Architect

The company appears to be a solo developer project named "Architect", submitted to the OpenAI 2026 hackathon. The author states that Architect is a Windows desktop tool designed to govern AI coding agents by introducing a "governed software production system". It uses a process-separated evaluator called Wraith to validate agent-generated code and includes mechanisms for bounded repairs, evidence hashing, and deterministic replay.

• The description states: Architect is a Windows desktop developer tool built with Go, Wails, React, TypeScript, PowerShell, Codex CLI, SRXML task contracts, and local evidence artifacts.
• The description states: It turns Codex into a governed software production system.
• The description states: It defines one mission, compiles bounded task packs, launches multiple Codex workers, records their execution, and treats every worker completion claim as untrusted until validated.
• The description states: A process-separated evaluator called Wraith tests the result using presealed validation evidence.
• The description states: If Wraith finds a hidden failure, Architect returns FALSE_SEAL, classifies the defect, unlocks only the responsible scope, permits one bounded repair, and revalidates the result.
• The description states: The final output includes real Codex execution receipts, worker claims and changed files, Wraith evaluation evidence, failure and repair history, a Sovereign Seal, an inspectable evidence graph, and deterministic replay.

Not evidenced: No information on actual functionality beyond the author's own qualification runs. No evidence of integration with existing development workflows or tools. No evidence of real-world usage by other developers.

• The description states: "AI agents can say they finished. Architect proves whether they did."
• The description states: The core idea is to solve the trust problem in AI coding agents.
• The description states: It was built to address the issue that AI coding agents can confidently claim a task is complete when implementation is incomplete, unsafe, or outside original intent.
• The description states: "Generating code is becoming easier. Determining whether autonomous work deserves to be trusted is becoming the harder and more valuable problem."
• The description states: Reliable agent systems need separation between human intent, execution authority, evaluation authority, repair authority, and final judgment.

Inference: The positioning evolved from a hackathon prototype focused on solving a trust problem in AI agents to a more structured system with defined roles for different components (workers, Wraith, etc.). However, this is based on the author's own description and lacks external validation or market traction evidence.

• The description states: Architect is a Windows desktop developer tool.
• The description states: It was built to govern AI coding agents.
• The description states: It is designed for developers who use Codex for software development.
• The description states: The current Build Week release uses process and repository boundaries on one Windows PC.

Not evidenced: No information about specific customer segments, personas, or target industries. No evidence of market research or customer interviews. No indication of whether the tool targets solo developers, teams, or enterprises.

• The description states: No explicit business model or pricing is mentioned.
• The description states: The project was submitted to a hackathon (OpenAI 2026).
• The description states: It was built using open-source technologies and tools like Codex, Git, Go, React, etc.

Not evidenced: No information about monetization strategy, pricing tiers, or revenue streams. No evidence of any commercial activity beyond the hackathon submission.

• The description states: Built with Go, Wails, React, TypeScript, PowerShell, Codex CLI, SRXML task contracts.
• The description states: Portable same-PC Codex execution.
• The description states: Three fixed Codex worker cells.
• The description states: Immutable mission and authority contracts.
• The description states: Process-separated Wraith evaluation.
• The description states: Presealed private validation.
• The description states: Real false-completion detection.
• The description states: One bounded repair path.
• The description states: Evidence hashing and receipt verification.
• The description states: Replay and safe termination.
• The description states: Judge Mode.
• The description states: Portable release packaging.

Inference: The project shows technical sophistication in building a multi-component system with separation of concerns, but this is inferred from the author's own description. No evidence of production deployment or scalability beyond one PC.

• The description states: Two clean Sovereign qualification runs.
• The description states: Eight current Codex thread executions.
• The description states: Three real worker cells per qualification.
• The description states: A real Wraith FALSE_SEAL.
• The description states: One scoped repair.
• The description states: Successful Wraith revalidation.
• The description states: Replay verification.
• The description states: Working baseline fallback.
• The description states: Clean extraction and launch.
• The description states: Portable Windows release.
• The description states: Zero operator code repairs during qualification.

Not evidenced: No evidence of external adoption, user feedback, or real-world usage beyond the author's own testing. No evidence of revenue, customers, or market traction.

• The description states: The project addresses a trust problem in AI coding agents.
• The description states: It aims to become the control and proof layer for agent-built software.
• The description states: It is designed to govern AI coding agents.

Not evidenced: No information about existing competitors or market positioning. No evidence of competitive analysis, pricing comparisons, or differentiation from other tools in the space.

• The description states: The project was built as a hackathon submission.
• The description states: It currently only works on one Windows PC.
• The description states: Future development will add stronger host-level isolation, broader repository support, additional model runtimes, enterprise policy controls, and team-based evidence review.
• The description states: No revenue, customers, or traction data are available beyond what is self-reported.

Inference: The project lacks commercial viability without clear evidence of market demand or adoption. It's unclear whether the author has plans to scale beyond a single developer tool or if there's any path to monetization.

1. What specific problems in AI agent-generated code are you solving, and how do you know these problems are real?
2. How does this tool integrate into existing development workflows, and what is the friction for adoption?
3. Have you tested this with other developers or teams beyond your own qualification runs?
4. What is your plan to move from a single-PC Windows tool to broader platform support and enterprise controls?
5. Is there any evidence of interest from potential users or partners outside the hackathon context?