Gemini 3.5 Flash 🤝 @Antigravity Watch how the model deploys multiple subagents to design and build an entire city.

May 19, 2026 · 7:58 PM UTC

39
80
14
1,007
81,195
RelevantRecentLikes
Find out more → goo.gle/3PwoinZ
Gemini 3.5: frontier intelligence with action

Gemini 3.5: frontier intelligence with action

At Google I/O we released Gemini 3.5, our latest series of models combining frontier intelligence with action.

blog.google
1
13
1
64
20,843
Everyone's eyeing the city layout, but the real win's the subagent orchestration. If it can't handle conflicting constraints between agents, it's just a fancy simulation.
4
6
209
wow, this is cool af.
1
133
Sick multi subagents demo
85
Antigravity is cool
273
teapawt.net/gerardsans/status/2053…
Gerard Sans | Axiom 🇬🇧 (@gerardsans) profile pictureGerard Sans | Axiom 🇬🇧
@gerardsansMay 9
Reaching 2M impressions on X! Thanks everyone for keeping it real against AI hype! Last 3 months have been a blast. Answering rumours: • No, I’m not an AI. • Realist: neither pro nor anti. • AI research: mention me on any paper for my analysis Trailblazing the manifold.
Image from Gerard Sans | Axiom 🇬🇧 (@gerardsans)
90
Used subagents to design our entire go-to-market structure last quarter. Main agent assigned territories, built the comp plan, hired 6 SDRs, and allocated budget. Our CFO found out when payroll ran. She said 'none of these people are real.' They are not real. The quota attainment is still impressive.
3
602
watching subagents coordinate to design and build an entire city in real time is the kind of demo that makes you put your phone down and just stare. this is the Google that people kept saying was sleeping. turns out they were just loading 🤭
2
769
New UI is 🤮
1
115
The interesting part isn’t even the benchmark jump anymore. It’s that models are starting to behave less like “chatbots” and more like operating systems coordinating specialized agents in parallel. We’re slowly moving from prompting intelligence → directing intelligence.
1
741
If nobody spends tokens does antigravity go away like Wave?
268
The coordination layer is kinda the whole bet. I'm curious how it handles rollback when one subagent's output breaks another agent's constraints.
1
408
To actually achieve this how much token usage is needed really ? I mean I assume you would hit limits fast with all these sub agents ?
1
569
multi-agent orchestration at this fidelity is the real unlock — single-model bottlenecks are the bigger constraint than raw parameter count.
3
Please retain Gemini CLI
2
hey ， like this
Image from teze (@fooyoon)
51
Gemini 3.5 flash+ code execution tools = god tier , in ai studio it auto test and do auto benchmark its amazing and efficient
92
Love google
66
hmmm? nie ogarnia
83
AI is moving WAY faster than most people expected.
46
Incrível de verdade! Muitos ainda não se deram conta da grandiosidade e poder do Gemini 3.5 Flash.
1
73
Subagents are where the demo starts to resemble real work. The operator question becomes coordination: who owns the plan, how conflicts get resolved, what evidence each subagent returns, and how the final result is judged before anything hits production.
26
Design and build an entire city ? Cool
189
Looks like Google fixed housing problem.
126
"What if we shoved AI into every aspect of your fucking lives!"
266
im sorry but this example is so bad, no? like its generating almost nonsense. im sure the tool is usefull for something ... but this?
248
We went from autocomplete to building cities. The jump in capability this year alone is hard to process
74
Subagent orchestration for complex spatial tasks is where things get genuinely interesting. Not "AI writes code" interesting, but "AI models understand dependencies and parallel workstreams" interesting. That's a different level of reasoning.
179
Please resolve the limits issue. The project can't be completed in three hours. Tokens are disappearing faster than people expect. Compare this to the Codex. @GoogleDeepMind
111
Was hopeful for antigravity but lack of extensions (like giving me access to Claude code) make it useless for my needs. Downloaded update, struggled as the help gave outdated advice, then deleted.
328
teapawt.net/aiofficetr/status/2057…
AIOfficeTR (@aiofficetr) profile pictureAIOfficeTR
@aiofficetrMay 20
Google DeepMind, "her şeyden her şeyi yaratabilme" vizyonunun ilk adımı olan yeni modeli Gemini Omni’yi duyurdu. Çok modlu zekada dev bir sıçrama kapımızda! İşte detaylar: 👇 🎬 Gemini Omni, geleneksel video araçlarının çok ötesinde. Videodaki karakteri kusursuzca değiştirebiliyor (Swap Character), sahnenin tarzını (anime, 3D, kukla) saniyeler içinde dönüştürüyor ve çevre unsurlarını sıfırdan inşa edebiliyor. Yapay zeka artık fizik kurallarını ve derinliği çok daha iyi anlıyor! 💡 AI Office Türkiye ekibi olarak her zaman vurguladığımız gibi: "Öğrenilen değil, kullanılan yapay zeka" dönemi tam anlamıyla başladı. Üretim süreçlerini tamamen değiştirecek Gemini Omni'nin tüm detayları ve özel analizimiz web sitemizde! #omni #google #ai aiofficeturkiye.com
45
Sorry, but splitting the old application into two separate ones wasn't a great move. Installing both Antigravity 2.0 and Antigravity IDE now just gives me exactly what we had before in a single application.
1
174
3.5 flash it's impressive but, hey! I was never reaching the limit for the previous model. Now I did it in record time. (Plus user)
224
can’t wait for the subagents to start unionizing
121
Load more replies