Flight Recorder

Project: Flight Recorder

Self-reported basis: The description is entirely from the author’s own submission to the OpenAI 2026 hackathon on Devpost. No external verification or historical data is available.

What it appears to be: A governance and control layer for AI agents in enterprise environments, positioned as a "black box" that sits between an agent and company systems to manage risk through policy checks, human approval gates, and evidence generation.

What changed: The project emerged from the recognition that as AI agents move beyond answering questions to taking actions (e.g., sending emails, updating CRMs, deploying code), enterprises require governance before execution and proof afterward.

Most important open question: Does the product demonstrate a real market need for agent action governance, or is it an experimental prototype with unclear commercial viability?

The description states that Agent Flight Recorder is a governance layer for AI agents, positioned between the agent and company systems. It captures proposed actions, evaluates them against policy, and either allows, blocks, or routes them to a human approver.

• The core flow involves:
  • Agent request → Flight Recorder gateway → Policy check → Human approval → Tool execution → Evidence packet
• It includes:
  • A React/TypeScript control console
  • Netlify Functions for gateway and approval logic
  • Policy-based risk classification
  • Human approval gates
  • Timed step-up approval for high-risk actions
  • Slack-style approval routing
  • ElevenLabs voice review
  • Exportable evidence packets with audit hashes

Inference: The product is described as a demo, not a production-ready solution. It was built for an enterprise action-governance demo and is not evidenced to be in use by any company.

The description states that the project was inspired by a shift in enterprise AI: agents are moving from answering questions to taking actions. The authors claim this shift necessitates governance before execution and proof afterward.

• The product is positioned as:
  • A "black box" for AI agents
  • A way to ensure trust through visibility, control, and evidence
  • A solution to the problem of “human in the loop” being too broad — focusing instead on the action boundary
• The authors claim that the hardest challenge was making governance feel operational rather than abstract.

Inference: The positioning is a response to perceived risks in AI agent adoption. It is not evidenced that this is a widely shared concern or that there’s a market demand for such a solution beyond the hackathon context.

The description states that the product is built for a future where agents do real work across:

• Finance
• Civic services
• Healthcare admin
• Supply chain
• Media rights
• Customer operations
• Software teams

It also says the mission is to let agents move faster while giving companies the control and proof they need to trust them.

Inference: The ICP appears to be enterprise organizations using AI agents, particularly those in regulated or high-risk industries. However, no specific customer segments or use cases are detailed beyond general domains.

The description does not state anything about a business model or pricing.

Not evidenced

The project is built with:

• Frontend: React, TypeScript
• Backend: Netlify Functions
• Database: PostgreSQL
• Other tech: ElevenLabs, Slack-style routing, audit hashes, policy engine, human-in-the-loop components

It includes:

• A control console
• Policy-based risk classification
• Approval routing logic
• Voice review via ElevenLabs
• Exportable evidence packets with audit hashes

Inference: The technical stack is minimal and likely experimental. It appears to be a proof-of-concept, not a scalable or production-ready system.

The description states that the project was built as a live enterprise action-governance demo for the OpenAI 2026 hackathon.

• No revenue, customers, or adoption data is provided.
• The team size is listed as 2.
• It is described as a demo, not a product in use.

Not evidenced

The description does not mention any competitors or existing solutions in the space of AI agent governance or action control.

Not evidenced

• The project is described as a hackathon demo, not a commercial product.
• No evidence of traction, revenue, or customers.
• The team size is small (2 people), which may limit execution capability.
• The technical stack suggests an experimental prototype, not a scalable solution.
• The positioning is based on a self-reported shift in AI agent use — no external validation.

Inference: The project lacks commercial viability and traction. It may be a speculative idea or early-stage experiment with unclear path to market.

1. What specific enterprise use cases have you identified where this solution would be needed?
2. How do you plan to scale beyond the demo environment?
3. Have you validated demand for this product with potential customers?
4. What is your roadmap for moving from a demo to a commercial offering?
5. How do you intend to monetize this product, if at all?