Judgement is the last job

Instant building makes taste the bottleneck

Every year—or faster—a new paradigm reshapes how engineers and product builders work. For decades, implementation was expensive enough that teams poured huge effort into upfront speculation before touching real software:

“Because building was expensive, it took a lot of time, effort, and resources. So we ideated more and invested more in that phase.”00:00:39

That constraint has flipped. When a sentence can materialize a working prototype—something you can click, feel, and iterate on immediately—upfront guesswork gives way to empirical testing. On his Judgement slide, Dey arrays definitions from Mitchell Hashimoto, Karri Saarinen, Paul Graham, and Nan Yu’s claim that most builders do not have better taste than AI—warning that cheap output invites lazy acceptance:

“This is exciting. But it's also a little bit deceptive. It's easier than ever to just turn our brains off and blindly accept what's made for us.”00:00:39

Evidence: transcript 00:00:00, 00:00:39

Tools scale with intelligence

Working inside Google DeepMind close to frontier research lets the Antigravity team prototype next-generation interfaces and ship them internally to over 100,000 builders. The first pattern from that loop is simple: as models grow smarter, the ceiling on what products can accomplish keeps climbing.

Earlier models could already build startling artifacts—even whole operating systems—yet only inside tightly scripted guardrails:

“At the time, it was awesome. You create operating systems and really cool things we've never seen before. But it was also pretty prescriptive because the models needed at the time.”00:02:57

As models got better at communicating and coordinating with one another, teams could strip away rigid step-by-step choreography—leading directly to Antigravity’s native agent swarms:

“There's more agents working together, and we're able to relax the constraints to achieve more.”00:02:57

Evidence: transcript 00:02:12, 00:02:57

Rebuild foundations in the right window

Scaling with intelligence is deceptive because teams instinctively bolt new features onto old assumptions. Turn-by-turn prompting—queueing tiny file edits with spoon-fed context—made sense when models were weaker, but chokes modern long-running, asynchronous agents:

“Evolving with intelligence is a little bit tricky because being ambitious does not always mean adding more.”00:04:35

“Maybe a better pattern looks like. Treating the agent more as an entity that can make better decisions, and giving it user intent immediately, just like this.”00:04:35

Build too far in the past and your system is obsolete; build too far in the future and it is visionary yet impractical. Staying in the productive present means periodically tearing down scaffolding:

“To create the most value, challenge your past assumptions, and rebuild your foundations to create good systems that scale with intelligence.”00:04:35

Evidence: transcript 00:04:35

A platform for user ambition

When software stops forcing one deterministic path, users build their own leverage: designers letting agents generate hundreds of UI explorations before zooming in, personal dashboards tracking bugs and user feedback, team demo tools, and internal Google science plugins:

“Users can create their own leverage and make room for their own ambition.”00:06:36

Extensible surfaces also double as product discovery. Antigravity’s own /swarm feature started as an internal plugin on its extensibility layer before becoming core—alongside internal wins like YouTube video decoders and memory optimizations:

“Let your users be ambitious, and watch it feed back into your product. To create more value for everyone.”00:06:36

Evidence: transcript 00:06:36, 00:11:58

Don't accept the Rust rewrite

As software factories automate coding, testing, and deployment, what remains distinctively human? Dey illustrates the risk with a deceptively simple prompt: ask an agent to “make this slide better,” and it optimizes for maximal busywork—dashboards, visualizations, or rewriting the slideshow in Rust—unless human judgment defines what actually matters:

“If I ask an agent to say make this slide better. Maybe it'll add some visualizations and a bunch of dashboards and cool colors and everything. Maybe it'll rewrite the whole slideshow in Rust. So, it'll solve every problem. But does every problem need to be solved?”00:08:50

Redesigning Antigravity’s Agent Manager taught the same lesson: an agent asked to improve it would expose every raw tool call and add knobs everywhere. Human judgment subtracted noise—showing only semantic state to lower cognitive load so builders can steer multiple tasks calmly:

“Just because you're finished creating something first, which can be done in an instant. It doesn't actually mean you're done. Think about what you're actually trying to achieve and how the thing you're making feels.”00:08:50

“Agents might be doing more and more, but it's also more important than ever to solve real problems and use good prompting. Protect your judgment. And build things that people want. Thank you.”00:12:59

• Taste replaces implementation cost as the scarce resource: generate many directions fast, then refuse the first draft until the product feels right (00:00:39, 00:08:50).
• Rebuild foundations for intent: drop prescriptive turn queues as models learn peer-to-peer swarm coordination (00:02:57, 00:04:35).
• Make products extensible: let users build personal and team workflows whose best ideas graduate back into the core (00:06:36).

Evidence: transcript 00:08:50, 00:12:59