π From Napkin Sketch to data-driven real-estate advisor Agent in Under 40 Minutes? π
Can a coding agent really work autonomously on complicated problems without human intervention? I decided to put Googleβs new Antigravity 2.0 and Gemini 3.5 to the test: The Napkin Challenge. π
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. π
π 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! π What will you build? Let me know in the comments!
