Skip to content

Improve agentic versus ask mode learning hour #84

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
emilybache merged 1 commit into from
May 15, 2026
Merged

Improve agentic versus ask mode learning hour #84

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 28 additions & 9 deletions _learning_hours/agentic_engineering/agentic_vs_ask_mode.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,21 +1,23 @@
---
theme: agentic_engineering
name: agentic_vs_ask_mode
title: Agentic vs Ask Mode
title: From Ask Mode to Agent Mode
kata: WarehouseDeskApp
difficulty: 1
author: larseckart
via: emilybache
tags: agentic tdd
tags: agentic copilot codex claude
---

# {{ page.title}}

This is a first learning hour for people who have not yet fully understood the difference between agentic AI and the previous forms of AI that were available to them like "Ask mode" and autocomplete in the IDE.
This is a first learning hour for people who have mainly used AI-powered inline suggestions, [ChatGPT](https://chatgpt.com/) or GitHub Copilot in *Ask mode*.
It is meant as an introduction for tools like [GitHub Copilot](https://github.com/features/copilot) Agent mode, [Codex](https://developers.openai.com/codex), [Claude Code](https://claude.com/product/claude-code), [Amp](https://ampcode.com/), [Pi](https://pi.dev/), [OpenCode](https://opencode.ai/) etc.
For the remainder of the Learning Hour, we'll use GitHub Copilot as an example.

## Learning Goals

* Describe how Agentic AI differs from AI "Ask mode" or autocomplete.
* Describe how GitHub Copilot Agent mode differs from "Ask mode" or AI-powered autocomplete.
* Use an AI Agent to inspect, change and verify code via prompting.

## Session Outline
Expand All @@ -34,17 +36,34 @@ Show the relationship diagram and explain it:

![agent loop.png](/assets/images/agent%20loop.png)

The numbers indicate the order that things happen in.
The agent loop is the basic pattern used by AI coding agents.
The agent takes user input and forwards it to the model.
When a response comes back, it usually contains tool call instructions.
The agent then uses its tools, like reading and editing files or executing bash commands such as `npm test`, and feeds the results back into the model.
This repeats until the model’s response contains no more tool calls.
Then the final response is presented to the user, and the agent waits for input again.

Agents also use additional instructions, for example from a system prompt, available skills, MCP configuration or the content of the `AGENTS.md` file.
We leave those out for this learning hour, so we can focus on the basic agent loop without introducing too many extra concepts.

## Demo - Using an AI Agent to refactor
Open a small project like [WarehouseDeskApp](https://github.com/LarsEckart/harness-engineering-kata.git) that has some poorly written code that could be improved by refactoring. Show how to prompt the agent to achieve that.
Open a small project like [WarehouseDeskApp](https://github.com/LarsEckart/harness-engineering-kata.git) that has some poorly written code that could be improved by refactoring.
Briefly explain the interface of the coding agent, highlight model choice and approval settings, and in case of Copilot, how to switch from Ask mode to Agent mode.

For the harness-engineering-kata, good prompts for a demo can be:

> Build this project, verify that it compiles and what it prints to the console.

> Refactor, introduce a value object for quantity. Compile the code afterwards and verify that the output stays the same.

Show them the AI settings so that it will use an Agent Loop that includes compiling and/or linting. Keep it to the bare minimum of choices so the tool details don't overwhelm them. For example, use the 'auto' LLM selection.
> Commit the changes we made. Follow the project's commit notation.

Show the normal workflow you recommend, which probably means asking it to improve the design in a specific way, inspecting changes, and using the AI to generate a commit message. The focus should be on the Agent loop and how it can check for itself whether it is making progress with the task you have set.
The focus should be on the Agent loop and how tools are used that allow the agent to verify its work.

## Concrete Practice - Use an AI agent
Split into pairs and give them the same code that you showed in the demo, with the same AI tooling already set up. Ask them to do similar design improvement tasks using the Agentic setup. Go round and help if anyone gets stuck and answer questions - try to get every pair to the point where the agent has made acceptable code changes without manual intervention, and they were prepared to commit the results.
Split into pairs and give them the same code that you showed in the demo, with the same AI tooling already set up.
Ask them to do similar design improvement tasks using the Agentic setup.
Go round and help if anyone gets stuck and answer questions - try to get every pair to the point where the agent has made acceptable code changes without manual intervention, and they were prepared to commit the results.

## Conclusions
Ask people to [explain the main idea]({% link _activities/conclusions/explain_main_idea.md %}) with a question similar to this:
Expand Down