Agentic learning engineering · Team framework

The learning engineering flower

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

What is this resource?

A guide to the learning engineering flower and its eight starting expertise domains, shared learner and stakeholder responsibilities, and controlled team comparisons. Earlier role contracts remain available for interpreting historical runs.

How should I use it?

  1. Inspect the flower and choose a domain to examine its requested contribution. Consider which expertise the particular challenge needs.
  2. Open Dynamic conversations, inspect each actor’s expertise and knowledge brief, and compare expertise omission, withdrawal, member removal and a broad but shallow generalist.
  3. Use the historical roster section when interpreting an earlier replay; retain the assignments and evidence from that run.

How does this connect to LENS and learning engineering?

Shared understanding connects domain contributions into a coherent design. Team composition may change with the challenge and stage. Flower domains organize expertise; they are not validated individual competency scores or a replacement for involving learners and stakeholders. LENS foundations remain Human flourishing, Learning sciences, and Systems engineering.

Learning engineering brings complementary expertise together around a shared challenge. The flower places shared understanding at the center and connects the whole team with learners and other stakeholders.

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.

Expertise contributes to the work

The default simulation assigns one member to each domain and a coordinator. A particular challenge may need a different mix. One person can contribute several domains, and several people can share a domain. These editable instructions organize contributions; they do not establish anyone’s professional competence.

Domain 1

Learning Sciences

Learning mechanisms, prior knowledge, practice, retention, and transfer.

Domain 2

Learning Experience Design

Learner journeys, activity structure, feedback, and revision.

Domain 3

Data Science

Data quality, analysis, uncertainty, and model assumptions.

Domain 4

Assessment, Measurement, Evaluation

Constructs, measures, comparisons, and limits of inference.

Domain 5

Learning Environment Engineering

Operating environment, dependencies, access, and recovery.

Domain 6

Software Engineering

Executable behavior, state management, testing, and maintenance.

Domain 7

Subject-Matter Expertise

Content accuracy, authentic tasks, and consequential errors.

Domain 8

Education and Training Professionals

Facilitation, learner support, adoption, and delivery conditions.

Source: Ali and colleagues, Learning engineering is a team sport (2025). The flower domains are starting points and are not conclusive. Composition may shift with the challenge and stage.

The center is everyone’s responsibility

  • Contribute to shared understanding by making assumptions, disagreements, decisions, and unresolved questions explicit.
  • Keep learners and other stakeholders connected to the work. Identify whose perspectives are supplied and what direct participation is still needed; an agent assignment does not represent their consent or agreement.
  • Distinguish supplied case facts, generated proposals, assumptions, observed checks, and information still needed from learners or stakeholders.

Assigning a responsibility does not substitute for involving learners, practitioners, and other stakeholders. Every configuration should distinguish supplied facts, proposed interpretations, and information still needed from those people.

Flower domains

Organize the expertise contributing to a team’s design work.

LENS foundations

Human flourishing, Learning sciences, and Systems engineering remain the three foundations of the program’s approach.

Evidence scorecard

Human flourishing support, system performance, and learning are assessed separately. They describe the design and available evidence; human outcomes need evaluation with people.

Five comparisons ask different questions

Distinguish a missing responsibility from losing a member’s time and contribution opportunities. Fix the case facts, challenge schedule, execution settings, and coordination policy, then inspect what each contrast actually changes.

ConditionChangeCapacity and interpretation
Full teamRetain the configured specialist assignments and available members.Route questions and objections dynamically within the common conversation budget.
Domain omitted everywhereWithdraw the selected domain’s explicit instructions and approval authority from every member.Retain identities and make the expertise gap visible.
Domain withdrawn from one memberWithdraw that domain from one selected member; other assignments remain.Keep this member available to ask questions; assigned overlap may compensate.
Member removedRemove the selected member and their available contributions.Retain any declared overlap held by other members; report actual participation.
Broad but shallow generalistOne generalist uses an introductory cross-domain brief and can make routine proposals and implementations.No specialist playbooks are injected into later turns; no specialist signoff is implied.

The pilot exercised the full specialist team and broad generalist configurations across two cases. The other three configurations remain available for future experiments.

The revised tool uses dynamic Design / Develop conversations followed by Deploy / Test scenarios. Questions, replies, objections, user updates and unresolved issues determine the next contribution. A bounded conversation preserves its full accepted transcript. The earlier fixed-workflow study used forty scheduled calls, or thirty-six after member removal; those counts do not describe the revised protocol.

Overlap may allow another member to identify a concern. Omission does not erase a model’s latent knowledge, and no condition is programmed to fail. Matching calls also does not match token volume, independent perspectives, or human effort.

Configure dynamic flower conversations

Inspect the design and its evidence

The current protocol freezes an implemented design before testing. A separate evaluator receives the assigned roles, complete accepted design conversation, frozen source and relevant scenario observations. The review is not blinded. A proposed repair becomes new evidence only after implementation, a new freeze and new tests.

The revised expert evaluation reads the complete design conversation, scenario conditions and observed browser evidence. It scores the process once per artifact version and the result separately for easy, medium and hard conditions. The preserved earlier study used a different fixed workflow and cannot validate the revised design. Neither agent judgments nor scripted browser observations establish human learning.

Inspect the dynamic conversations, tests and prototypes · Read Exploring the learning engineering flower with agentic simulation · Read the execution and evaluation documentation

Historical roster and role contracts

These profiles document the earlier replay collection. Their original duties and requested LENS targets remain attached to those saved runs. They are not a crosswalk from the flower domains to a validated competency framework.

Earlier roster

Learning engineer

Integrate learning sciences, human-centered design, systems requirements and evidence. Preserve dissent and choose the earliest stage to revisit.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Synthesize drafts into a usable stage artifact and explicit decision.

Earlier roster

Learning scientist

Observable capability, descriptive feedback, retention, deliberate practice and transfer. Completion is not competence.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Systems engineer

Human-system coupling, constraints, requirements, recovery and non-training levers.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Human-centered designer

Affected people’s voice, accessibility, workload, participation and agency.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Measurement specialist

Task evidence, comparison, transfer, causal threats, subgroup coverage and unknowns.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Responsible AI specialist

Delegation, privacy, surveillance, correction and accountable human decisions.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Implementation specialist

Small runnable artifact, technical limits, instrumentation and fallback.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Domain specialist

Use supplied domain context; state limits of expertise and required practitioner validation.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Challenge the plan within your discipline and propose a reviewable artifact; do not supply human approval.
  • Tie each proposed product requirement to a specific user interaction, its purpose, and an observable acceptance check; preserve uncertainty and explain what changed after a challenge.
  • Tie the contribution to the current seed and actual earlier journal events.

Earlier roster

Scenario agent

Introduce a plausible hypothetical objection or unexpected event that challenges a named assumption and requests a requirement change.

Original responsibilities
  • Distinguish supplied facts, assumptions, hypothetical events and missing evidence.
  • State human and system capability requirements; examine agency, dignity, access and workload.
  • Preserve dissent, correction, non-AI alternatives and accountable human gates.
  • Tie the contribution to the current seed and actual earlier journal events.

Registry 1.2.0; framework 1.1.0. Requested targets have no demonstrated proficiency evidence by themselves.

Inspect the original resolved role contract · Open a historical replay