🚀 From Napkin Sketch to data-driven real-estate advisor Agent in Under 40 Minutes? 🚀

Can a coding agent really work autonomously on complicated problems without human intervention? I decided to put Google’s new Antigravity 2.0 and Gemini 3.5 to the test: The Napkin Challenge. 📝

The goal: Build and deploy a real estate investment advisor - based on real-world data, starting with nothing but a rough sketch on a napkin.

The Result? With the right context, yes!

I used:
Antigravity 2.0 & Gemini 3.5: Architect and execute the plan
Agent CLI Skills: Scaffold, build, test and evaluate the agent using ADK.3
Developer Knowledge MCP: Provided necessary context to connect to BigQuery MCP and integrate the census dataset for grounded investment advice.
Parallelized Workflow using sub-agents: While the system ran the evaluation suite, it simultaneously deployed the agent to Cloud Run.

The Outcome:

• Built and deployed in under 40 minutes.
• Zero human input beyond a napkin sketch (other than approving the plan and setting the model and region).
• 100% passing scores on evaluation cases, with responses delivered in under 30 seconds.
• A fully functional real estate advisor providing data-driven analysis on short and long term investments.

The "Napkin Challenge" proves that when you combine the right context with powerful models, the barrier between an idea and a deployed product has virtually disappeared.

I challenge you: What is your "napkin" project? Try it with Antigravity 2.0 and Gemini 3.5. 📝

🚀 How to join the challenge -

• Sketch an architecture or app idea on a literal napkin.
• Feed it to Antigravity 2.0 + Gemini 3.5. (I recommend making sure you have relevant skills in place for the task - context makes all the difference!)
• Drop a quick video or screenshot of your results on socials with the hashtag #NapkinChallenge.
• Add the link to your demo in the comments to this post!
• Tag 3 other folks and give them 48 hours to match the challenge! 👇 What will you build? Let me know in the comments!