Every developer I talk to in 2025 is either all-in on AI tools or completely ignoring them. Both extremes are wrong. Here’s my honest, opinionated take after 18 months of daily use.
The Tier List
Tier S: Daily Drivers (Use These)
Cursor IDE — The only AI coding editor that actually understands your codebase. The tab autocomplete is good, but the real power is Cmd+K (inline edits) and the Composer (multi-file changes). If you’re not using Cursor, you’re leaving 20–30% productivity on the table.
Claude (Sonnet/Opus) — For complex reasoning: architecture reviews, debugging weird production issues, writing technical docs. Better than GPT-4 at following nuanced multi-step instructions. My workflow: Claude for thinking, Cursor for implementation.
GitHub Copilot — Still useful for boilerplate in languages where Cursor’s context is weaker (Terraform, YAML, regex). Good fallback.
Tier A: Situational Power
ChatGPT (GPT-4o) — Better for creative/marketing copy, image generation workflows, and situations where you want a different perspective from Claude. Less reliable for long technical chains.
Perplexity — AI-powered search with citations. Use it when you need current information (library versions, recent API changes) that Claude’s training data might miss.
v0 (Vercel) — Generates React UI components from text descriptions. Surprisingly good for prototyping. Save it for when you need to build a UI fast and don’t want to write CSS.
Tier B: Interesting but Overhyped
Devin/SWE-agent style tools — Autonomous coding agents that can “do the whole task.” In practice, they work well for isolated, well-defined tasks and poorly for anything requiring real domain knowledge. Don’t expect them to replace engineers yet.
AI code reviewers — Tools that auto-comment on PRs. They catch obvious issues (null checks, naming) but miss architectural problems. Useful as a first pass, not a replacement for human review.
How to Actually Get Value (The Workflow)
The mistake most developers make is using AI as a “query engine” — you ask a question, get an answer, use it. That’s the lowest-value pattern.
The higher-value pattern is iterative refinement:
1. Write a rough version yourself (even if bad)
2. Ask AI to identify weaknesses ("What could go wrong here?")
3. Ask AI to suggest alternatives ("Show me 3 different approaches")
4. Pick the best, implement, and ask AI to reviewThis works because AI is excellent at critique and generation but you need a human to drive the steering wheel.
The Prompt Engineering That Actually Matters
You don’t need to learn 20 prompt techniques. Master these three:
1. Role + Context + Constraint
You are a senior .NET architect reviewing production code.
Context: This service handles payment processing for ~10k transactions/day.
Review this method for thread safety and error handling issues.
[paste code]
Constraints: Focus on issues that could cause data loss or double-charging.2. Chain of Thought for Complex Problems
I need to design a caching strategy for user sessions.
Walk me through the trade-offs step by step before giving me a recommendation.
Consider: memory cost, consistency, expiration policy, and session invalidation on logout.3. Concrete Examples Over Abstractions
Instead of: “Explain async/await in C#“
Use: “Show me the difference between sync and async versions of the same database call, with timing implications explained in comments”
What AI Cannot Do (Yet)
- Understand your team’s architectural decisions and constraints
- Know which library version your company has approved
- Debug issues that require reading logs + metrics + code together
- Replace the judgment call of “should we build or buy this?”
Use AI for execution speed, not for decision making. The decisions are still yours.
My 2025 Setup
- Cursor as primary IDE (replaced VS Code completely)
- Claude in browser tab, always open
- Perplexity for anything time-sensitive
- Custom Cursor rules file (
.cursorrules) with our team’s coding standards — this alone saves hours per week
If you want to understand how to build AI-integrated applications (not just use AI tools), our Prompt Engineering & AI Workflow Automation course covers exactly that.
This post reflects my personal workflow as of December 2025. AI tooling evolves fast — the tier list will shift.
