I've tried over 40 AI tools this year. Coding assistants, research tools, writing helpers, image generators, automation platforms. Most of them are gone from my machine. Here's what survived and why, as of November 2024.

Tier 1: Daily essentials

Claude (Anthropic) is my primary AI. I use it for architecture decisions, complex debugging, code review, and any task that requires sustained reasoning over a long conversation. The Claude 3.5 Sonnet update in June made it faster without sacrificing quality, which resolved my only real complaint. I talk to Claude more than I talk to most humans on a workday.

What makes it essential: instruction following is the best in class. When I give it specific constraints, it respects them. The 200K context window means I can paste entire codebases. And it's honest about uncertainty, which matters when I'm making decisions based on its recommendations.

Cost: $20/month for Pro, plus API usage for automated tasks. Worth every cent.

Cursor is where I write code. It's a VS Code fork with AI deeply integrated into the editing experience. The Cmd+K inline editing, the project-aware chat, and the tab completions are all best-in-class. I switched from VS Code + Copilot in March and never looked back.

What makes it essential: the inline editing paradigm is addictive. Describe a change in English, see it applied as a diff, accept or reject. It's the fastest way to make small-to-medium code changes I've found. The project indexing means the AI actually understands your codebase, not just the current file.

Cost: $20/month for Pro. Replaced both my Copilot subscription and most of my ChatGPT usage for coding.

Perplexity replaced Google for research. Ask a question, get a synthesized answer with citations. I use it for library evaluation, error diagnosis, architecture research, and staying current on technology trends.

What makes it essential: citations. Every claim is linked to a source. This makes the answers verifiable, which no other AI tool offers for research queries. The Pro tier's GPT-4 and Claude-backed responses are meaningfully better than the free tier.

Cost: $20/month for Pro. Easily saved me $20/month worth of time previously spent clicking through Google results.

Tier 2: Regular use

Ollama for local model inference. I run Llama 3 8B for commit messages, docstrings, and simple code generation on sensitive projects. Free, fast, private. It's not a daily essential because the use cases are narrow, but when I need it, nothing else works.

ChatGPT Plus I still keep around mainly for the code interpreter (Advanced Data Analysis). Upload a CSV, ask questions, get charts and statistical analysis without writing code. For data exploration, nothing else matches this workflow. I also use DALL-E through it occasionally for quick mockups and diagrams.

v0 by Vercel for rapid UI prototyping. Describe a component, get a working React implementation with Tailwind styling. It's narrow in scope but excellent at what it does. When I need to quickly prototype a landing page or dashboard layout, v0 gets me there faster than Cursor because it's optimized for this specific task.

Tier 3: Occasional but valuable

Notion AI for meeting notes and document summaries. Not essential for coding, but I use it weekly for organizing technical documentation and summarizing long threads. The AI is deeply integrated into Notion's interface, which matters more than raw model quality for this use case.

GitHub Copilot I no longer use in my editor (Cursor replaced it), but I still use Copilot for PR descriptions and commit messages through the GitHub CLI. It's trained on enough commit history to generate decent, conventional commit messages automatically.

What I cut and why

Bard/Gemini: decent model, but I already have Claude and GPT-4. Three general-purpose AIs is two too many. Grok: fun personality, insufficient quality for professional use. Cody (Sourcegraph): good codebase understanding, but Cursor's project indexing covers the same need. Tabnine: tab completions aren't as good as Cursor's. Jasper/Copy.ai: I don't need a dedicated writing AI when Claude writes well enough. Replit AI: interesting for browser-based development, but I develop locally.

Total monthly cost

Claude Pro: $20. Cursor Pro: $20. Perplexity Pro: $20. ChatGPT Plus: $20. Total: $80/month. I genuinely believe this saves me 15-20 hours per month. At any professional rate, that's an absurd return on investment. The key is that each tool has a clear, non-overlapping role. No redundancy, no waste.

My advice

Pick one tool per category and commit to it. One reasoning AI (Claude), one coding editor (Cursor), one research tool (Perplexity). Master those three and you'll outperform someone using ten tools badly. The productivity gain comes from depth of use, not breadth of tools.