Everyone recommends AI tools. Nobody talks about the ones they stopped using. Over the past three months, I tried 12 different AI tools for coding, writing, and productivity. I kept 4. I deleted 8.
Here is the honest breakdown of what did not work and why.
What I kept
Before the deletions, here is what survived:
- Claude Code for coding agent work
- Cursor for IDE-integrated AI
- ChatGPT for quick questions and brainstorming
- Perplexity for research and web search
These four cover 95% of my AI needs. Everything else was redundant or not worth the cost.
What I deleted and why
1. GitHub Copilot
Why I tried it: Everyone uses it. The autocomplete is supposed to be amazing.
Why I deleted it: After using Cursor’s tab completion, Copilot feels primitive. Cursor predicts multi-line changes and understands context. Copilot predicts the next line. The difference is massive for experienced developers.
Cost: $10/month (Individual plan) Used for: 3 weeks Replaced by: Cursor ($20/month, worth the extra $10)
2. Aider
Why I tried it: Open source, supports any model, terminal-based.
Why I deleted it: Aider is powerful but complex. Configuring models, managing context, and handling errors requires significant setup time. Claude Code does the same thing with zero configuration.
Cost: Free (open source) Used for: 4 weeks Replaced by: Claude Code (simpler, better integration)
3. Continue.dev
Why I tried it: Free, open source, VS Code integration.
Why I deleted it: The free tier is too limited. 30 premium credits per month runs out in a day of serious use. The tab completion is decent but not as good as Cursor. For the same price as Cursor, you get less capability.
Cost: Free (limited) / $15/month (Pro) Used for: 2 weeks Replaced by: Cursor (better tab completion, worth the $5 difference)
4. Windsurf
Why I tried it: Budget-friendly AI IDE, good reviews.
Why I deleted it: Windsurf is fine for beginners. For experienced developers, the agent mode is too limited. Multi-file changes fail frequently. The tab completion is good but not great. Cursor at $20/month is worth the extra $5 over Windsurf at $15/month.
Cost: $15/month (Pro) Used for: 3 weeks Replaced by: Cursor (more reliable agent mode)
5. LM Studio
Why I tried it: Run models locally, no API costs.
Why I deleted it: Local models are slow and quality is poor compared to cloud APIs. Running Llama 4 on my RTX 3060 takes 10-30 seconds per response. The same task on GPT-5.6 Luna costs $0.004 and takes 2 seconds. The math does not work unless you have no internet.
Cost: Free Used for: 2 weeks Replaced by: GPT-5.6 Luna ($0.20/$1.20, faster and better)
6. Ollama
Why I tried it: Simple local model runner, good ecosystem.
Why I deleted it: Same problem as LM Studio. Local models are too slow and too expensive (in time) compared to cloud APIs. The only use case is offline work or privacy-sensitive data.
Cost: Free Used for: 1 week Replaced by: GPT-5.6 Luna (cloud API, faster, cheaper in time)
7. LangChain
Why I tried it: Build AI applications with chains and agents.
Why I deleted it: LangChain is overengineered for 90% of use cases. The abstraction layers add complexity without value. For simple API calls, raw Python with the OpenAI/Anthropic SDK is simpler. For complex agents, Claude Code or custom code is better.
Cost: Free (open source) Used for: 4 weeks Replaced by: Raw Python SDK calls (simpler, more control)
8. Helicone
Why I tried it: AI observability and cost tracking.
Why I deleted it: Helicone is good for teams with multiple developers and complex AI workflows. For a solo developer, the provider dashboards (OpenAI, Anthropic, Google) are sufficient. The added complexity of another tool is not worth it.
Cost: Free tier / $20/month Used for: 2 weeks Replaced by: Provider dashboards (simpler, sufficient for solo use)
What I learned
1. Simplicity wins
The tools I kept are simple. Claude Code: type what you want, get code. Cursor: tab to complete. ChatGPT: ask a question, get an answer. No configuration, no setup, no learning curve.
2. Cost per task matters more than monthly price
A tool that costs $20/month but saves 2 hours/day is worth it. A tool that costs $10/month but adds friction is not. I evaluate tools on time saved, not price paid.
3. Local AI is not worth it for most developers
Unless you have no internet or work with sensitive data, cloud APIs are better in every way. Faster, cheaper (in time), better quality, and zero setup.
4. One tool per workflow
I tried to use multiple tools for the same workflow (coding with Claude Code + Aider + Continue.dev). That added complexity without value. One tool per workflow is optimal.
5. Delete aggressively
If a tool does not deliver value in the first week, delete it. Do not keep trying because “it might get better.” The opportunity cost of learning a bad tool is higher than the cost of switching.
My current stack
| Workflow | Tool | Cost |
|---|---|---|
| Coding (complex) | Claude Code | $100/month (Max) |
| Coding (daily) | Cursor | $20/month |
| Quick questions | ChatGPT | $20/month |
| Research | Perplexity | $20/month |
| API calls | GPT-5.6 Luna | ~$30/month |
| Multimodal | Gemini 3.6 Flash | ~$15/month |
Total: ~$205/month
This covers 100% of my AI needs. No local models, no complex frameworks, no unnecessary tools.
FAQ
Should I try local AI models?
Only if you have no internet or work with sensitive data. For most developers, cloud APIs are faster, cheaper (in time), and better quality. The setup time for local models is not worth it.
Is LangChain worth learning?
For simple use cases, no. Use raw Python SDK calls. For complex multi-step agents with memory and tool use, LangChain adds value. But 90% of developers do not need that complexity.
What is the minimum AI tool stack?
For developers: Cursor ($20/month) + ChatGPT ($20/month) + GPT-5.6 Luna API (~$10/month). Total: $50/month. That covers coding, questions, and API calls.
How do I know if a tool is worth keeping?
If you use it daily and it saves time, keep it. If you use it weekly and it adds friction, delete it. The tools you use every day are the ones that matter.
Should I switch to Claude Code?
If you code in the terminal and want the most powerful AI agent, yes. If you prefer IDE-integrated AI, stick with Cursor. If you are on a budget, use Continue.dev (free) or Windsurf ($15/month).
Related: Best AI Coding Tools 2026 | Cursor vs Claude Code vs Windsurf | Claude Code vs Aider vs Continue.dev | GPT-5.6 Luna Price Drop | AI API Pricing Compared