Tools
7 AI tools worth trying in May 2026
There are too many AI tools and most are noise. These are the seven we actually open weekly — what each one is genuinely good for, and where it falls short. No hype. Some links may be affiliate; opinions are not for sale.
1. Claude (Anthropic) — daily thinking partner
Best for long-form reasoning, writing, and code review. Opus 4.7 is the model we reach for on hard problems; Sonnet 4.6 is the value pick for routine work. The Projects feature is genuinely useful for keeping context across sessions.
Falls short: Image generation is not its strength. Web access is more limited than competitors.
2. ChatGPT (OpenAI) — broadest capability
Best for the breadth — voice, vision, web browsing, code interpreter, and computer-use agent in one product. GPT-5.5 is the safest single-model bet for a team that wants one tool covering everything.
Falls short: Memory can become noisy across long projects. Expect to clean it occasionally.
3. Cursor — AI-first IDE
Best for daily coding flow. Tab-complete and the agent panel are still the cleanest in-IDE experience. The composer is good for multi-file edits up to a moderate scope.
Falls short: Long autonomous refactors. Pair it with a CLI agent for those.
4. Claude Code — autonomous CLI agent
Best for the hard tickets. Plans, edits across many files, runs the test suite, fixes failures, presents a diff. The right tool when the alternative is asking an engineer to spend half a day on it.
Falls short: Cost and overkill on small tasks. Do not use it where Cursor would have been fine.
5. Perplexity — research search
Best for research questions where you want sources, not vibes. The Pro Search and Spaces features work well for ongoing research threads. Faster than reading ten tabs.
Falls short: Citation quality varies — verify load-bearing claims. It will sometimes synthesize confidently from a thin source set.
6. Granola — meeting notes that hold up
Best for the “what did we decide?” problem. Records, transcribes, and produces structured notes you actually want to send. The auto-extracted action items are accurate enough to forward without rewriting.
Falls short: Privacy posture matters — only use it on calls where everyone has consented to recording.
7. ElevenLabs — voice generation
Best for narration, voice notes, and audio versions of articles. The voice quality is now good enough that listeners do not flinch. Useful for podcasters, course creators, and anyone publishing content.
Falls short: Long emotional range still trips it occasionally. Best for informational content, not drama.
What we cut from this list
We removed three tools that were on last quarter’s shortlist: a video editor that has not kept pace, an automation product that has gotten flakier with scale, and a code assistant that is now strictly behind Claude Code and Cursor on every benchmark we care about. We are not naming them — we used them, they did not work for us, and that is a different post.
How to actually choose
- Start with a daily-use category: thinking partner, IDE, research, meetings.
- Pick one tool per category and use it every working day for two weeks.
- If you are not opening it by week three, it is not for you. Cancel.
- Resist the urge to try ten things at once. The cost of switching is bigger than the marginal quality difference.
We may earn a small commission if you sign up via certain links on this site. It does not affect what we recommend or what we cut. Full disclosure →