When to Use AI vs. When to Think

TL;DR

AI accelerates tasks with clear inputs and outputs: boilerplate, fast feedback loops (front-end, unit tests, REST APIs). It struggles with novel problems, long build cycles, and "it depends."
The 70% problem: not all tasks benefit equally. Use AI to augment routine work; maintain critical thinking for complex decisions and learning. Heavy AI use compounds tech debt when you skip review.
When in doubt: use AI to explore, you to decide. Core logic and security configs remain human domains.

Not everything should go through AI. Some tasks get worse when you delegate to a model. Knowing the difference saves time and prevents bad outcomes.

The Quick Decision Test

Use AI when:

The task has a clear format (code, doc, summary, test). Routine, well-understood work.
Fast feedback loops: front-end, unit tests, web/mobile dev, standard back-end. Popular libraries with lots of online examples.
Rapid prototyping: designs → working prototypes, concept → MVP in hours.
You can verify the output (you know the domain). Objectives are well-specified.
The stakes of a wrong answer are low or recoverable.

Think it through (or think first, then use AI) when:

Novel or complex problems: AI struggles with novel projects, deep architectural understanding.
Long build cycles: slow feedback (e.g., 2-hour deploys) makes AI dependency harder.
Learning fundamentals: relying on AI prevents understanding core concepts.
Avoiding technical debt: heavy AI use compounds quality concerns when you skip review.
The answer depends on context AI doesn't have (org politics, team capacity, legacy constraints).
The stakes are high (architecture, security, hiring, legal). Core logic and security configs stay human.

The "It Depends" Trap

AI loves to give a definitive answer. But many real problems are "it depends on X, Y, Z." When you ask "Should we use microservices or a monolith?", AI will give you a balanced essay. It won't know your team size, your deployment cadence, or your tolerance for operational complexity.

Use AI to list options and trade-offs. You supply the context and make the call.

When AI Actively Hurts

Premature convergence — AI gives you an answer. You stop exploring. Sometimes the best answer is the third one you hadn't thought of.
False confidence — AI sounds sure. You stop questioning. Verify.
Context blindness — AI doesn't know your codebase, your customers, or your org. Generic advice can be wrong for you.
Ethics and judgment — AI can't weigh fairness, culture fit, or "how will this play in the all-hands?"

A Simple Framework

Situation	Approach
"Write a test for this function"	Use AI. Review. Ship.
"What's the best way to structure our event pipeline?"	Think first. Use AI to brainstorm options. You choose.
"Draft a status update for the exec team"	Use AI for draft. You add nuance, politics, tone.
"Should we refactor this now or later?"	Think. AI can't weigh opportunity cost, team capacity, or risk.
"Explain this error"	Use AI. Then verify the explanation matches your setup.

You spend 45 minutes writing a status update email. You research the topic manually. You draft, edit, re-draft. You second-guess the tone. You send it and immediately think of something you forgot.

Click "With AI" to see the difference →

Quick Check

Your team is debating whether to refactor a critical payment service or build a new feature. What's the best use of AI here?

Do This Next

Audit one decision you made this week. Would AI have helped? Would it have led you astray? Reflect. Note whether it had fast feedback (AI-friendly) or slow cycles (think first).
Create a personal "AI or me?" rule of thumb for your role. E.g., "For technical decisions with 3+ options, I think first, then use AI to stress-test. For boilerplate and prototypes, I ship AI drafts and review."