Ideas Radar: Mar 22, 2026

A unified stock and crypto portfolio dashboard with live news, market moves, AI-driven predictions, and clear buy/sell/hold actions — all in one view. Every existing tool treats information display and actionable intelligence as separate products. The gap is in the synthesis layer. Over 15,000 impressions with strong engagement.

No-KYC stock trading app for on-chain assets. As stocks move on-chain, there's demand for a simple trading interface where normal users just tap buy or sell while settlement happens invisibly on-chain in the background. The key insight: make the crypto rails invisible to the end user.

A "Strava for AI" — an activity chart like GitHub's contribution graph, but for aggregated token usage across all AI tools. AI-native workers would show how hardcore they are; normies would demonstrate they're learning to avoid getting fired. Could evolve into a social platform around AI usage patterns.

A multi-model AI agent that dynamically routes to hundreds of specialized models for different tasks, including local models. Called for by the CEO of the largest AI model hub. The infrastructure exists via inference providers, but no one has built the routing + orchestration layer that makes this seamless for end users.

A formal proof reconstruction benchmark: progressively strip proofs from verified systems and measure how much AI can rebuild from scratch. High value for the AI safety and formal verification research community. Over 6,000 impressions from a niche but influential audience.

MCP servers for banks — enabling AI agents to directly query bank statements and automate tax filing. The banking API landscape is fragmented, but the demand for AI-powered financial operations is massive. Over 1,800 impressions with high engagement.

An AI agent identity verification system. As autonomous agents proliferate, there's no standard way to verify that an agent is what it claims to be. Think SSL certificates but for AI agents — a trust infrastructure layer that doesn't exist yet.

A tool that monitors what AI models say about your product when users query ChatGPT, Perplexity, or other AI assistants. GEO (Generative Engine Optimization) analytics is an emerging category with almost no tooling. Over 11,000 impressions — founders need this yesterday.

A platform that does daily AI benchmark testing across models. Users report Claude performance regressions of 9%+ with no public accountability. A neutral, continuous benchmarking service would serve the entire AI ecosystem. Over 2,500 impressions.

An Apple Silicon-optimized local AI development environment. As more developers run models locally on M-series Macs, there's a growing gap in tooling specifically designed for Apple's hardware. Over 7,700 impressions with 63 likes from a single tweet.