AI curriculum

Course outline for Human–AI Workflow Design

Design people, technology, and work practices together. An agent deployment changes who reviews, decides, learns, and recovers when something fails. Study actual work, assign roles and authority, and test human review. Preserve the skills needed to supervise and recover automation through manual rotation, escalation practice, and incident drills. Design adoption and trust repair too.

Human–AI Workflow Design Course Objectives

  • Explain joint optimization of people, technical tools, and work practices.
  • Compare work as documented with work as actually performed.
  • Identify each technical change's effects on roles, workload, judgment, and expertise.
  • Allocate work, decision rights, approval, and execution authority explicitly.
  • Make independent judgments and measure assisted-review errors and effort.
  • Design and exercise manual rotation, escalation practice, and incident recovery.
  • Define evidence-producing skill checks and later reassessment.
  • Design phased adoption, worker feedback, and trust-repair responsibilities.
  • Evaluate the combined design without claiming short-term tests prove retained skill.

Pre-requisites

  • Describe a workplace process, its decisions, consequences, and people.
  • Basic awareness of AI limitations and workflow models.
  • Read short evidence packets and use a timer or spreadsheet.
  • No programming, live model, or socio-technical research background is required.

Before this course

Complete the required courses or bring equivalent experience. Recommended courses add useful depth; they are not enrolment gates.

Required preparation

Recommended preparation

Readiness checks

  • Describe a workplace process, its decisions, consequences, and people.
  • Basic awareness of AI limitations and workflow models.
  • Read short evidence packets and use a timer or spreadsheet.
  • No programming, live model, or socio-technical research background is required.

Describe a workplace process, its decisions, consequences, and people.

Find your place in the learning map

Lab Setup

Delivery requirements: this environment, access, and the exercise materials must be prepared and verified before delivery. This outline does not certify that a lab package is ready.

Editor, timer/spreadsheet, process diagram, fictional policy, staff accounts, role cards, matched evidence sets, controlled recommendations, hidden answer keys, and an outage/escalation simulation. Support paired and solo participation. Pretest case difficulty and reference judgments with a domain reviewer. No live model, real customer decisions, personnel monitoring, or production system access is required. Staff accounts are synthetic teaching evidence, not a claimed field study.

Detailed Course Outline

Module 1: Design the social and technical system together (2 hours)

  • Socio-technical system: Socio-technical systems: people, tools, practices, and operating environment.
  • Joint optimization: Joint optimization versus isolated technical performance or staff convenience.
  • Actual and documented work: Work-as-imagined versus work-as-done and useful informal coordination.
  • Human effects of technical choices: Technical choices and changes to review, discretion, workload, and learning.
  • Human-role inventory: Human-role inventories, affected people, accountability, and worker participation.
  • Before/after design: A before/after design linking social and technical consequences.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.75 hours for guided and independent practice, 0.5 hours for checks and feedback, and 0.25 hours for questions and contingency.

Practical outcome: Identify three differences in actual work and pair each proposed automation change with an explicit human-work design.

Module 2: Share authority and make adoption deliberate (2 hours)

  • Work allocation: Allocation by capability, consequence, reversibility, and total human effort.
  • Authority and escalation: Advice, judgment, authorization, action, and reachable escalation owners.
  • Human judgment: Human roles in ambiguity, accountability, trust, and domain context.
  • Review and exception burden: Review capacity, incentives, job meaning, and uneven exception burden.
  • Designed adoption: Shadow operation, limited rollout, worker feedback, and adoption gates.
  • Trust repair: Visible incident explanation, correction, and trust-repair communication.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.75 hours for guided and independent practice, 0.5 hours for checks and feedback, and 0.25 hours for questions and contingency.

Practical outcome: Produce a role/authority map and staged adoption plan with evidence and accountable owners at each gate.

Module 3: Give reviewers a real chance to disagree (2 hours)

  • Review evidence: Evidence visibility, missing information, contradictions, and uncertainty.
  • Confidence and support: Model-stated confidence versus independently checked support.
  • Independent-first judgment: Independent-first judgments, automation bias, and persuasive errors.
  • Review decisions: Accept, correct, reject, and escalate with explicit review criteria.
  • Observed review effort: Observed review errors, time, disagreement, and interruption burden.
  • New-case retesting: A redesigned review control tested on new matched cases.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.75 hours for guided and independent practice, 0.5 hours for checks and feedback, and 0.25 hours for questions and contingency.

Practical outcome: Complete a controlled review experiment and test a correction without treating every changed judgment as bias.

