CoDiscover

Turn a real work challenge into a responsible, prioritized, and testable Human-AI use case.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for CoDiscover

Inspiration

Organizations often begin AI adoption with a tool or a generic list of use cases. That creates many ideas, but often weak problem fit, unclear ownership, and little evidence that a proposed workflow will create meaningful value.

CoDiscover starts from a different question: where should people and AI work together—and why? It combines the founder's experience in problem discovery, project design, inclusive practice, and Human–AI collaboration with the implementation capabilities of Codex and GPT-5.6.

What it does

CoDiscover is an installable ChatGPT and Codex plugin that turns a real work challenge into a prioritized, responsible, and testable Human–AI use case.

It supports three reusable actions:

  • Discover: generate up to three materially different opportunities from a challenge, workflow, KPI, role, decision, or document context.
  • Compare: evaluate existing ideas across context fit, value, readiness, evidence gaps, risk, and non-AI alternatives.
  • Sharpen: turn a broad or tool-led idea into a workflow-level experiment.

The selected opportunity becomes a Minimum Testable Use Case (MTUC) with a human owner, clear human and AI roles, boundaries, approval checkpoints, evidence signals, and Go / Revise / Stop criteria.

How it works

  1. Capture a concise context snapshot and separate facts, assumptions, and unknowns.
  2. Locate leverage points in the current work system.
  3. Generate a small set of context-specific options, including a non-AI alternative.
  4. Profile Productivity, Impact, Inclusion, and Innovation independently.
  5. Recommend transparently, including uncertainty and evidence gaps.
  6. Design the smallest responsible test with explicit human accountability.
  7. Apply nine machine-checkable responsibility gates before recommending Go, Revise, or Stop.

This keeps the user experience simple while preserving evidence, inclusion, responsibility, and governance backstage.

How we built it

This OpenAI Build Week prototype was built with Codex and GPT-5.6.

Codex helped transform the founder's earlier problem-discovery framework into a product contract, scaffold the ChatGPT plugin and remote MCP server, implement the Discover, Compare, and Sharpen tools, write the documentation, and create deterministic validators, test fixtures, and automated QA.

GPT-5.6 was the reasoning partner used through Codex to synthesize the architecture, refine tool contracts, challenge product decisions, generate test cases, review outputs, and iterate after QA.

The repository is a clean-room implementation. It includes an installable plugin, a reusable skill, progressive reference files, a structured output contract, synthetic examples, a judge testing guide, and an automated GitHub Actions workflow. Confidential organizational materials, paid-course content, and unverified third-party materials are excluded.

Challenges we ran into

  • Preserving a broad “for all value-creating work” ambition without producing generic recommendations.
  • Keeping discovery separate from execution so the first product remains coherent.
  • Balancing fast answers with responsible evidence and human accountability.
  • Treating external patterns as conditional inspiration rather than an always-on answer library.
  • Translating a rich conceptual manual into a compact, installable, and testable product.

Accomplishments we are proud of

  • A working plugin rather than a prompt document.
  • A four-value compass that does not reduce AI value to time savings alone.
  • A decision-ready MTUC as the destination of discovery.
  • Explicit human accountability and non-AI alternatives.
  • Nine machine-checkable responsibility gates.
  • A redistribution-safe repository with positive and negative QA fixtures.
  • One product surface that can support individuals, teams, and emerging Human–AI organizations.

What we learned

The hardest part of AI adoption is often not generating an answer. It is discovering where intervention is justified, deciding how humans and AI should share the work, and designing evidence that supports a responsible next decision. Strong use-case discovery is therefore a form of work-system design—not merely brainstorming.

What's next

We will run a bounded pilot across five workflow families, calibrate scoring and question selection from observed failure modes, and improve conditional pattern retrieval. Later versions can hand off a validated opportunity from Discover to Design and Build without turning the core product into an unfocused agent bundle.

Success will be measured not only by speed, but also by decision quality, inclusion, experiment readiness, and meaningful impact.

Built With

Share this project:

Updates