There are hundreds of AI tools now. Most of them wrap the same underlying models with a different UI and charge you $20/month for it. Here's what we actually use at starOS Labs.
Claude (Anthropic)
Our primary model for everything that requires reasoning, long-context work, or code generation. Claude Code — the CLI — is how we build. It reads your codebase, writes and edits files, runs commands, and maintains memory across sessions. It's the closest thing to a junior developer we've found.
Worth it: yes, especially for code-heavy workflows.
Cursor
IDE built on top of Claude/GPT. If you live in VS Code and don't want to switch to the terminal, Cursor is the best AI-native editor. Tab completion is genuinely good. The chat sidebar gets the context right more often than Copilot.
Worth it: yes, if you're not already using Claude Code.
Perplexity
AI search that actually cites sources. We use it when we need current information — prices, recent news, documentation that might have changed. ChatGPT and Claude have training cutoffs. Perplexity doesn't.
Worth it: yes, as a Google replacement for research queries.
Midjourney / Flux
Image generation. Midjourney v6 is still the quality leader for photorealistic renders. Flux is faster and cheaper for iterations. Both are worth knowing.
Worth it: yes, if you make visual content.
What we skip
- Jasper, Copy.ai, Writesonic: You're paying for a prompt wrapper. Just use Claude directly.
- Most "AI productivity" apps: Chrome extensions that summarize emails, etc. The time savings rarely justify the cost.
- GPT-4 for code: Claude is better. This isn't opinion anymore, it's reproducible.
The real rule
If a tool saves you more time than it costs to learn and pay for, use it. If it doesn't, skip it regardless of how impressive the demo looked.