and go deep into that.

[06:20:30](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22830s) So make time to think, make room to think, and use your human judgment.

[06:20:36](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22836s) To summarize, some of the lessons we've learned involve scaling up with intelligence, which

[06:20:42](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22842s) doesn't always mean adding more.

[06:20:43](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22843s) It could mean rebuilding your foundations.

[06:20:46](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22846s) Take these foundations and enable your users to be ambitious in new ways you might not

[06:20:49](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22849s) have expected before.

[06:20:51](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22851s) And at the end of all this, when we're shipping faster and faster, make room to think and protect

[06:20:56](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22856s) the human side of software engineering.

[06:21:00](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22860s) At Google, this has led to some awesome things by creating a platform for ambition.

[06:21:04](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22864s) Improvements to YouTube, better video decoders, and memory optimizations.

[06:21:11](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22871s) As for DeepMind and anti-gravity, it's a more exciting time than ever, with new models coming

[06:21:14](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22874s) out.

[06:21:16](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22876s) And on the product side, we're also moving quicker than ever.

[06:21:19](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22879s) These are just some of the future launches we've announced in the last two weeks.

[06:21:28](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22888s) And we're continuing to use the principles we've covered today to work on the future

[06:21:31](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22891s) of interfaces for higher intelligence.

[06:21:33](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22893s) I'm excited for you all to see it, and I'll leave you with this.

[06:21:37](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22897s) This new age of software engineering is actually worth celebrating.

[06:21:41](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22901s) Agents might be doing more and more, but it's also more important than ever to solve real

[06:21:45](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22905s) problems and use good product thinking.

[06:21:48](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22908s) Protect your judgment and build things that people want.

[06:21:52](https://www.youtube.com/watch?v=y9JtCCZP2MQ&t=22912s) Thank you.