Learn with Codex

Create personalized paths with interactive curriculum on demand with Codex (or any other agent) to make real progress on your learning goals.

Comment

Inspiration

Humans working alongside AI agents will often produce remarkable explanations and learning opportunities, but that work usually vanishes into the chat history. We wanted to create a learning experience built on durable knowledge and goals over time alongside an agent instead of restarting from zero every conversation. Real learning needs more than just a answer or chat completion. It often requires a durable goal, an intentional sequence, practice, interactive representations, evidence of progress, and a useful next step. (or path).

Key Inspirations:

What it does

PathMX is a methodology and framework for markdown-first learning repositories. It is used for executable paths, durable sequences of context, interaction, decisions, and evidence that humans and agents can play, adapt, and improve together.

A document tells you something. And an app lets you do something. A PathMX path can guide you through understanding, doing, deciding, and creating. And it remains yours when the session ends.

As we approached building a “learn anything” path, we wanted to build a simple method where any human (or their mom) could download an agent interface (ChatGPT/Codex) and be able to interact closely with the agent about their learning journey.

A learner starts with one simple prompt and a personal goal. Codex creates a local learning repository, asks about the starting point and constraints, and proposes a milestone map before building the first lesson. This is where the chat interface with the AI assistant/agent becomes the orchestration surface, while the generated path becomes the lasting product.

To try it, open an empty writable folder in Codex Desktop and provide our hosted bootstrap URL together with any learning goal. The agent installs the necessary tools, creates the learning space, and opens the first playable map. The primary user journey requires no API keys, plugin development, or manual PathMX setup.

How we built it

We built Learn with Codex with Codex/ChatGPT, using the same durable-source method we were investigating. The repository is the project manager: typed tasks with human owners, a playable change log agents must update before every push (enforced by a check script), research briefs, and evaluation evidence that all ordinary Markdown that humans and agents read from the same context.

PathMX was created by team member Mark Johnson before Build Week, including a published Markdown-first methodology and a proprietary development framework and Player. During Build Week, our team used Codex and GPT-5.6 to meaningfully extend that foundation while creating the new Learning Labs work: teaching skills, starter materials, interactive labs, reusable components, evaluation evidence, and a durable human-agent research workflow.

Codex and GPT-5.6 helped us research learning patterns, author and revise experiments, implement interactive components, diagnose Player behavior, run verification, maintain the task graph, and synthesize findings. Humans selected the questions, reviewed the experiences as learners, and decided which results were worth preserving.

The system has four parts:

1) the pathmx-skills repo with three canonical Agent Skills: /pathmx (authoring), /learn (personal learning), /teach (shared curricula), plus one hosted bootstrap entry point;
2) the pathmx-learning-starter template 3) with milestone maps, module scaffolds, and validity checks;
4) the PathMX methodology and framework, which turns those Sources into focused, interactive experiences; and
5) an eval harness that runs the real Codex CLI (0.142.5–0.145.0) through a five-phase learner journey: onboard → proposed map → confirmed map → complete module → return, which is scored by deterministic repo checks, an independent model judge, and latency telemetry.

Challenges we ran into

Our biggest challenge was having too many promising ideas for the time available. Our first direction attempted to generate a complete “learn anything” product, and testing exposed several problems:

  • Interactive components needed careful lifecycle, focus, and navigation behavior.
  • Deciding which of the many possible features that we should build to bring a better UI/UX for users/learners
  • Waiting time could cause real friction: The time it took AI Agents to build out the course materials is time that may cause an easily distracted human (us) and their attention to wander.

Accomplishments that we're proud of

  • A one-prompt learner flow a nontechnical person can run: one hosted URL → installed skills → a personal, durable learning repo.
  • 16+ playable lab demos from a Kepler orbit instrument and an interactive chess opening lesson to a practice-interview voice simulator, a research bench, and personal-knowledge experiments.
  • A real eval harness with published evidence: final candidate flows scored 100% deterministic + 100% judge at 6m 56s and 10m 07s, with a first useful update in 4–13 seconds on every measured turn (down from 18m55s early runs).
  • A public Build Week Space at build-week.pathmx.net generated from these same sources, where the submission deck itself is a playable, cloneable Path.
  • A human-agent workflow whose tasks, decisions, evals, and changes remain inspectable as plain Markdown.

What we learned

  • Durability changes the agent relationship. A shared file workspace lets learning and decisions compound across sessions.
  • Artifact quality is not learning quality. Pacing, focus, practice, and feedback matter as much as correctness, which is why responsiveness became its own scored dimension.
  • Repository contracts beat prompting. Encoding rules in skills, templates, and checks means the next agent recovers correct behavior from the repo, not hidden context.
  • Learner actions should become evidence. Durable responses give the learner and the agent a shared basis for what comes next.
  • The process can embody the thesis. Our own work with Codex got more coherent when research, tasks, and reviews lived in the same durable graph.

What's next for Learn with Codex

  • Validate collaboration properly: rerun larger modules with subagents disabled vs. required before claiming any speed effect.
  • Ship the second-wave labs (negotiation rehearsal, debugging dojo) and graduate reusable learning patterns into the Starter.
  • Explore stateful and multi-user Paths, deliberately future work until the single-player loop is boringly reliable.
  • Open-source the pieces of PathMX that can be separated from Core as its boundaries stabilize.

Built With

Share this project:

Updates