Fencepost

The description states that Fencepost:

• Takes a line of code written by a student (identified via Git)
• Changes one character in that line
• Runs the student's existing tests on the modified code
• If those tests don’t catch the change, GPT-5.6 generates a new test to show the difference
• A sandbox environment runs this test and reports the result
• The output is a short report for teachers to ask questions — not a grade or verdict

It uses:

• Codex with GPT-5.6 as the engine
• AST (Abstract Syntax Tree) for code manipulation
• Docker sandboxed execution
• Python, pytest, git, mutation testing concepts

The tool is described as being built in a way that avoids API keys and internet access — running locally on the teacher’s plan.

Inference The product appears to be a proof-of-concept prototype with no commercial or production deployment yet. It is not evidenced to have been used beyond the author's own testing.

The author states:

• The tool is intended for use in CS courses where AI agents may be used to complete assignments
• It aims to detect whether students understand their code, rather than just whether it passes tests
• It avoids traditional AI detectors that are seen as unfair or ineffective
• It is framed as a way to prompt discussion with students, not to judge them

Inference The positioning is centered on educational integrity and understanding, not on AI detection or grading. The claim evolution shows an intent to move from a tool that simply checks code correctness to one that probes comprehension.

The description states:

• The primary users are teachers in CS courses
• It is intended for use in programming education contexts
• The author mentions “a real CS2 class this fall” as the next step, suggesting a classroom setting

Inference The target customer is likely educators or teaching assistants in computer science programs. The ICP (Ideal Customer Profile) appears to be early-stage educational institutions or individual instructors using it for assessment.

The description states:

• No pricing model is described
• It is MIT licensed
• The report stays local — no data leaves the user’s environment
• It runs on the teacher's own GPT-5.6 plan, with no API key required

Inference There is no evidence of a monetization strategy or pricing structure. The tool is self-hosted and open-source (MIT license), suggesting no commercial business model at this stage.

The description states:

• Built using GPT-5.6, AST, Docker, Python, pytest, git
• Uses mutation testing principles to generate meaningful test cases
• Runs in a sandboxed environment with no internet access
• The author walked through the product himself multiple times for design and validation

Inference Technical execution shows some sophistication (e.g., AST manipulation, sandboxing), but there is no evidence of scalability or production-grade delivery. It is a prototype built by one person.

The description states:

• Submitted to the OpenAI 2026 hackathon
• Built by one person (Kevin Soto Burgos)
• No revenue, customers, or adoption data are provided
• The demo was created with fake student data for demonstration purposes
• Next step is to use it in a real CS2 class

Inference There is no evidence of traction or commercial maturity. It is a hackathon project with no known users beyond the author.

The description does not mention any competitors or existing tools in this space.

Inference No competitive landscape is described, nor is there any indication that similar tools exist or have been used in education. The author explicitly states they did not want to make another AI detector, which implies a niche or underserved area.

• No commercial traction or adoption: The tool is a hackathon submission with no evidence of real-world use.
• Single-person development: No team or external support is evident.
• Unproven educational impact: There is no data on effectiveness in real classrooms.
• Limited scalability: The product appears to be built for local, single-user use.
• No monetization strategy: No indication of how the tool would generate revenue.

1. Has the tool been tested in a real classroom setting?
2. What is the expected user experience for teachers and students?
3. Are there any plans to scale beyond the MIT-licensed prototype?
4. How does the tool handle edge cases or complex code changes?
5. Is there any feedback from educators who have tried it?
6. What are the technical limitations of running GPT-5.6 locally?