TRAINING
Hands-on AI-Assisted Development
IT track
Prepared for TechnoPro, Inc. / Givery, Inc.
3 days, 8 hours each. Most of the time is hands-on.
This course goes beyond having generative AI do the work and stopping there. It covers checking what comes out yourself before you use it. The aim over the 3 days is that you can build something that works by giving instructions in English, and verify the result with your own eyes.
Each day's goals and how the time is used. Each day is [480min], plus the lunch break. The gap between Session 1 and Session 2 is 2 weeks, and the gap between Session 2 and Session 3 is 1 week.
| [100min] | Environment setup (VS Code, GitHub account, Copilot invitation, connection check) |
| [30min] | The world of AI-driven development (how LLMs and agents work) |
| [95min] | Copilot basics (Ask, Agent, Plan, Keep), handling confidential information, and prompt patterns |
| [10min] | Break |
| [60min] | A working web app from a 5-line spec (vibe coding, checking in the browser) |
| [10min] | Break |
| [75min] | Adding features to an existing app (have the AI read it first → add filtering and totals → cross-check) |
| [10min] | Break |
| [90min] | Choosing the right model, rolling out instructions files across the team, and review |
| [60min] | Extracting the handout ZIP, and debugging with AI (break one line, tell the AI only the symptom, and have it fix the bug) |
| [37min] | Dividing test work between humans and AI |
| [10min] | Break |
| [114min] | Extracting test viewpoints and generating test cases (the test case table for the features you added on Day 1) |
| [60min] | Test design basics (source of expected results, requirement IDs, decision tables) |
| [10min] | Break |
| [70min] | Test data and test report |
| [10min] | Break |
| [73min] | Implementing the import and unit tests (a person judges the tests the AI wrote) |
| [27min] | Gaps in display and export (security) |
| [9min] | Review and a preview of Day 3 |
| [10min] | Review of Day 2 and preparation for today |
| [90min] | E2E tests that run in the browser alone (read it, break it to check, and add execution results to the test report) |
| [10min] | Break |
| [120min] | Locking in the steps and the reviewer (skills, instructions, agents) |
| [40min] | Certification and Literacy Check |
| [10min] | Break |
| [120min] | Practical exam (import the September data in the new format and submit your decisions) |
| [10min] | Break |
| [60min] | Feedback session, responding to comments, and the Certificate of Completion |
| [10min] | The 3 days' deliverables and next month's steps |
There are only 3, and each has a clear role.
| Tool | Role |
|---|---|
| VSCode | This is where you write code and text. Version 1.116 or later comes with Copilot Chat built in, so Copilot Chat is there from the start. |
| GitHub Copilot | A partner you can consult inside VS Code. You can make your requests in English. |
| Browser (such as Edge) | This is where you run what you built and check it. To load data, drag the file onto the page and drop it. |
The only thing to do before the day is to get your GitHub account ready.
Slides and handouts for each session are gathered here.
After setting up your environment, you build a working web app from a 5-line spec. In the late afternoon, you have the AI read the Worklog app from the handouts first, then add 2 features and put instructions files in place.
Using the app you extended on Day 1, you start with debugging. Then you build a test case table and test data from the spec, implement the import and unit tests, and cover gaps in the display.
You run an E2E test from import to reload, and move repeated steps into skills and agents. In the afternoon, you finish the import of the new format in the practical exam.
These cover topics left out of the main course for lack of time, plus a cheat sheet of the files you create in the course. Read them before or after the course.
There is one ZIP per session. Extracting it creates a folder named tpro-work. Put the folder on your desktop.
| Session | For Windows | For Mac |
|---|---|---|
| Session 1 | IT_Day1_handouts.zip | ZIP for Mac |
| Session 2 | IT_Day2_handouts.zip | ZIP for Mac |
| Session 3 | IT_Day3_handouts.zip | ZIP for Mac |
| Harness Cheat Sheet | Harness Cheat Sheet (PDF, one A4 landscape page) | |
As participants, you handle information from your client sites. In the exercises, assume the following.
No. The course works with only VS Code and a browser. Data aggregation and charts both run in the browser, using the libraries bundled in the handouts.
The course assumes participants with development experience, but it is structured so that you can follow it even if you are new to using AI.
Yes. The files you create stay on your own device. However, there are limits on what you can submit. See the next item.
You submit only the text you have put into words. You do not submit code or real data. We will explain how to submit on the day.
Raise your hand. Each exercise has an "If you're stuck" section. You do not have to finish within the time. Finishing everything is not the goal.
Whether you can use it, and to what extent, depends on the agreements with your own company and with the client site where you work. We cannot give a definite answer during the course, so please check through your company's procedures. Here is only how the cost works. GitHub Copilot has a free plan, but it has a usage limit and is not enough for everyday work. Paid plans for individuals start at $10 per month, and the Business plan for organizations is $19 per seat per month. In June 2026, the billing method changed to GitHub AI Credits, which are used up according to the model and the amount of tokens you use. Inline suggestions are not billed, so credits hardly decrease if you mainly use them. (Source: GitHub Docs, checked on September 9, 2026)
If a file name is merely listed in the Explorer, nothing is sent. Sending starts when you open the file in the editor or select text in it. VS Code attaches the open file and your selection to the conversation even if you do not specify them.
Take care with how .gitignore works. Even if a file is listed as excluded, opening that file bypasses the exclusion. "It is safe because I listed it as excluded" does not hold.
Files are also sent for a question such as "What does this project do?" VS Code keeps an index of the workspace and automatically picks up related files according to what the conversation is about. The official documentation says that content matched by a search is included in the conversation even if the AI does not open the file.
The sure way is not to open a folder that contains confidential files in VS Code. If you have to open it, deal with it in this order: first move the confidential files to another folder so that the workspace does not include them, then use files.exclude to hide them from the Explorer so that you do not open them by accident. (Source: VS Code Docs, checked on September 9, 2026)
Here are cases where the company name and figures are public. Hitachi ran an internal evaluation with about 200 people, and 83% said "I can complete tasks quickly." Fujitsu had over 10,000 users and reduced workload by 20%. Toshiba Tec ran a PoC with about 300 people. Kakaku.com rolled it out to all developers after a trial. In talks with clients, figures from a company of a similar size are more persuasive than the name of a large company.
These 3 are a good starting point: Workspace index and file handling, Custom instructions files, and Plans and billing. The first is the page cited in the confidential files question above.
These are the terms you will meet during the course. If one is unclear, check it here before you move on.