Most "best AI tools" lists rank products by brand. That is how you end up opening the same chat window for a job it is bad at. A better habit is to start from the job, pick the tool built for it, and know what that tool is weak at before you commit an afternoon.
Here are six jobs and the first pick for each, taken from the published lists at Ranked By Task. For every pick I have included the catch, because every tool has one.
Vibe coding: Cursor
Pick: Cursor, from the vibe coding ranking. It is an editor with an agent that edits across files. It can take a multi-file change, apply it, and let you reject individual hunks without leaving the project.
The catch: vague prompts send it wandering, and it will spend tokens proving it. Be specific about the change you want, then review the diff.
If you would rather work from a terminal, the same list puts Claude Code second. It is best when the job is "go do this in the repo", and it plans well across a large codebase. It is the wrong pick if you want inline diffs as your main interface. OpenAI Codex is third, useful when the task should run in a cloud sandbox away from your laptop.
Video: Runway
Pick: Runway, from the video ranking. The reason is control. Camera moves, inpainting and a workflow that resembles editing make it the one people finish a shot in.
The catch: clips are still short, and keeping a scene consistent is a fight. Runners-up on the list are Kling, which has more convincing motion and holds up better on longer generations, and Luma, which is fast and easy to direct with simple camera language like a dolly or an orbit.
Agent building: n8n
Pick: n8n, from the agent builder ranking. It is a workflow tool that grew agents, which is what most business tasks actually need. Triggers, branches and a human approval step are first-class, and you can self-host.
The catch: if your engineers want explicit state, retries and a graph they can test, the second pick, LangGraph, is a better fit. You are writing software there, not dragging boxes. Skip it if nobody wants to own that code. For company assistants over internal docs and tools, the list points to Dust.
Writing: Claude
Pick: Claude, from the writing ranking. It is still the one to open for a long draft that has to sound like a person.
The catch: it is weaker when the job is searching the web and citing this afternoon's news. That is research, not prose. ChatGPT is second on the list as the generalist for drafts and rewrites, though a careful draft from it still wants an edit. Gemini is third, convenient when the draft lives next to Gmail and Docs, with a flatter voice for ordinary prose.
Images: Midjourney
Pick: Midjourney, from the images ranking. It has the strongest default taste. Composition out of the box beats most rivals, especially for concept frames and photography-like images.
The catch: prompting is its own dialect, text inside the image is a gamble, and it is not a production asset pipeline. If you need to follow a fussy prompt closely or run or fine-tune the model yourself, look at Flux in second place, which has open weights and the best prompt adherence among widely used image models. If the image has to follow an instruction or include readable text, ChatGPT is third.
Research: Perplexity
Pick: Perplexity, from the research ranking. It is the fastest way to get an answer with sources you can click, good for "what is the state of X" and for current events.
The catch: it will still cite a weak page with a straight face, so read the links. When you already have the sources, NotebookLM is second: upload the papers, transcripts and notes and it stays inside them. For academic literature reviews, Elicit is built to find papers and pull columns of methods and findings. It is narrow on purpose.
The pattern
Look at what repeats. Each pick wins on one specific strength: Cursor on multi-file edits, Runway on control, n8n on workflows a team can run, Claude on tone, Midjourney on taste, Perplexity on sources. Each loses on something just as specific. Choosing by the job, and reading the "wrong pick if" part, saves more time than chasing whichever model launched this week.
Get the one-page version
I put all six picks and the runners-up for each on a single page, as a free cheat sheet you can keep open next to your editor. The PDF arrives after you sign up for the weekly roundup.
Get the free cheat sheet: the top AI tool for every job
Editorial ranks are opinions, and some tool links on the site may earn a commission at no cost to you. Rankings are never paid for or changed by that. If you want a pick for a job that is not on this list, the site has a three-question picker.
The Ranked By Task team
Top comments (0)