Why I Tried So Many
I have a problem. When there’s a new AI tool, I need to know if it’s useful for my work. So I test it.
This is my honest assessment after using these tools extensively.
The Contenders
Claude (Anthropic)
Best for: Complex reasoning, code generation, thoughtful analysis
What surprised me: Claude actually thinks about why you’re asking something, not just what you asked. When I asked about Kubernetes networking, it didn’t just explain the concept — it anticipated follow-up questions.
# Claude generated this Terraform module for me
# It actually understood the security requirements I mentioned
# and added appropriate IAM restrictions
Weakness: Sometimes too cautious. I had to push to get it to take bold architectural positions.
GPT-4 (OpenAI)
Best for: Quick code snippets, boilerplate, familiar patterns
The undisputed king of “write me a function that does X.” Speed is unmatched.
Weakness: For complex DevOps work, it often suggests the “textbook” answer rather than the practical one. Sometimes misses recent tool updates.
Gemini (Google)
Best for: Google Cloud integrations, context from my own documents
The multimodal capabilities are genuinely useful. I can upload architecture diagrams and ask questions about them.
Weakness: API costs add up fast. The pricing model is confusing.
Specialized Tools
GitHub Copilot
Good for autocomplete-style code suggestions. Lives in my editor. But for DevOps, it’s mostly helpful for boilerplate, not complex configurations.
LangChain
Overhyped for simple use cases. The abstraction adds complexity without benefit unless you’re building something genuinely complex.
Vector Databases (Pinecone, Weaviate)
Actually useful for knowledge management. I now have a searchable database of our internal runbooks and documentation.
My Current Stack
Daily: Claude (complex work) + Copilot (autocomplete)
Research: GPT-4 (speed) + Gemini (multimodal)
Internal: Vector DB + Custom scripts
What I’d Tell My Past Self
Don’t chase the latest model. The productivity gains come from:
- Finding your workflow with one tool, not trying all tools
- Building patterns you can reuse, not one-off prompts
- Focusing on integration — tools that work together beat single powerful tools
The best AI tool is the one you actually use consistently.