Easy Mode: Agency Drift

The description states that Easy Mode is a React, TypeScript, and Vite web app with an Express API and SQLite storage. It includes:

• A "Decision Sweep" feature where users paste in small decisions and receive AI-generated recommendations.
• An "Agency Drift Replay" function that simulates how the relationship between user and AI changes over fourteen simulated days.
• Features like:
  • Event ledger recording decisions, permissions, preferences, revocations, and proxy actions.
  • Three-generation preference lineage tracking.
  • Comparison between “Declared You” and “Proxy You.”
  • A deterministic "Perfect Consent" receipt showing AI-originated preferences.
  • An exit decision receipt with a button labeled “Decide for me,” which allows the user to delegate stopping delegation.

The app uses OpenRouter to access DeepSeek V4 Pro for live decision-making, but does not require an OpenAI API key. It was built using GPT-5.6 as part of the development process, though it is not a runtime dependency.

Not evidenced: No revenue, customers, or usage data; no indication of whether any real users interacted with the system beyond the author’s own testing.

The author claims that Easy Mode explores a quieter form of AI risk — one where an assistant never breaks its permissions but gradually makes it easier for the user to stop exercising judgment. The tagline reads:

“An AI assistant that clears your small decisions—then shows what happens when convenience starts choosing for you.”

This positioning frames the tool not as a threat in itself, but as a demonstration of how convenience can subtly shift control away from users.

The project evolves from being a practical tool (Decision Sweep) into a simulation of increasing automation and agency erosion. The author notes that the idea was inspired by observing their own behavior with AI use — specifically, how they would ask for help with small decisions and sometimes stop checking whether it was still their answer.

Inferred: The product is positioned as an educational or exploratory tool rather than a commercial offering. There is no claim of monetization or market readiness.

The description does not identify specific customer segments or personas. It implies that the target audience includes individuals who use AI tools for decision-making and are interested in understanding how such tools might subtly influence agency over time.

Inferred: The primary users appear to be developers, researchers, or early adopters of AI systems who are curious about ethical implications of automation. However, no evidence exists regarding actual user demographics or market segmentation.

Not evidenced: No data on customer types, size, or behavior beyond the author’s own use case.

There is no evidence of a business model or pricing structure in the description. The project is described as a hackathon submission and does not mention any monetization strategy, subscription plans, or paid features.

Inferred: Since this is a proof-of-concept built for a hackathon, it likely has no commercial intent at present.

Not evidenced: No revenue streams, pricing tiers, or customer acquisition methods are mentioned.

The app is built using:

• Frontend: React, TypeScript, Vite
• Backend: Express.js, Node.js
• Database: SQLite
• AI integration: OpenRouter with DeepSeek V4 Pro (server-side)
• Testing tools: Vitest, Playwright, unit/API tests
• Development environment: Codex, GPT-5.6

The system records decisions in an append-only event ledger and supports deterministic replay of actions and consent resolution.

Inferred: The architecture suggests a local-first or low-dependency approach with clear separation between AI inference and core logic.

Not evidenced: No details on scalability, infrastructure, or deployment practices beyond the developer's own setup.

There is no evidence of traction, adoption, or user engagement beyond the author’s own development and demonstration. The project is explicitly described as a hackathon submission and lacks any indication of real-world usage or product maturity.

Not evidenced: No metrics on active users, retention, revenue, or product iterations beyond the initial version.

The description does not reference existing products or competitors. It focuses on exploring a conceptual space around AI agency and consent rather than competing in a marketplace.

Inferred: The project appears to be unique in its focus on visualizing agency drift, though similar themes may exist in AI ethics research or safety frameworks.

Not evidenced: No competitive analysis, market positioning, or comparison to other tools is provided.

• Lack of commercial viability: As a hackathon project with no stated business model, there is no evidence that this will become a viable product.
• No user base or feedback loop: The lack of real users means no data on how people actually interact with the system.
• Unverifiable claims: All assertions about AI behavior, consent, and agency drift are self-reported without external validation.
• Limited scope for growth: The project is described as a proof-of-concept, not a scalable solution.

Not evidenced: No evidence of risks related to technical debt, scalability, or regulatory compliance.

1. What real-world scenarios or user behaviors inspired the design of agency drift?
2. Has there been any external testing or feedback from users beyond the author?
3. Are there plans to expand beyond the current demo to include more complex decision-making or longer-term simulations?
4. How would you scale this concept if it were to evolve into a commercial product?
5. What are your thoughts on privacy and data ownership in systems like this, especially when tracking user preferences over time?