IntentTrials

What the company appears to be: IntentTrials is a developer tool that claims to address ambiguity in AI-assisted software development by generating multiple implementation alternatives from underspecified feature requests, then locking human decisions into executable criteria before code is built.

What changed: The author states they built this tool to solve a problem where AI coding agents silently choose interpretations of ambiguous requests and implement them perfectly while still building something the user never intended. This represents a shift from purely automated implementation toward structured human oversight.

Single most important open question: Is there evidence that developers actually struggle with ambiguous feature requests in ways that require this specific solution, or is this an abstract problem that may not reflect real-world friction?

The description states IntentTrials is a developer tool that:

• Turns underspecified feature requests into runnable, evidence-backed counterfactual implementations
• Uses GPT-5.6 for structured ambiguity analysis to identify material product-decision axes
• Creates isolated implementation variants using Codex inside Git worktrees
• Evaluates each variant through deterministic verification including type checking, automated tests, production builds, browser journeys, accessibility analysis, screenshot comparison, artifact hashing, source allowlist validation, and Git provenance
• Presents one product decision at a time to human reviewers
• Generates a versioned intent.lock.json containing selected decisions, rejected alternatives, rationale, evidence references, acceptance criteria, protected executable tests, and integrity hashes
• Detects tampering with locked intent and protected tests

The tool is implemented as a TypeScript monorepo with React/Vite dashboard, Fastify/Node.js orchestrator, shared contracts package, and PulseBoard synthetic application for demonstration.

Evidence strength: Self-reported. No independent verification of functionality or usage.

The description states the core principle behind IntentTrials is:

"AI should build and prove the alternatives. The human should make and lock the product decision."

This positions the tool as a solution to the problem that AI coding agents can silently choose interpretations of ambiguous requests, implement them perfectly, but still build something unintended.

The author claims this addresses a deeper problem than conventional testing — that AI can "implement the wrong interpretation of a request perfectly" while passing every test.

Inference: The positioning suggests IntentTrials is positioned as a tool for improving product decision-making in AI-assisted development workflows, not just code generation speed or quality.

The description states:

• IntentTrials is a developer tool
• It targets the problem of ambiguous feature requests in AI-assisted development
• It presents one meaningful product decision at a time to human reviewers
• Future directions include supporting product managers, developers, and reviewers in the same workflow
• It aims to integrate into CI pipelines

Inference: The primary customer appears to be software developers working with AI coding agents who want to ensure correct interpretation of feature requests. Secondary users may include product managers involved in decision-making.

Evidence strength: Self-reported. No evidence of actual customers or user base.

Not evidenced.

The description does not contain any information about:

• Revenue streams
• Pricing models
• Monetization strategy
• Customer acquisition costs
• Unit economics

The description states:

• Built with: codex, git, gpt-5.6, node.js, openai, openai-codex, playwright, pnpm, react, sdk, sqlite, typescript, vite, vitest, worktrees, zod
• Implemented as a TypeScript monorepo with four principal workspaces:
  • React/Vite dashboard for reviewing alternatives
  • Fastify/Node.js orchestrator
  • Shared contracts package (Zod schemas and TypeScript types)
  • PulseBoard synthetic application
• Uses Git worktrees for isolation
• Implements deterministic verification including type checking, automated tests, production builds, browser journeys, accessibility analysis, screenshot comparison, artifact hashing, source allowlist validation, and Git provenance
• GPT-5.6 used for structured ambiguity analysis with strict schemas
• Codex used throughout development process for implementation tasks requiring repository-level reasoning

Evidence strength: Self-reported. No evidence of actual deployment or operational delivery.

Not evidenced.

The description does not contain any information about:

• Revenue
• Customers
• User adoption
• Product-market fit
• Market traction
• Growth metrics
• Customer feedback
• Product usage data

Not evidenced.

The description does not contain any information about:

• Competitors
• Market landscape
• Competitive positioning
• Differentiation from existing tools
• Industry trends
• Market size or opportunity

Risk 1: The tool is described as a hackathon project with only one team member (Youssef El Mahmi). No evidence of scaling beyond prototype level.

Risk 2: The description states the tool uses GPT-5.6, which may not be available to most developers or organizations, creating a dependency risk.

Risk 3: The solution involves complex technical implementation with Git worktrees, isolated environments, and deterministic verification gates. This suggests high development complexity and potential maintenance burden.

Risk 4: The author claims the tool treats human intent as a "durable engineering artifact" rather than temporary chat messages — but there is no evidence of adoption or practical utility in real-world settings.

Red Flag: The entire description is self-reported, unverified, and lacks any demonstration of actual usage or impact beyond the author's own development experience.

1. What specific problems do developers encounter with ambiguous feature requests that this tool solves?
2. How does the tool integrate into existing development workflows (CI/CD, IDEs, etc.)?
3. Are there any real-world use cases or pilot programs where this has been tested?
4. What are the technical limitations of using Git worktrees and isolated environments at scale?
5. How does the tool handle edge cases or unexpected inputs from ambiguous requests?
6. What is the roadmap for making the tool accessible to organizations beyond individual developers?
7. How do you plan to monetize this tool, and what pricing model would be appropriate?
8. What are the security implications of running Codex builders inside isolated Git worktrees?