Author-led mentored course · 36 academic hours · 9 sessions
Build a personal workspace powered by OpenAI Codex and Claude Code: from defining a task to producing lectures and reviews, assessing student work and transferring ready-to-use processes to colleagues. No programming required. Built around your real project.
Every session ends with a working artefact in your repository — from a task brief to a workspace that can be handed over to a colleague.
Профессиональные задачи
The course is built around the real work of people who create, evaluate and transfer knowledge.
A lecture and practical session that can be updated in hours rather than days; an AI-use policy and an agent for preliminary review of student work — while the final assessment remains yours.
An evidence-based review with a source registry and a ‘claim — evidence’ map; a reproducible research process that stands up to scrutiny.
Materials with instructions, criteria and a runbook can be transferred without weeks of verbal explanation; a foundation for departmental standards.
No programming experience is required: every session has an operator track using ready-made templates and a builder track for technically confident participants.
От ответа к системе
Most educators have already tried ChatGPT: a quick draft, polished prose — and a fabricated reference in the bibliography. A one-off conversation does not provide what matters most: verifiability, repeatability and the ability to transfer the result. This course teaches a different way of working.
The agent proposes and executes. Fact-checking, pedagogical decisions and accountability remain with the human. The course teaches this balance.
The first session begins with your real task
ApplyРезультат в репозитории
Not a collection of chat responses, but seven connected working artefacts that remain in your repository.
A repository with instructions, project memory and a decision log: the agent works according to your rules, not the other way around.
AGENTS.mdA delegation matrix, agent task queue and checkpoints: measure the real time saved, rather than relying on a sense of moving faster.
agent-queue.yamlA source registry and claims map; a script catches fabricated references before your students or a reviewer sees them.
check_sources.pyEvery claim can be traced to its source; adapting the material for a new audience becomes a procedure rather than a rewrite.
lecture-manifest.yamlA packaged method that can be repeated and transferred: a rubric, reference solution and counterexamples.
SKILL.mdAn AI-use policy, a process evidence package and an agent for preliminary review; the educator makes the academic decision.
AI_USE_POLICY.mdThe course's final outcome: a workspace that a colleague can reproduce without you, tested through a ‘cold handoff’ during final assessment.
cold-handoff-report.mdEvery artefact remains yours and continues working after the course
ApplyВитрина артефактов
Four professional scenarios and real files that you will work with.
Instead of rewriting it, create a TASK.md task brief with acceptance criteria; the agent reads your materials, proposes a plan and clearly marks anything missing as [SOURCE REQUIRED].
TASK.md
A sources.csv registry, a claims.csv claims map and a validation script that finds a phantom reference in seconds.
sources.csv · claims.csv
The student submits the text plus a process evidence package; the agent checks completeness, identifies contradictions and prepares questions for the defence. The decision remains yours.
submission-manifest.yaml
A manifest, runbook and ‘cold handoff’: the recipient reproduces the result without asking a single verbal question.
manifest.yaml · RUNBOOK.mdFees and the next available dates will be provided in response to your application
Apply3 modules · 9 sessions
We examine the limitations of a one-off chat and turn a real task into a verifiable project. We establish a baseline benchmark, acceptance criteria and delegation risks.
Ключевой результат: A real-task brief, baseline benchmark and risk map
We assemble the workspace structure, persistent agent instructions and a decision log. We establish a short cycle: one task, one check, one documented result.
Ключевой результат: A workspace structure, persistent instructions and an atomic workflow
We break down the working week by delegation type and create an agent task queue. We compare time spent before and after using real data.
Ключевой результат: A delegation matrix and personal agent-enabled working week
We connect the search protocol, source registry and claims map. A validation script detects phantom references and gaps between conclusions and evidence.
Ключевой результат: An evidence-based review with a source registry and claims map
We turn a lecture from a single document into a package of connected artefacts. Every claim and slide can be traced to a source and the session plan.
Ключевой результат: A lecture and presentation linked to an evidence base
We design a practical session, rubric and reference answer, then package the repeatable method as a Skill. We test it on both a standard and a problematic case.
Ключевой результат: A practical session, rubric, short-course map and tested Skill
We define AI-use rules and the components of a process evidence package. The preliminary-review agent prepares flags and questions but does not replace the educator's assessment.
Ключевой результат: An AI-use policy, evidence package and agent for preliminary review
We separate the roles of producer, critic and integrator. We formalise the handoff through contracts, permissions and a measurable acceptance report.
Ключевой результат: Handoff contracts, roles and a measurable group workflow
We assemble the final workspace and hand it over to a colleague without verbal guidance. An independent run reveals what is truly reproducible and what requires improvement.
Ключевой результат: A transferable workspace and an independent reproducibility report
How learning works
Each session is structured as a production cycle, not as a lecture about what AI can do.
One continuous personal project connects all nine sessions: each new artefact builds on the previous one.
The final assessment is a cold handoff: a colleague reproduces your workspace without verbal guidance.
Operator track: work with ready-made templates. Builder track: adapt instructions, scripts and workflows to your own practice.
Interested in a departmental course? Let’s discuss it in your application
ApplyFAQ
No. Every session includes an operator track with ready-to-use commands and templates. The builder track is available to participants who want to modify scripts and workflow architecture, but it is not required to complete the course.
You will need a laptop, Git and access to OpenAI Codex or Claude Code. Before the course, you will receive a short preparation checklist; during the first session, we will check your workspace and baseline benchmark.
Bring one real professional task and materials you already use: a lecture plan, a set of sources, a practical exercise or a student-work package. The course is built around your project, not a disposable training example.
A prompt is only one element of the process. The main outcome is a repository with context, instructions, criteria, checks and a decision history that can be reproduced six months later and handed over to a colleague.
Yes. The course can be tailored to your department's materials and AI-use policy. State the size of your team and its task in the application, and the author will suggest an appropriate format for the discussion.
Your continuous project. Ready-made course demonstration packages are used when necessary, but the final instructions, rubrics and workflows are adapted to your professional context.
All the files you create remain yours: the workspace, evidence-based review, lecture package, practical session, AI-use policy and handoff package. They are stored in your repository and updated through your normal working process.
You hand a mirror copy of your workspace to a colleague, who tries to reproduce the result without verbal guidance. The cold-handoff report identifies gaps and becomes the plan for the final revisions.
Fees and the next available dates will be provided in response to your application
You will receive a reply within 1–2 working days. No newsletters or spam.
$ course apply --project=my-work→ заявка без рассылок и спама