FIT2109 Chap.12 Agentic Coding
Agentic Coding
Agentic Coding is governed by the current assessment rule, not by a tool's capability. The captured Assignment 2 specification permits chatbot-style assistance but bars agentic coding tools that directly write or apply changes in the editor or terminal.
It requires a record of prompts, responses and the student's explanation of what was accepted, changed or rejected.
Permitted assistance does not replace verification. Convert a suggestion into claims about current behaviour, expected behaviour and evidence. Compare those claims with the specification, docstrings, tests and observed output.
Make changes personally within the rule, run the relevant checks and be able to explain every accepted decision during the demonstration.
What this chapter covers
- 01
Assessment-specific rules checked before tool selection
- 02
Chatbot assistance distinguished from direct code-changing agents
- 03
Prompts and responses preserved for the required declaration
- 04
Student decisions to accept, change or reject suggestions
- 05
Project specifications and tests used as evidence
- 06
Personal verification of every accepted change
- 07
Sensitive information kept outside external tools
- 08
Demonstrated understanding aligned with the submitted artifact
Worked example · free
Reject advice that weakens the specification
- +1Identify the task statement that treats the supplied test as a correct behaviour contract.
- +1Reject deletion because it would weaken the specification rather than repair the implementation.
- +1Read the failing assertion, docstring and actual values.
- +1Reproduce the mismatch and locate the first incorrect implementation boundary.
- +1Repair the code and run the focused test plus the complete suite.
- +1Record the suggestion and the evidence-based rejection in the required declaration.
- +1Explain the cause and repair independently during the demonstration.
Key terms
- Agentic tool
- A tool capable of directly applying changes or actions in an editor or terminal. Assignment-specific rules may distinguish it from conversational assistance.
- Chatbot assistance
- A conversational interaction that proposes explanations or options without directly changing the project. Permitted use still requires review and declaration.
- Use declaration
- The required record of prompts, responses and the student's explanation of what was accepted, changed or rejected and why.
- Verification evidence
- Tests, observed program state, specification text and reproducible commands used to evaluate a suggestion rather than trusting plausibility.
- Tool boundary
- The assessment-specific line between permitted and prohibited forms of assistance, determined from the current task instructions.
- Personal ownership
- The ability to explain and demonstrate every submitted decision and change without depending on generated wording or hidden automated actions.
Agentic Coding FAQ
Are direct code-changing AI tools allowed for Assignment 2?
The captured specification says no. It permits chatbot-style tools under stated conditions but bars agentic coding tools that directly write or apply changes in the editor or terminal. Check the current task instructions before use.
What must be recorded when chatbot assistance is used?
Keep the prompt and response, then explain in your own words what you accepted, changed or rejected and why. The record should also identify how the resulting decision was verified.
How should I evaluate a generated code suggestion?
Turn it into a claim about behaviour, compare that claim with the specification and tests, inspect the proposed diff, run focused and wider checks, and reject any part that lacks evidence or violates the task rule.
Does declaring a prohibited tool make its use acceptable?
No. Transparency is required, but a declaration does not change the permission boundary. Check the rule first and seek course guidance if a proposed workflow is unclear.
Assessment move
Begin with the current individual task instructions and write a one-line permission boundary before opening any AI tool. If conversational assistance is permitted, ask narrow questions that expose alternatives or test ideas rather than delegating ownership of the repository. For every response, extract the claims, identify the evidence needed and decide whether to accept, modify or reject each suggestion.
Make and verify changes personally within the rule. Keep a contemporaneous declaration entry with the prompt, response, decision and verification. Practise explaining one accepted suggestion and one rejected suggestion from repository evidence. Confirm that no personal data, credentials or restricted material entered the interaction.
Before submission, compare the declaration, Git history and report so they tell a coherent development story that you can demonstrate independently. For one permitted interaction, preserve the original proposal and annotate each claim with the repository check that supported or refuted it. Rewrite the final explanation from observed behaviour rather than generated phrasing.
Rehearse the demonstration without the conversation open, locating the relevant code, test and commit from memory. This proves that assistance informed a decision without becoming a substitute for personal ownership.
Working through Agentic Coding in FIT2109? Sia is AskSia’s AI Computer Science tutor — ask any FIT2109 Agentic Coding question and get a clear, step-by-step explanation grounded in how FIT2109 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
