Photograph: Shane Rounce on Unsplash

Learn Courses preview Course plans by Shane Rounce. Enrolment is not open.

← All courses

Warm sunlight dissolving into soft circles of light

Photograph: Shane Rounce on Unsplash

Course plan · Enrolment not openAI and agent workflows · Self-paced

Maintain AI tooling, model choices and long-running work

With Shane Rounce

I teach maintain AI tooling, model choices and long-running work through a practical project, worked examples and checks that another person can repeat. This is a complete course plan; lesson recordings, workbooks and delivery are still to be produced.

4 weeks20 learning hoursSelf-paced

I’m developing this course as part of Learn Courses. This is the full teaching plan, ready to review before I produce the lessons and open enrolment.

IS THIS FOR YOU?

Who I’m building this for

Technical leads running repeatable AI workflows.

Before you begin

A computer with a current browser and permission to use the project material.
Basic familiarity with Git and the relevant development tools. Work takes place in an isolated test environment.
No client records, credentials or production changes are needed for the exercises.

PROPOSED COURSE FEE

£149

Provisional planning figure, not an approved price or an offer for sale. I have assumed reusable self-paced lessons and bounded support. Production effort, access period, assessment support and tax treatment need confirming before sale.

Planning price only. Enrolment is closed and no payment is taken.

Explore a sample learning space →

The sample is fictional. It is separate from this course’s future lessons.

A PRACTICAL NEXT STEP

What you’ll be able to do

Find model identifiers, settings and integration boundaries without exposing credentials.
Build a small comparison set including useful failures and cost observations.
Exercise resumed sessions, worktree recovery and durable decision notes.
Record version changes, known constraints and a rollback route with official references.

THE TEACHING PLAN

Your next 4 weeks

A clear route through the subject, with something to make, test or reflect on at every stage.

MODULE 1

Inventory saved assumptions

I explain the decisions behind inventory saved assumptions using worked examples, a deliberately broken case and a corrected version. The lesson distinguishes what the evidence establishes from what still needs checking.

Your practical activity: Find model identifiers, settings and integration boundaries without exposing credentials.
How I’ll check your understanding: Submit the working artefact and a short evidence note for module 1: find model identifiers, settings and integration boundaries without exposing credentials. Include the initial state, the chosen change, a meaningful failure case and the result of the check. I review whether the evidence supports the conclusion and return specific revisions where it does not.

MODULE 2

Evaluate representative work

I explain the decisions behind evaluate representative work using worked examples, a deliberately broken case and a corrected version. The lesson distinguishes what the evidence establishes from what still needs checking.

Your practical activity: Build a small comparison set including useful failures and cost observations.
How I’ll check your understanding: Submit the working artefact and a short evidence note for module 2: build a small comparison set including useful failures and cost observations. Include the initial state, the chosen change, a meaningful failure case and the result of the check. I review whether the evidence supports the conclusion and return specific revisions where it does not.

MODULE 3

Recover a longer task

I explain the decisions behind recover a longer task using worked examples, a deliberately broken case and a corrected version. The lesson distinguishes what the evidence establishes from what still needs checking.

Your practical activity: Exercise resumed sessions, worktree recovery and durable decision notes.
How I’ll check your understanding: Submit the working artefact and a short evidence note for module 3: exercise resumed sessions, worktree recovery and durable decision notes. Include the initial state, the chosen change, a meaningful failure case and the result of the check. I review whether the evidence supports the conclusion and return specific revisions where it does not.

MODULE 4

Make an update reversible

I explain the decisions behind make an update reversible using worked examples, a deliberately broken case and a corrected version. The lesson distinguishes what the evidence establishes from what still needs checking.

Your practical activity: Record version changes, known constraints and a rollback route with official references.
How I’ll check your understanding: Submit the working artefact and a short evidence note for module 4: record version changes, known constraints and a rollback route with official references. Include the initial state, the chosen change, a meaningful failure case and the result of the check. I review whether the evidence supports the conclusion and return specific revisions where it does not.

PUT IT INTO PRACTICE

Your final project

Build and document a complete practical project for maintain AI tooling, model choices and long-running work. Use an invented organisation or an expressly permitted personal project, join the four module artefacts into one coherent result and demonstrate a normal journey, a failure state and a recovery or revision.

What you’ll produce

Module 1 artefact: Find model identifiers, settings and integration boundaries without exposing credentials.
Module 2 artefact: Build a small comparison set including useful failures and cost observations.
Module 3 artefact: Exercise resumed sessions, worktree recovery and durable decision notes.
Module 4 artefact: Record version changes, known constraints and a rollback route with official references.
A concise project brief, source and rights record, evidence log and handover with known limitations.

What good work looks like

Inventory saved assumptions: 20 marks for a correct artefact, justified decisions and repeatable evidence.
Evaluate representative work: 20 marks for a correct artefact, justified decisions and repeatable evidence.
Recover a longer task: 20 marks for a correct artefact, justified decisions and repeatable evidence.
Make an update reversible: 20 marks for a correct artefact, justified decisions and repeatable evidence.
Integration and handover: 20 marks for a coherent project, source acknowledgement, appropriate access and a clear explanation of remaining limits. Pass requires at least 70/100 and correction of any privacy, access or destructive-operation defect before completion.

Resources for your learning

A worked example and a deliberately broken fixture for inventory saved assumptions.
A worked example and a deliberately broken fixture for evaluate representative work.
A worked example and a deliberately broken fixture for recover a longer task.
A worked example and a deliberately broken fixture for make an update reversible.
A maintain ai tooling, model choices and long-running work workbook with evidence prompts and the capstone marking sheet.
A transcript, caption file and text-first lesson version for every recording.
A downloadable source-and-rights checklist and a sample completed project handover.

A course you can access

Provide edited captions, transcripts and a text-first route through every lesson.
Use keyboard-operable exercises, visible focus, legible contrast and no essential autoplay or motion.
Supply downloadable instructions and low-bandwidth alternatives; avoid timed assessment as a default.
Offer equivalent written evidence where live attendance or spoken presentation is a barrier and document support before enrolment.

FINISH WITH SOMETHING USEFUL

How you’ll complete the course

Submit all four module artefacts and respond to assessment feedback.
Complete the capstone at 70/100 or above with no unresolved access, privacy or destructive-operation defect.
Explain a meaningful failure case and the limits of the result in the evidence log.
Completion records describe assessed participation; I am not claiming external accreditation.

Completion will describe the work you have done. No external accreditation or professional qualification is claimed.

BEFORE I OPEN ENROLMENT

What I still need to finish

I’ll turn this plan into a delivered course once the teaching material and the learning service are ready.

Produce and editorially review the four lessons, worked examples, broken fixtures and marking guidance for maintain ai tooling, model choices and long-running work.
Build and test the course-specific workbook and capstone reference project using synthetic or permitted material.
Pilot the complete learner journey, including mobile, keyboard, captions, assessment feedback and recovery from interrupted access.
Approve the actual price, tax treatment, refund terms, access period, support boundaries and any cohort dates before accepting payment.
Implement and test authentication, server-side paid entitlements, webhook reconciliation, private asset access and tenant isolation before unlocking paid lessons.
Confirm tutor availability, feedback turnaround, moderation responsibilities and capacity before advertising tutor-led support.
Review source licences, sample rights and current official technical guidance before recording or publishing the material.

The ideas behind the course

My published articles are background reading for this plan. The full lessons, activities and assessment material still need producing.

EXPLORE THE ACADEMY

See how the learning space feels.

Try the fictional learner demo →