AI Use Cases
2026-11-05
When using AI, identify which parts of your work benefit most from your own expertise, judgment, and creativity. Prioritize doing those things yourself, and use AI to assist with the rest.
Use Cases
The following are some situations where I would recommend using AI. All of these come with a caveat that you should be reviewing all artifacts that the AI generates for accuracy and correctness; AI tools are a lot better than they were a year ago, but still often make mistakes.
- Proofreading: Write your content first, then use AI to proofread it for grammar and spelling errors. You can also use it to fact-check your content by finding supporting sources online that you can verify. Don't let it make major changes to your writing.
- Scripts / Personal Software: Generate scripts or personal software where you don't care about what the code looks like as long as it works. Just give the code a skim to make sure it isn't doing anything unexpected before using it.
- Converting Mockups to Code: Rather than letting the AI decide what the UI/UX is for your software, create a mockup and design the user experience and let the AI convert it into code. This keeps you in control of how your software looks and feels, while removing the tedious and boilerplate work.
- Converting Pseudocode to Code: Write high-level pseudocode, comment/function placeholders, or a rough draft of the code first so it stays true to your intended logic. Then use AI to convert it into working code. This keeps you in control of the business logic while saving time on the actual syntax and implementation details. If you prefer visuals, you could create a flowchart or diagram of your logic first and then use AI to convert that into code.
- Converting Code to Different Languages: Just tell it to convert your code to another programming language without changing the logic.
- Implementing Boilerplate or Framework Code: There is little value in writing repetitive boilerplate code yourself, so use AI to generate it for you or create wrappers around it matching an interface you define. The way I think about it is that this boilerplate/framework code is necessary but not where your software's value lies.
- Implementing Well-Known Algorithms: There is also little value in writing well-known algorithms from scratch and offloading this to AI lets you define business logic at a higher level.
- Generating Automated Tests: Use AI to generate automated tests for your code. Guide it with the expected behavior and edge cases you want to cover, but don't worry much about how the test code looks.
- Debugging: Give the AI a bug description and the relevant code and let it reproduce it, gather details about the root cause, and suggest a fix. Make sure you understand why the bug occurs and how the suggested fix resolves it. If you are learning about the system, it may be useful to do this without AI so you can build up your own understanding first.
- Reviewing Code: Use AI as a secondary reviewer for your code (as in, in addition to a human). The built-in code review skills in agent harnesses usually work well, though they can be nitpicky at times. Don't take its suggestions as written in stone. I recommend pointing it at your existing code as well; it may uncover bugs you have not encountered before.
- Optimization: Use AI to optimize your code on a metric you choose. Provide clear guidance on the bounds of the optimization so it doesn't interfere with your business logic and give it a target to aim for.
- Writing Documentation: Documentation and user guides are often just factual content with little creative input. Set up the expectations for the structure, style, and level of detail you want, and then use AI to generate the content. When you make a change to the software, it can automatically update the documentation for you.
- Challenging Requirements and Gap Analysis: Ask the AI to challenge your requirements and identify potential gaps or inconsistencies. If you use a skill like 'grill-me', you can have it ask a series of questions to settle edge cases. You can also use it to help combat scope creep by ensuring what you are planning on implementing aligns with the original requirements and objectives.
- Creating Proof of Concepts: Use AI to quickly generate proof of concepts for your ideas. This allows you to validate feasibility and explore different approaches without investing too much time in manual coding. Once you know something is possible, then you can refine it.
- Learning: Use AI to find resources, create learning plans, and test your understanding, but be cautious about it giving you bad information. If you don't know the topic well yourself, you may not be able to determine if the AI is providing inaccurate information.
Anti-Use Cases
The following are some situations where I would not recommend using AI.
- Generating Creative Work: Avoid letting AI replace your own creativity.
- Creating Requirements: While AI can help you refine requirements, you should be the one driving and creating them. The requirements define what gets built and that shouldn't be delegated to AI.
- Designing Architecture: A core part of software engineering is architecting the overall structure and design of the system. Let the AI fill in the implementation, but you should be the one making the high-level architectural decisions. The AI can help soundboard ideas, but it has a strong tendency to agree with you.
- Designing User Interfaces / User Experiences: As mentioned above, don't let the AI decide how people use your software.
- Human to Human Interaction: AI can change what you mean, mask your point behind a wall of text, misrepresent your abilities or intentions, make your message feel inauthentic, and insert false information. It is fine to have it proofread your messages, but don't let it write for you.
- Automating Enjoyable Tasks: If you enjoy doing certain tasks for fun, don't let the AI do them.
In short, use AI as a tool to increase your capabilities but not as a replacement for your expertise, judgment, and creativity.