When to reach for AI tools: Real scenarios

November 20, 2025

Part of the Claude Code in Practice series

Series outline: 1. Why Claude Code is different • 2. How Claude reads your codebase • 3. Writing tasks Claude can complete • 4. Plan mode workflow • 5. Real-world scenarios • 6. MCP and skills • 7. Optimizing for speed • 8. Building your workflow

One of the most common questions I hear from engineers: “Should I use Claude for this?” The answer is not always yes. Claude Code is powerful, but using it for everything slows you down. Knowing when to reach for AI versus traditional tools is a critical skill for productive engineers. Let me walk through five scenarios that illustrate when each approach works best.

Scenario 1: Refactoring patterns across multiple files

Situation: You need to replace all instances of useState with useReducer in your form components because you are standardizing on a reducer-based pattern.

How Claude Code handles it: This is where Claude excels. In plan mode, Claude searches for form components, identifies all the useState calls, understands the state structure, and creates a plan to migrate each component to useReducer. You review the plan, approve it, and Claude systematically updates each file. The plan mode workflow means you catch any edge cases before they break your forms.

Verdict: ✅ Excellent use case. The combination of systematic search, pattern understanding, and reviewable plans makes multi-file refactoring much faster than doing it manually.

Scenario 2: Debugging a specific error

Situation: Users report getting a “Cannot read property ‘id’ of undefined” error on the checkout page, but you cannot reproduce it locally.

How Claude Code handles it: Claude starts by searching for the error message and exploring the checkout code. It reads the relevant components, traces the data flow, identifies where ‘id’ is accessed, and finds that the error happens when a specific payment method object is missing. The plan shows you the root cause and proposes a fix with proper null checking. You review, verify the logic makes sense, and approve.

Verdict: ✅ Excellent for debugging with context. The exploration → plan → fix workflow helps you understand the problem before applying a solution. Much better than randomly adding null checks until the error goes away.

Scenario 3: Architecting a new feature

Situation: You need to add a notification system to your app. It should support multiple notification types, toast messages, and persistence.

How Claude Code handles it: Plan mode helps with exploration and initial structure, but falls short on architectural decisions. Claude can research existing patterns in your codebase and propose an approach, but it cannot make judgment calls about trade-offs. Should notifications use a global context or a separate service? Should they persist to localStorage or a backend? These are decisions you need to make.

Verdict: ⚠️ Mixed results. Use Claude to explore options and draft initial implementations, but drive the architectural decisions yourself. The plan mode workflow is perfect for this: review Claude’s architectural suggestions critically before approving execution.

Scenario 4: Learning an unfamiliar framework

Situation: Your project uses a framework you have never worked with before, and you need to add a new feature using that framework’s patterns.

How Claude Code handles it: Claude reads your existing code, identifies the framework patterns, and proposes implementations that match those patterns. For well-known frameworks, this works well. Claude can spot “this project uses React Query for data fetching” and apply the same patterns to your new feature.

However, you need to verify the results. Claude sometimes mixes patterns from different versions or makes assumptions about your setup. The iterative workflow helps: implement one piece, verify it works, then move to the next.

Verdict: ⚠️ Good but verify everything. Use plan mode to review proposed approaches before execution, and test thoroughly. Claude is helpful for learning patterns faster, but it is not a substitute for understanding the framework yourself.

Scenario 5: Quick one-line fixes

Situation: You spot a typo in a variable name or need to add a missing import.

How Claude Code handles it: Technically Claude can do this, but by the time you describe the task, wait for exploration, review the plan, and approve execution, you could have fixed it yourself three times over.

Verdict: ❌ Use your editor. Claude Code shines on tasks that require context and multiple changes. For trivial edits, traditional tools are faster.

The pattern

Looking across these scenarios, a pattern emerges:

The sweet spot for Claude Code is tasks that involve multiple files, require understanding relationships, and benefit from systematic exploration. Plan mode makes these tasks reliable by letting you review before execution.

When to reach for Claude Code

Use Claude Code when:

• The task involves more than 2-3 files
• You need to understand how pieces connect
• You want systematic changes across patterns
• The implementation is tedious but not conceptually hard

Stick with traditional tools when:

• The change is obvious and localized
• You need to make a quick fix
• The task is more about thinking than typing

The key is recognizing that Claude Code is not about replacing your editor. It is about delegating tedious, multi-file work while you focus on decisions and verification. Plan mode makes this partnership work by keeping you in control of what actually gets implemented.

Next, let’s look at how to extend Claude Code with MCP servers and custom skills to make it even more powerful.

Previous: ← Plan mode workflow | Next: MCP and skills →