Module 4: Keep supervision and recovery skills exercised (2 hours)

  • Skill-preservation dynamics: Skill-preservation dynamics and reduced opportunities to practice critical judgment.
  • Recovery expertise: Automation's remaining exceptions and the expertise needed for recovery.
  • Manual-work rotation: Rotation through manual work with meaningful decisions and feedback.
  • Escalation and incident drills: Escalation-path practice, absent-owner scenarios, and incident drills.
  • Practice evidence: Evidence-producing practice, role coverage, and spaced reassessment.
  • Recovery and reassessment: Outage recovery exercise and a retention/adoption follow-up plan.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.75 hours for guided and independent practice, 0.5 hours for checks and feedback, and 0.25 hours for questions and contingency.

Practical outcome: Perform an unaided decision and outage/escalation drill, then schedule skill-preservation practice with evidence and reassessment.

Practical exercise

  • Use a synthetic repair desk with a documented process, staff accounts of actual work, a short domain policy, and evidence packets. Identify three differences between the documented and performed process. Examine informal peer cross-checks, private reference collections, and exception handling as possible design evidence, not automatically misconduct or practices to remove. Validate interpretations with the affected role in a role-play or written case response.
  • Build a before/after socio-technical design. For each technical change, record who now performs the work, who reviews, who exercises judgment, who learns, and who recovers failures. Include the time and staffing needed for these roles. Distinguish what the process needs humans to retain from whatever work automation happens to leave behind. Include one harmless reversible action without blanket approval and a consequential action with a named decision/authorization boundary. The authority table is a design artifact, not deployed technical enforcement.
  • Define shadow operation and limited-rollout gates. Record what affected workers gain or lose, who can report a problem, who can stop the rollout, and what evidence permits expansion. Do not set adoption success to nominal tool usage alone. Include visible correction and communication after an agent error and an owner responsible for repairing the workflow and trust. Trust does not mean persuading people to ignore warranted concerns.
  • Record unaided judgments before showing controlled recommendations on matched cases. Review six recommendations including correct advice, an unsupported claim, a plausible policy error, and conflicting evidence. Record accept/correct/reject/ escalate, supporting evidence, time, and unresolved uncertainty. Counterbalance case sets where practical and report order/learning effects otherwise. Keep the answer key hidden until attempts are saved. A justified correction is not bias. For solo learners, report reference disagreement rather than invented peer data.
  • Revise one review control and test new matched cases. Then perform a manual verification task with no model recommendation. Run an agent-down incident drill and an escalation drill in which the usual decision owner is unavailable. Exercise the actual needed judgment: inspect evidence, identify the error, select a safe fallback, and hand off unresolved work. Merely clicking an escalation button is not a skill-preservation exercise. Preserve decisions, errors, time, feedback, and next practice needs as evidence.
  • Design a recurring manual-work rotation, escalation rehearsal, and incident-drill schedule. Assign owners, coverage, protected time, required judgment, evidence, and later unaided reassessment. Choose frequency from consequence and observed proficiency, not a universal monthly/quarterly rule. Account for the throughput and staffing cost of practice instead of treating it as free. A fallback plan that assumes unpracticed expertise is not a complete recovery design.

Delivery and timing

  • Guided hours include explanations, demonstrations, learner practice, checks, feedback, questions, and contingency as allocated above. Breaks, bulk downloads, and prerequisite account/environment setup are additional. These allocations require a timed pilot before delivery; they are not measured completion times.

What you will produce

Design human and technical work together, test review, and exercise the judgment needed for recovery.

How the work is checked

  • The design treats social and technical choices with equal rigor.
  • Differences between documented and actual work lead to justified design decisions.
  • Each automation change identifies review, workload, judgment, and expertise effects.
  • Roles, authority, escalation, adoption gates, and trust repair have named owners.
  • Independent review precedes recommendations and answer disclosure.
  • Review observations include errors, evidence, effort, uncertainty, and new-case retesting.
  • Manual verification and outage/escalation drills exercise critical judgment.
  • Recurring practice has protected time, coverage, evidence, feedback, and later assessment.
  • Same-day practice is not described as proof of long-term retained skill.
  • Overall recommendations consider quality, workload, human capability, and simpler alternatives.

8 guided hours. Design roles, authority, adoption, review, and recovery jointly. Synthetic staff accounts and controlled recommendations are workshop evidence, not a field study. Manual rotation, escalation practice, and incident drills exercise critical judgment and have staffing costs. Same-day observations cannot prove long-term retention, universal skill decay, or organization-wide adoption success. Later unaided reassessment is a scheduled follow-up.

When another approach is better

  • Do not add approval to every harmless reversible action if it creates an unmanageable review queue.
  • Do not delegate consequential decisions when reviewers lack evidence or the ability to challenge the result.

How we train

Contact us for full course details, including duration, delivery options and lab requirements.

Ask about exercises, instructor feedback and the prior knowledge you need. Tool-specific courses marked provisional may become modules in a broader course.

Discuss your learning goals