SendIt

text claude to make decisions for you

Comment

Inspiration

Decision fatigue is real in college, every choice feels overwhelming when you're balancing fun and responsibilities. We wanted to create an AI companion that doesn't just give you the "right" answer, but shows you both sides of every decision through a fun angel vs devil debate, then helps you find the balance that actually fits your life right now.

What it does

SendIt turns decision-making into a three-way conversation delivered via iMessage. When you text a question like "should I go out tonight?", two AI agents, an angel and a devil, debate the pros and cons in real-time, considering context like time of day, your tone, financial implications, and urgency. A judge then synthesizes their arguments into personalized advice that balances spontaneity with responsibility, helping you make decisions that feel right for your actual situation.

How we built it

We built SendIt using TanStack Start (React) with an iMessage-style UI, powered by Claude AI models (Haiku for the debaters, Sonnet for the judge) through Anthropic's AI SDK. The backend uses server functions to stream debate messages in real-time, while a context engine analyzes each question for financial keywords, urgency signals, emotional tone, and temporal factors. We store debate history in SQLite with TanStack Query for seamless data management, and deployed the whole system to handle SMS/iMessage integration for on-the-go decision support.

Challenges we ran into

Getting the AI agents to actually debate rather than just agree was surprisingly tricky—we had to carefully engineer prompts that encouraged genuine back-and-forth while keeping responses under 30 words for authentic texting feel. The context-aware system also required extensive testing to ensure it could accurately detect urgency, financial implications, and emotional tone from brief text messages. Streaming responses while maintaining message order and adding realistic typing delays for the iMessage experience added complexity to the real-time architecture.

Accomplishments that we're proud of

We're proud of creating AI personalities that genuinely feel like friends in a group chat, not robots—the angel and devil have distinct voices that make debates entertaining while still being helpful. The context engine intelligently parses questions without any external data sources, extracting time-of-day awareness, financial implications, and emotional urgency from text alone. Having the agent run in iMessage gives authentic messaging experience as AI feel natural and fun.

What we learned

We discovered that good AI personality design is about constraints—limiting responses to 30 words forced more punchy, authentic dialogue than unlimited text ever could. Context matters enormously for advice; the same question at 2am on a Tuesday versus 8pm on Friday should get radically different responses. We also learned that users want to see the reasoning process, not just the answer—the debate format builds trust by showing both perspectives before synthesizing advice.

What's next for SendIt

We're planning to add persistent user profiles that learn from past decisions and outcomes (did you regret staying in last Friday?), making advice increasingly personalized over time. We want to expand context awareness with calendar integration, weather data, and location services to give even smarter recommendations. We're also exploring group debate modes where multiple users can get collective advice, and building a reflection feature that checks in the next day to see how your decision played out, helping SendIt learn what actually works for you.

Built With

Share this project:

Updates