Built This Week/AI Tools

AI Tools

Gemini

Gemini is our pick when an app needs AI built in or has to read a lot of data, and guests praise its speed and cost, but for coding we have kept reaching for other models.

Our verdict on Gemini

We use Gemini most when an app needs AI inside it, or when the input is too big for a smaller model. Through Google AI Studio it powered seven of our builds, from a finance dashboard to a talking Halloween mirror. For writing code, though, Jordan kept saying through 2026 that other models were ahead, and that you have to test each new Gemini release yourself.

Our guests are more upbeat than we are on speed and cost:

  • Speed. Anand, whose company runs Gemini through Vertex AI in production, said Gemini 3 and Gemini 2.5 Flash respond extremely fast.
  • Cost. Carina said Gemini 3 is much cheaper to run than GPT. David Petrou said in September 2026 that Gemini 3.8 Flash looks really good on cost versus quality.
  • Marketing content. Sagi Waitzman's team writes LinkedIn posts and images with Gemini, partly because it comes with Google Workspace.
  • Multi-model products. Emblem's August Kiles said Claude and Gemini are the two models they use most in their product.
And the speed at which it responds, especially Gemini three and the Gemini 2.5 flash, they are just too good.

— Anand Chandrasekaran, Episode 24

What we built with Gemini

  • Solopreneur finance dashboard (September 2025): built for Hooman Radfar's Collective. Gemini reads an uploaded P&L and balance sheet and writes next-month, next-quarter and next-year projections, an insurance compliance check and a lending risk report. Jordan admitted the overall AI insights panel may not be working that well.
  • Weather and time app (October 2025): one prompt in AI Studio, built in seconds. Sam doubted the 82-degree Nashville reading.
  • Mirror, Mirror on the Wall (October 2025): speech-to-text sends your question to Gemini, which answers through text-to-speech. It was slow and laggy, and it said Microsoft, not Nvidia, was the most valuable company.
  • Guess My Costume app: a webcam app that correctly identified Jordan as a vampire.
  • Trick-or-Treat route planner: for Nashville it picked Sylvan Park, which Sam, a former resident, called pretty much spot on.
  • LLM Math Roaster (November 2025): Gemini 2.5 Pro was one of four models writing Lean 4 proofs. The ChatGPT judge gave Gemini the top score, 98, on Fermat's Little Theorem.
  • AI start-of-care packet generator (December 2025): Sam pasted a PRD into AI Studio with Gemini Pro and had a working prototype after four or five tweaks in about 30 minutes.

How we use Gemini

  1. Start in Google AI Studio's Build mode. Jordan's argument in Episode 19 (at 13:30) is that Gemini is already integrated, so you skip the backend, API choice and key setup you would need in Bolt or Lovable.
  2. Bring a PRD. Sam got his from ChatGPT, pasted it into AI Studio and iterated from there.
  3. Click the suggested features. While building, AI Studio proposed weather alerts for the weather app, a difficulty setting and timer for the costume guesser, and real-time weather for the route planner. Picking a few is the fastest way from v0 to v1.
  4. Swap to Gemini when the data is big. On the Collective dashboard, Jordan's first model could not take in the financial statements, so he moved those calls to Gemini.
  5. Write the app elsewhere if you need more control. That dashboard started in Bolt, then moved to Cursor with OpenAI Codex for the front end, with Gemini only behind the API.
  6. Use Gemini Pro for research and screen help. In Episode 38 (at 13:50) Jordan showed sharing his screen in AI Studio so Gemini can help with what he is working on.
  7. Don't trust an LLM judge alone. Axiom's Carina told us to compile the Lean proofs instead of letting ChatGPT score them.
I needed a bigger model, and so in fact, I moved to Gemini, and I'm using Gemini here because I was able to take the data in.

— Jordan Metzner, Episode 14

Where Gemini falls short

  • Coding, for us. When Gemini 3 launched in November 2025, Jordan burned through his tokens in Google Antigravity and Cursor and liked it. Sam hit hiccups building a meditation app. By January 2026 Jordan said Anthropic's models were still the best way to build, and that we would move over if Gemini became top notch.
  • Not quite there for engineering work. Anand Chandrasekaran said in Episode 24 that his team can do some coding with Gemini 3, but it is "still not yet there."
  • A thinner product. Amar Goel of Bito said the Gemini app can't make a PDF the way ChatGPT can.
  • Wrong answers in live demos. The mirror got the most valuable company wrong on air.
I personally haven't felt like Gemini three is like so groundbreaking per se. I know on the benchmarks it's done really well, but I do think it's quite a good model.

— Amar Goel, Episode 23

Gemini compared

  • ChatGPT vs Gemini: Gemini won our math proof test and is cheaper to run per Axiom; ChatGPT has the richer product.
  • Claude vs Gemini: Claude for coding in our builds; Gemini Flash on cost versus quality.
  • Gemini vs Grok: Jordan switched from a smaller model to Gemini when the financial data got too big.

Episodes featuring Gemini

  • Episode 14: Gemini analyzes P&Ls for the Collective dashboard.
  • Episode 18: the weather app built live in AI Studio.
  • Episode 19: three Halloween apps, and why built-in Gemini matters.
  • Episode 22: the Math Roaster and first takes on Gemini 3.
  • Episode 24: Sam's care-packet prototype and Arya Health on speed.
  • Episode 29: why we still code with Anthropic's models.
  • Episode 43: Jordan plans to test Gemini 3.5 Flash in Antigravity and says to trust your own tests over the charts.

FAQ

Is Gemini good for coding?

Jordan was impressed with Gemini 3's coding and design taste when it launched in November 2025, but in January 2026 he still said Anthropic's models were the best way to build. Arya Health's Anand Chandrasekaran said Gemini 3 can do some coding work but is not there yet.

Gemini vs ChatGPT: which is better?

In our LLM Math Roaster, the ChatGPT judge gave Gemini 98 on Fermat's Little Theorem versus 70 for ChatGPT. Bito's Amar Goel said Gemini 3 sometimes gives better answers, but the product is not as rich as ChatGPT.

How do you build an app with Gemini?

The quickest way we found is Google AI Studio's Build mode, where Gemini is already wired into the app. Jordan built a talking mirror, a costume guesser and a trick-or-treat route planner there in one episode.

Is Gemini fast?

Yes, according to our guests. Anand Chandrasekaran said Gemini 3 and Gemini 2.5 Flash respond extremely fast, though our own talking mirror felt a little slow and laggy.

Why use Gemini for large documents?

Jordan started his solopreneur finance dashboard on a smaller model, but the P&Ls and balance sheets were too much for it. He switched to Gemini because it could take the data in.