✳ IskraAI

Test Plan: How to Combine Manual Work and AI

Testing is not only about finding bugs, but also about thinking through scenarios that could break the system. Manual test plan creation takes time, and the AI tool "Test Plan" helps quickly generate a draft with successful, failed, and boundary cases. However, blindly trusting the machine is impossible: it is important to understand where automation ends and your responsibility begins. In this article, we will effectively combine manual work and AI using the example of a password reset form.

Ouvrir l'outil: Constructeur de plan de test ↗

What the "Test Plan" tool can do

"Test Plan" is a web tool that, based on your textual task description, generates a structured test plan. You enter an idea or functionality in the "Your text or idea" field, specify the platform or application in the "Platform or application" field and press "Generate". As a result, you get plain text with a set of test cases: successful scenarios, failed ones, and boundary conditions. The result can be copied or downloaded as TXT.

The tool does not require registration, does not automatically save data, and does not integrate with other systems. It works only with text: it does not support document uploads, OCR, export to DOCX or PDF, batch processing. It is important to consider when planning: if you need a test plan for a complex system with many integrations, you will have to supplement the result manually.

Input limits: first field up to 8000 characters, second up to 1000. The model can make mistakes, and queue and service availability vary. Therefore, the tool should be perceived as an assistant for generating ideas, not as a replacement for an analyst or tester.

Example input: "Password reset form, which sends a one-time link to email". Platform: "Web application". Result: a list of tests, including successful reset, invalid email, expired link, reuse of link.

Which parts of the task to assign to AI

AI is good at routine enumeration of options. It can be assigned to generate typical successful scenarios: for example, for a password reset form this is entering a correct email, following the link, setting a new password. The tool will also propose failed cases: non-existent email, empty field, incorrect format, password mismatch. Boundary conditions - password length, special characters, resending the letter - also often end up in the draft.

Delegate AI the primary structuring: breaking down into categories, formulating steps, expected result. This saves time at the brainstorming stage. But you should not ask the tool to set priorities or assess risks - for this you need project context, which you know better.

If you have a checklist or requirements, you can insert them into the input field so AI takes them into account. For example, indicate that the link is valid for 15 minutes, and the password must contain at least 8 characters. Then generated cases will be more accurate.

Input: "Password reset: email, link valid for 15 minutes, password from 8 characters, must contain a digit and a letter". AI will add checks for link expiration and password complexity.

What to check yourself

After generation, necessarily check that all key scenarios are covered. AI may miss specific cases for your product: for example, password reset for a user who is already logged in, or behavior when the account is blocked. Also check that there are no duplicate or unrealistic tests.

Manually evaluate boundary values. The tool may propose a password of 7 characters, but if the requirement is minimum 8, this is not a boundary, but a failure. Make sure boundaries correspond to your documentation. Check that negative scenarios with unexpected data are covered: emojis, spaces, SQL injections.

Manually set priorities: which tests are critical for release, which can be postponed. AI does not know your deadlines and risks. Also check step formulations - they must be unambiguous so any tester can reproduce the scenario.

AI suggested test "entering a 7-character password". You check the requirements: minimum 8, so this is a failed case, not a boundary. You fix it.

When the service is not suitable

The tool will not suit if you need a test plan for a system with many integrations, where important are API call sequences, database work, message queues. AI does not see the architecture and cannot propose comprehensive scenarios. In such cases it is better to use specialized tools or manual design.

Also do not rely on the service if high accuracy and security are required. For example, testing financial transactions or medical data: errors in the test plan can lead to missing critical bugs. Here an experienced tester and a strict process are needed.

If you already have a detailed test plan, AI will not give a significant gain. It is useful at early stages when you need to quickly sketch ideas. For regression testing or complex scenarios it is better to use proven checklists.

For testing bank transfer with balance check, limits and blocks AI will not account for all business rules. Better to make a plan manually.

Practical example: password reset form

Consider a fictional example: a web application with a password reset form. A user enters an email, receives a letter with a one-time link, follows it and sets a new password. We want to make a test plan with the help of AI.

We enter into the tool: "Password reset form, sends a one-time link to email. Link valid for 15 minutes. Password from 8 characters, must contain a digit and a letter". Platform: "Web application". We press "Generate".

The tool outputs a draft. Successful cases: correct email, follow link, enter valid password, successful reset. Failed: non-existent email, empty field, incorrect email format, password mismatch, weak password. Boundary: email with spaces, link after 14 minutes 59 seconds, password exactly 8 characters, reuse of link.

We check manually: we add a case when the user is already logged in and requests reset; we check that the link is single-use; we clarify that on repeated sending the old link is invalidated. We also add a check for SQL injection in the email field. We set priorities: critical - successful reset and link expiration.

Final test plan: 1. Successful reset. 2. Invalid email. 3. Empty field. 4. Link expired. 5. 7-character password. 6. 8-character password with digit and letter. 7. Reuse of link. 8. SQL injection.

How to check the result and limitations

After generation review each test case for compliance with requirements. Make sure steps are reproducible, expected result is unambiguous. Check that there are no tests that cannot be executed in your environment. For example, AI may suggest checking email, but you may not have access to the mailbox.

Remember the tool limitations: it does not save history, does not export to DOCX, does not support file uploads. The model can make mistakes, and the queue for generation may delay. Do not use the result without adaptation to your project.

The tool does not guarantee full coverage. It only helps to start. Always supplement the plan with your knowledge of the product, risks and business logic. Testing is a creative process, and AI here is only an assistant.

Check: if AI suggested test "entering email without @", make sure your validation really should reject such an address. If yes - the test is correct.

The AI "Test Plan" tool accelerates the creation of a draft test scenario, but does not replace analytics. Delegate to it the generation of typical cases, and manually check boundaries, priorities, and specifics. Use the service at early stages when you need to quickly get ideas. Remember the limitations: text up to 8000 characters, no integrations, possible model errors.

Ouvrir l'outil: Constructeur de plan de test ↗

Les traductions de l'interface sont assistées par machine et peuvent nécessiter des améliorations.