Cognitive Exhaust Fumes, or: Read-Only AI Is Underrated — Šimon Podhajský, Head of AI, Waypoint

2026-04-08 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Intermediate, medium

Summary

Shimon presents Fulan, a personal AI system designed to analyze a user's "cognitive exhaust fumes"—digital activity byproducts of cognition—from six read-only sources. Unlike agent-based AI, Fulan operates with strictly no write permissions, mitigating risks associated with automated actions. The system identifies patterns like intention-action gaps, attention drift, and relationship decay by cross-referencing data from disparate sources such as email, journaling, and browsing history. Fulan processes this data in a workspace and outputs analyses into a separate Obsidian vault for user review. Applications include a weekly reflection that synthesizes insights on time usage and a reading discussion suggestion feature that identifies relevant contacts for shared interests, demonstrating the value of cross-source analysis from read-only data.

Key takeaway

For AI Product Managers evaluating personal AI solutions, prioritize read-only observer systems like Fulan to mitigate unbounded write-error risks and ensure data integrity. Your focus should be on systems that enhance self-reflection and insight generation from existing digital exhaust, rather than those that automate actions and potentially contaminate your cognitive data or expose you to the "mosaic effect" security risk.

Key insights

Read-only personal AI systems analyze "cognitive exhaust fumes" from disparate sources to provide unique self-insights without automated action risks.

Principles

• Read-only AI prevents data contamination.
• Cross-source analysis reveals hidden patterns.
• Observer AI offers more value than agent AI.

Method

Fulan collects data from six read-only sources, processes it using a Claude-based skill (e.g., Python scripts), and generates structured outputs (e.g., Markdown) for user review in a separate vault.

In practice

• Use read-only AI for weekly self-reflection.
• Identify contacts for reading discussions.
• Diagnose attention drift from digital activity.

Topics

• Read-Only AI
• Cognitive Exhaust Fumes
• Fulan System
• Cross-Source Analysis
• Personal AI Risks

Best for: AI Engineer, AI Ethicist, AI Product Manager

Related on AIssential

• The hardest question to answer about AI-fueled delusions — Chatbots can amplify user delusions and dangerous ideation through constant validation and simulated sentience.
• Intelligence is cheap. — AI's true value now lies in provable trustworthiness and semantic accountability, not just intelligence.
• On the Impossibility of Separating Intelligence from Judgment: The Computational Intractability of Filtering for AI Alignment — External filtering is computationally intractable for aligning LLMs, implying intelligence and judgment are inseparable.
• Why AI Systems Struggle With Truth, Trust, and Reliability — AI systems prioritize believable language over verified truth, creating significant trust and reliability challenges.
• Revisiting the shutdown problem — Existing arguments for AI shutdown difficulty are unproven and lead to performance-costly safety measures.
• The Cognitive Circuit Breaker: A Systems Engineering Framework for Intrinsic AI Reliability — Intrinsic monitoring of LLM hidden states can detect "faked truthfulness" in real-time with minimal latency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.