Testing on Work You Can Explain

From Autocomplete to Agents: How AI Coding Methods Are Evolving

Thu, 30 Jul 2026 00:00:00 -0500

AI coding methods are moving from local suggestions toward repository-aware, tool-using, and delegated work. The important change is not that typing code disappears. It is that a person can ask a system to take more actions across a larger part of the software process.

As the action surface grows, the durable human work becomes more explicit:

define the behavior;
select the context;
state permissions and boundaries;
create acceptance checks;
inspect actions and diffs;
integrate the result; and
explain why the final software should be trusted.

Use the smallest assistance level that fits the task. More autonomy is useful only when its saved effort exceeds the cost of context, supervision, review, and recovery.

A Beginner’s First Programming Workflow

Fri, 24 Jul 2026 00:00:00 -0500

Use one repeatable loop for beginner programming work: understand the task, plan one small change, change the code, run it, test the expected behavior, and create the checkpoint required by the project. A Git project can use a commit; another course may require a working-log entry, saved version, or submission. Each step produces evidence for the next.

The loop prevents a vague assignment from becoming a large, unexplained edit.

1. Understand the task

Before editing, state:

A Project Structure That Encourages Good Habits

Fri, 24 Jul 2026 00:00:00 -0500

A useful project structure makes the next correct action easy to find. Keep source, tests, documentation, scripts, and examples in predictable locations. Start small, name each directory by responsibility, and add complexity only when the project needs it.

The goal is not a universal folder tree. It is a project another person can inspect, run, test, and explain.

Begin at one clear project root

The project root contains the files that describe the project as a unit:

AI Can Be Confident and Wrong

Fri, 24 Jul 2026 00:00:00 -0500

AI confidence is a feature of presentation, not evidence of correctness. Generated output can contain false citations, nonexistent APIs, code that works only for the example, hidden assumptions, or unnecessary complexity.

Break the response into claims. Verify factual claims in primary sources and technical behavior with official documentation and reproducible tests. Mark unsupported claims as uncertain or reject them.

Confabulation can sound specific

NIST’s Generative AI Profile uses confabulation for erroneous or false content that a generative AI system presents confidently. The report notes that output can also contradict the prompt or earlier output.

Automate Deterministic Work with Code

Fri, 24 Jul 2026 00:00:00 -0500

Use code instead of repeated AI prompts when the same explicit inputs should produce defined outputs through known operations. Keep AI at the interpretation boundary: it can help explain an unfamiliar case or draft code, but deterministic execution should be inspectable, tested, and repeatable.

Define the contract

Before scripting, write:

Inputs:
Outputs:
Files or systems changed:
Valid and invalid conditions:
Expected failures:
Verification:
Rollback:

For example, renaming assignment files by a documented convention can be deterministic. Deciding whether a student’s reflection demonstrates understanding requires judgment and should not be hidden in a rename script.

Define “Done” Before You Start

Fri, 24 Jul 2026 00:00:00 -0500

Define “done” before implementation. A useful definition names the scope, expected behavior, tests, documentation, and submission or deployment evidence. It turns “work on the project” into claims that can be checked.

Separate outcome from activity

“Spend three hours coding” is an activity. “Given a valid CSV file, produce a summary containing row count and the mean of the score column” is an outcome.

Start with:

The project is done when [reader or user] can [observable outcome] under
[conditions], and the result is verified by [evidence].

Do not write criteria after seeing what the current implementation happens to do. That turns the checklist into a description, not a target.

How to Review AI-Generated Code

Fri, 24 Jul 2026 00:00:00 -0500

Treat AI-generated code as an untrusted proposal. Read every line before execution, compare it with the specification, identify inputs, outputs, assumptions, dependencies, and effects, then test expected, boundary, and invalid cases.

Do not accept code you cannot explain. Passing one sample is evidence about one sample, not proof of general correctness or safety.

Confirm that generated code is allowed

Course policies can distinguish among explanation, completion, code generation, review, and disclosure. Before generating or using code, determine:

Test the Expected, the Boundary, and the Invalid

Fri, 24 Jul 2026 00:00:00 -0500

A basic meaningful test set covers three categories: expected use, important boundaries, and invalid input. This does not prove a program correct, but it tests more than one friendly example.

Begin with a specification

Suppose a function calculates the mean of a non-empty list of scores from 0 through 100. It returns a numeric mean and rejects an empty list, nonnumeric values, and values outside the range.

The specification defines the tests. Without it, you may only confirm whatever the current code happens to do.