Create custom AI mini-apps with Gems
TL;DR
Google Labs is rolling out experimental Gems, allowing users to create or remix custom AI mini-apps like 'Recipe Genie' for specific everyday tasks.
## What changed Google Labs started rolling out experimental Gems on December 16, 2025. The feature lets users build or remix custom AI mini-apps for narrow tasks inside Gemini. Premade examples include Recipe Genie, which turns leftovers into meals, and Claymation Expert, which converts any topic into a claymation infographic.
Remix options and blank templates are available from the same interface. No paid plan is required to try the experimental version.
## Why it matters Vibe Builders gain a fast way to create reusable task agents without leaving the Gemini chat. This reduces the need for separate custom GPTs or multi-step Zapier flows when the job stays inside text and image prompts. The move pressures tools that rely on external orchestration by making simple specialization native to one model.
Google is betting that users will invest time in these mini-apps if they feel personal and quick to iterate. Success depends on whether shared Gems become common or stay isolated experiments.
## How to use it Open the Gemini app or gemini.google and navigate to the Gems section in the side menu or settings. Choose a premade Gem, select remix, or start from scratch by describing the exact task and output format. Test the result immediately in the same window and adjust instructions until the behavior matches your workflow.
Results appear in the regular chat thread. Export or share links work for basic distribution, though full editing rights stay with the creator.
## Watch for Public sharing volume and consistent output quality will confirm whether Gems stick. The bet weakens if Google adds usage caps or if responses drift after a few turns. The next logical step is a searchable directory or direct integration with Google Docs and Sheets.
Harsh’s take
For a solo operator running client work or content systems in 2026, Gems give a low-friction way to lock a specific prompt pattern into daily use. The real cost is lock-in: once you refine a Gem, moving the same logic to Claude or a local model requires starting over. Most builders will discover the output still needs heavy editing on anything beyond simple recipes or lists.
Do not treat this as a replacement for documented workflows in Notion. Build one Gem this week for your highest-volume task, run ten real examples, and measure how often you still rewrite the result. Drop it if the rewrite rate stays above 30 percent.
by Harsh Desai
About Gemini
View the full Gemini page →All Gemini updatesGo deeper
More AI news
- Daily RoundupGLM 5.3 and Vercel agent tools land, Gemini SAT practice, plus fresh models on Hugging Face
Google and Vercel released new agent and model features while Hugging Face and Replicate added trending models; OpenAI tightened safety after agent issues.
- Daily RoundupVercel-Origin beta and GPT-5.6 Sol discount, plus agent tools for builders
Vercel rolled out Origin repository deployments and a month-long GPT-5.6 Sol discount on AI Gateway while Product Hunt featured new agent layers and Hugging Face highlighted an uncensored Qwen model, with industry moves around data use and automation exits.
- Weekly DigestHermes Agent v0.20 voice and A2A protocol, OpenClaw mobile updates, and multi-agent tools to test now
Hermes Agent rolled out v0.20 releases with voice streaming, A2A interoperability, desktop artifacts, and self-recovery tools while OpenClaw added Android switching, iOS caching, and plugin provenance checks across the week.