Agentic learning engineering · Problems → Products

Explore learning engineering teams

About this resource · How to use it · Learning engineering connections

What is this resource?

A flower-based experiment studio and a collection of preserved agent runs on fictional problems. Configure team contributions, examine the original journals and candidate tools, and identify the observations still needed with real learners and practitioners.

How should I use it?

  1. Open Dynamic conversations to configure Design / Develop roles and an editable case, then inspect easy/medium/hard Deploy / Test conditions.
  2. Choose an earlier case and replay its original discussion. Compare a challenge, decision, and artifact; those records retain their original team assignments.
  3. Inspect the requirements and manifest. Name an affected-person review or field observation that could challenge the candidate before use.

How does this connect to LENS and learning engineering?

The flower makes complementary expertise and shared understanding explicit across a learning engineering cycle. LENS connects Human flourishing, Learning sciences, and Systems engineering; the simulation asks how contributions enter decisions and executable designs. Output alone does not demonstrate human capability or learning impact.

Configure complementary expertise around the learning engineering flower. Follow how shared understanding, contributions, and operating conditions shape a candidate design, then inspect the evidence behind it.

A shared framework for the experiments

Eight starting expertise domains surround shared understanding. Learners and other stakeholders remain connected to the whole team. Assign one or several domains to a member, share responsibilities, and select the expertise the challenge needs.

Eight starting domains of expertise around a shared challenge. Team composition can change with the problem and stage. Schematic based on Ali and colleagues (2025); it is not an exhaustive taxonomy or a proficiency scale.

Open dynamic design conversations Understand the five comparisons

Use two phases: dynamic Design / Develop conversations, then Deploy / Test under easy, medium, and hard conditions. Specialists have bounded responsibilities; the generalist has introductory cross-domain knowledge. An expert evaluator examines the complete conversation and observed result separately. Edit and export a case here, run it locally, then import the retained record. Earlier fixed-workflow results remain available below.

Earlier case records · 3 problems · 6 runs

Preserved agent execution on fictional problems, using the original team configurations. These replays are historical examples, not new flower study results. Each refinement keeps its original run available; human pilot approval and field evidence remain pending.

Case inventory and journal hashes · Expansion checklist

Trace contributions into the design

Shared understanding means integrating domain concerns into requirements, decisions, implementation, and evaluation. Every team member has responsibility for the learners and stakeholders affected by the design.

Explore the flower domains and shared responsibilities. The guide also preserves the original role contracts for interpreting historical replays.

The scorecard examines human flourishing support, system performance, and learning separately. Ratings concern the design and available evidence. Claims about learning or impact require suitable evaluation with people.

Inside the workforce candidate

Conversation Lab: facts, perspectives and next moves lets a new team lead compare authored openings, separate facts from explanations, and reconsider a proposal when protected time and coverage are withdrawn. Access, authority, capacity and agreement remain independent. Its reflection is descriptive; it does not assess employees.

The handoff contains executable HTML, requirements, versioned artifacts, the project journal and a source-linked manifest. It is a candidate product requiring affected-person review and field validation.

Inspect the product handoff · Read the requirements

The preserved original workforce record

The saved fictional workforce project contains 29 model requests, including a rejected correction, and 14 artifact versions. Scenarios changed interruption privacy and reviewer-conflict requirements. An actual missing-code finding reopened Build.

The replay preserves original drafts, dissent, prompts, provider metadata, failed validation and the corrected candidate. Recorded browser checks support only the behavior exercised in that environment.

Inspect the source journal · Follow the original workforce revisions

From candidate to operational product

Understand → Map → Design → Build → Instrument → Experiment → Evaluate → Refine is a revisitable cycle. Build produces the executable candidate; refinement can return to any invalidated requirement. The learning engineer integrates the work and keeps human and system requirements visible.

Real use needs accountable ownership, affected-person review, appropriate data handling, accessible operation, a meaningful comparison and stopping rules. Learning, workplace transfer and flourishing remain claims to investigate. Inadequate conditions can justify changing the product, changing the work system, or stopping.

The hosted site serves saved products and replay. New model runs use the local runner in this repository; hosted model access and a shared job queue remain future work.

Return to project seeds · Runner source · Specification and operating documentation