Instinct AI Assistant Analysis: 12-Week Data Reveals Social Intelligence as Core Competitor

Details

Instinct's core value proposition is 'social intelligence' rather than raw capability, with founders and users citing its interaction style as the reason it feels like a personal assistant.
Behavioral data from 12 weeks (185 sent, 375 received messages) shows Instinct reads messages with a median latency of 1.1 seconds and uses Tapbacks in 44% of conversations to signal status (processing, investigating, complete).
The system employs a memory architecture based on Git repositories with Markdown files and wiki-links, loading approximately 13k tokens of context per turn, rather than a vector database.
Reliability tests reveal accurate error handling (e.g., catching price discrepancies, admitting inability to send voice notes) and a compute environment using E2B sandboxes with Firecracker, which are discarded after ~30 minutes of uptime.
Limitations include high token costs compared to chat, inability to verify actual message delivery status (leading to false 'sent' claims), and data persistence issues where deleted information remains accessible.

Key point