My AI Stack for Shipping Fast Without Losing Control of the Code
Everyone talks about using AI to code faster. Nobody talks about which tool for which job.
Here is exactly how I use AI day to day as a Lead Full Stack Engineer.
Claude - Documentation, Boilerplate, Code Reviews & UI
Genuinely the best at writing clean readable code from a description and reviewing PRs for issues I might have missed.
If I need a component built from scratch or a function documented properly, Claude is the first tool I open.
Where I use it:
- Documentation - it captures intent, not just what the code does
- Boilerplate generation - NestJS modules, Next.js routes, Stripe handlers
- Code reviews - drop in a diff, get specific feedback, not generic advice
- UI components - goes from a description to a well-structured Tailwind component
Cursor - Functional & Logical Changes Inside an Existing Codebase
When I need to refactor something that touches multiple files or trace how a bug is flowing through the system, Cursor is unbeatable.
It understands the context of the entire project, not just the snippet I paste.
Where I use it:
- Refactors that ripple across multiple files
- Tracing bugs through the system
- Wiring up state, side effects, data flows
- Any change where the full project context matters
Kiro - Spec-Driven Development
When I need to think through a feature properly before writing a single line of code, Kiro helps me define requirements, design the system, and generate implementation tasks in order.
It brings structure to the build before the build starts.
Where I use it:
- Breaking down features into requirements before touching the codebase
- Designing system behaviour and edge cases upfront
- Generating ordered implementation tasks from a spec
Gemini - Google Ecosystem Research
Anything involving the Google ecosystem. Docs, APIs, Google Cloud, Firebase, Workspace integrations. Native knowledge, faster answers.
Where I use it:
- GCP and Firebase architecture questions
- Current best practices - grounded in live search
- Anything where freshness and Google-native context matters
ChatGPT - R&D, Exploration & Quick Q&A
When I am figuring out an approach before I start building, GPT is good for bouncing ideas quickly without switching contexts.
Where I use it:
- R&D sessions - "How would I architect a system that does X?"
- Quick questions to unblock myself fast
- Exploring approaches before committing to one
The Stack at a Glance
| Tool | Lane |
|---|---|
| Claude | Docs, boilerplate, UI, code reviews |
| Cursor | Functional and logical changes in-codebase |
| Kiro | Spec-driven feature planning |
| Gemini | Google ecosystem, current research |
| ChatGPT | R&D, exploration, quick Q&A |
Why This Works
The mistake most developers make is picking one AI tool and trying to force it to do everything. Each one has a lane. Respect the lane and you move twice as fast.
I ship faster now than at any point in my career. Not because AI writes my code. Because it handles the parts that used to slow me down so I can focus on the parts that actually require judgment.
What does your AI stack look like?