Radish Blog
What we are building, learning, and shipping
Product notes, engineering write-ups, and the decisions behind the assistant we are putting into the world.
How to effectively automate lead research in the AI world
Why public web Deep Research falls short for sales and investor research - and how Radish closes the gap with grounded, multi-source person and company enrichment.
Permissions: How we make sure your assistant does not go rogue?
How Radish keeps AI assistants from going rogue — OAuth scopes you control, tool-level Ask/Auto permissions, and a permission layer the agent can't bypass.
Why We Built a Package Management System for Skills
Radish Skills aren’t static prompts; they’re versioned packages with dependencies, constraints, and install/upgrade flows.
The AI Pendulum: Why the move to the Desktop is a Step Backwards
Why the trend of running AI agents on local servers is a hurdle for true AI democratization and how cloud-native orchestration wins.
3x cheaper, 2x better: Radish vs Claude Cowork for lead gen
A head-to-head benchmark on a real lead-gen task: finding verified LinkedIn profiles for regional US airline leadership teams—measuring valid links and cost.
ChatGPT vs. Radish for Language Learning
ChatGPT can explain a language. Radish can help you actually learn one — with structure, audio, progress tracking, and workflows that continue across the tools you already use.
Your AI Assistant Is Only as Good as Its Skills
Most AI tools stop at answering questions. Radish goes further — with Skills that give your agent the ability to do things across the tools you already use.
The Last Knowledge System You Will Need is an Agent
Stop organizing your notes. Connect your messy files to a Radish agent and let AI handle search and recall — via Telegram, instantly.
Agent Skills Ecosystem Map: A Data-Driven Look at 25M Workers and 50K skills
AI tooling is booming but only for developers. We measured the democratization gap across 25M workers and found most professions locked out of the composable AI ecosystem.
Gemini undocumented limits for strongly typed schemas
Gemini silently fails when your tool schemas get too complex. We found the undocumented limits and how to work around them
A 3-step fix for the persistent MALFORMED_FUNCTION_CALL error in production agents
How three layered fixes brought our Gemini MALFORMED_FUNCTION_CALL production error rate down to near zero.