A2A Trade

Company: A2A Trade

Self-reported basis: The description is entirely self-reported and unverified; it contains no evidence of revenue, customers, or traction.

What the company appears to be: A governed AI trading system that connects to a live E*TRADE account and uses GPT-5.6 for reasoning and language processing, but enforces strict code-based guardrails to prevent unsafe trades.

What changed: The project was built as part of an OpenAI 2026 hackathon submission. It represents a proof-of-concept for bounded AI autonomy in financial trading.

Single most important open question: Is the system truly safe and bounded, or does it rely on assumptions that may fail in production?

The description states that A2A Trade is a **governed GPT-5.6 trading pocket connected to a live E*TRADE account**. It uses:

• **E*TRADE API** for live market data and order placement
• GPT-5.6 as the reasoning and language layer
• Code-based guardrails, not prompts, to enforce safety
• Audit logging of all actions

The system is described as allowing users to interact with an AI assistant (Aiden) via a public app interface, where Aiden can:

• Answer grounded market questions with live quotes
• Convert explicit instructions into governed order proposals
• Refuse unsafe requests like selling all holdings or placing orders above a $100 cap

The project is built using Next.js, TypeScript, Supabase, OpenAI Responses API, Yahoo Finance, and **E*TRADE OAuth 1.0a**.

Inference: The product appears to be a prototype for AI-assisted trading with strict safety boundaries, not a commercial product or service yet.

The author states that A2A Trade is built to show what an AI provably cannot do, rather than what it can do. It is positioned as:

• A governed AI system for trading
• A bounded autonomy model for real money
• A safe experiment for ordinary people using AI on their brokerage accounts

The project is framed as a demonstration of safety in AI trading, not a commercial offering.

Inference: The positioning is focused on safety and control, not scalability or monetization. It is a prototype, not a product.

The description states that the system is built for:

• A normal person who wants to experiment with AI on real money
• Ordinary investors seeking a safer way to interact with AI in trading
• Veterans, as a broader mission to support and improve lives of veterans

It is not clear if there are specific personas or segments targeted beyond this general audience.

Inference: The ICP appears to be individual retail traders or AI experimenters, not institutional or enterprise users. No evidence of segmentation or targeting beyond the general public.

There is no evidence in the description of:

• A pricing model
• Revenue streams
• Monetization strategy
• Subscription plans or fees

The project is described as a hackathon submission, not a commercial offering.

Inference: No business model or pricing is evidenced. The system appears to be a prototype, not a product with a monetization path.

The system is built using:

• Next.js
• TypeScript
• Vercel
• Supabase
• OpenAI Responses API
• Yahoo Finance API
• **E*TRADE OAuth 1.0a**

Key technical decisions include:

• Safety in code, not in prompts
• Broker adapter enforces guardrails before any request reaches E*TRADE
• Audit trail of all actions
• Deterministic execution underneath GPT-5.6

The system is described as using Codex to scaffold and iterate on the build.

Inference: The technical architecture shows a deliberate focus on safety through code, not just prompt engineering, which is a strong signal for a controlled system.

There is no evidence of:

• Customers
• Revenue
• User adoption
• Product usage metrics
• Live deployment beyond the hackathon

The project was submitted to the OpenAI 2026 hackathon, and the only demonstration is a pilot with real orders placed during that time.

Inference: The system has no traction or maturity beyond a prototype. It is not a product in use, but a proof-of-concept.

The description does not mention:

• Competitors
• Market positioning relative to others
• Existing solutions in AI trading or bounded autonomy

It is clear that this is a novel approach to AI-assisted trading with safety boundaries, but no competitive landscape is described.

Inference: No competitive context is evidenced. The project appears to be unique in its approach, but not validated in the market.

• No revenue or customer data: The system has no traction.
• Unverified safety claims: The description states that safety is enforced by code, but there is no independent verification of this.
• Prototype only: Not a product, not deployed for real users.
• Limited scope: Only one brokerage (E*TRADE) and one AI model (GPT-5.6).
• No scalability or generalization plan: The next steps are described as "generalizing beyond one brokerage", but no evidence of progress toward that.

Inference: Risks include unproven safety, no commercial viability, and limited scope.

1. What specific guardrails are enforced in code, and how are they tested?
2. How is the system validated to ensure it does not bypass its own safety boundaries?
3. Is there any evidence of real-world usage beyond the hackathon pilot?
4. What are the plans for expanding beyond E*TRADE and GPT-5.6?
5. Are there any known limitations or edge cases in the current implementation?
6. How is audit logging used, and what is its purpose beyond compliance?