You are here:Workshops IndexCourse Index][Workshop A][ Appendices The main course explains how modern generative AI works, where it helps, where it fails, and how to use it without drifting into vibe-coding. The workshops turn those ideas into controlled practice.
These are small labs for technical professionals who want to test AI tools against real working conditions. They cover comparing coding environments, writing repository guidance, building retrieval-backed workflows, and designing AI-assisted code review without treating the AI as the reviewer of record.
Each workshop can stand on its own, although it will be more useful after the related course parts. The purpose is to develop judgment that transfers across tools and projects rather than memorize one product-specific workflow.
The workshops can be read as guided articles, but they are designed to be performed.
Bring a real repository, a small document set, a pull request, or a realistic technical task. Avoid inputs invented only to make the exercise easy. Keep the scope narrow, make the acceptance criteria visible, and treat every AI output as draft material until it has been checked.
The working pattern remains consistent:
This is how an experiment becomes a repeatable professional practice.
Most AI coding tool comparisons are uneven before they begin. One tool receives a full repository, another gets a pasted snippet, and a third is judged mainly by autocomplete. The resulting ranking says little about how the tools would perform under comparable conditions.
This workshop runs the same bounded engineering task through several AI coding environments using shared acceptance criteria. It examines how each tool uses context, plans work, edits files, exposes assumptions, handles validation, and affects the reviewer’s workload. The outcome is a clearer view of tool fit for a particular task shape, not a universal winner.
Best read after
What you will practice
What you should leave with
A practical scorecard for comparing AI coding tools by task shape rather than brand preference.
When an AI tool repeatedly misunderstands a project, the model may be only part of the problem. Important expectations often live in tribal knowledge, scattered comments, old wiki pages, and the heads of a few experienced developers.
This workshop turns that hidden context into explicit repository guidance. You will document build and test commands, coding conventions, architectural boundaries, risky areas, generated files, review expectations, and the conditions under which an AI assistant should stop and request human input.
The result is an operational artifact intended to make AI-assisted work more predictable around the codebase. It should be concise enough to maintain and specific enough to influence behavior.
Best read after
What you will practice
What you should leave with
A first version of an AI-facing repository guide that can be adapted for Copilot instructions, agent guidance files, Claude Code context, Codex workflows, or internal team standards.
Teams often reach for the term RAG when their immediate need is simpler: the AI should answer from a known document set, cite the material it used, and make it possible to check whether the answer is supported.
This workshop keeps retrieval deliberately small. You will work with a limited document set, compare answers produced with and without retrieval, inspect source relevance, and run a simple evaluation pass.
The exercise avoids enterprise search architecture so that the important mechanics remain visible: document selection, chunk quality, irrelevant context, unsupported claims, and the difference between a plausible answer and a grounded one.
Best read after
What you will practice
What you should leave with
A grounded-answering workflow that is small enough to inspect directly and realistic enough to expose the main tradeoffs of retrieval.
AI-assisted code review is useful when its role is explicit. It can summarize a change, flag suspicious details, classify possible issues, and direct human attention. It should not become the authority that decides whether code is safe to merge.
This workshop designs a pull request review pipeline with defined limits. You will specify what the AI should inspect, what it must not decide, how findings should be structured, how severity and confidence should be represented, and how human reviewers should consume the output.
The pipeline adds another review signal while preserving human responsibility for validation, risk assessment, and the merge decision.
Best read after
What you will practice
What you should leave with
A review-pipeline blueprint that treats AI as an additional source of evidence rather than delegated engineering judgment.
For a linear reading of the course, it’s perfectly fine to complete the workshops after Part 20. To use them alongside the main course, follow this sequence:
This sequence follows the course’s learning arc: grounding and evaluation, tool comparison, repository guidance, and review discipline.
The workshops develop a repeatable operating habit. When AI enters a technical workflow, the team should be able to explain what context it receives, what actions it may take, how its output will be checked, and who remains responsible for the result.
That is the standard applied throughout the workshop series.
AI without illusions: Workshops was originally published in Stackademic on Medium, where people are continuing the conversation by highlighting and responding to this story.