Which tool for which job. For each one: what it does, what it does not do, and the link to its vendor. Type what you are trying to do, the list narrows.
Nothing for that search. Tell us which tool is missing.
The most widespread general assistant: write, summarise, translate, code a little, analyse a file.
What it does not do It invents confidently when it does not know. Every number needs checking at the source.
chatgpt.com →General assistant, strong on long documents and sustained writing.
What it does not do Same rule: unsourced is unverified.
claude.ai →Assistant from a European vendor, EU hosting on business plans.
What it does not do Smaller extension ecosystem than the two above.
chat.mistral.ai →Code editor where you describe the change in plain words and the agent applies it across the project.
What it does not do It depends on third-party models: when a provider closes the tap, the tool stops.
cursor.com →Completes code as you type and answers inside the editor, wired into GitHub.
What it does not do It proposes plausible code, not verified code: tests remain your job.
github.com/features/copilot →Describe an application in plain words and get a working site, database included.
What it does not do Easy for the first version, costly to take over once the project grows.
lovable.dev →Builds web interfaces from a description or a screenshot.
What it does not do The interface only. The rest of the application is on you.
v0.app →Guided assumptions in, formatted financial forecast out, in the shapes a bank or investor expects.
Price à partir de 7 $ par mois, remboursement sous quinze jours
What it does not do It formats, it does not check your assumptions. A wrong forecast, well presented, is still wrong.
upmetrics.co →Describe a home in one sentence, get a floor plan, refine it in conversation, see it in 3D.
Price gratuit pour démarrer
What it does not do The vendor writes it itself: a professional must review before any construction or permit application.
www.maket.ai →Interior floor plan in 2D then a 3D render, aimed at layout and furnishing.
What it does not do Interior layout, not a regulatory document and not structure.
planner5d.com →Turns a text or an outline into a laid-out deck you can export.
What it does not do The layout is done, the argument is still yours. A handsome document does not save a vague idea.
gamma.app →General layout, decks, posters, social, with built-in AI features.
What it does not do The templates are recognisable. A distinct identity means leaving them behind.
www.canva.com →Image generation from a description, very strong on style and texture.
What it does not do Text inside images stays unreliable, and precise composition control takes craft.
www.midjourney.com →The industry standard design tool, with AI features to rough out a screen.
What it does not do It mocks up, it does not produce the site. Turning it into code is still work.
www.figma.com →Video generation and editing, effects, background replacement.
What it does not do Long coherent shots remain hard, and each attempt burns credits.
runwayml.com →Edit video or a podcast by correcting the transcript like a text document.
What it does not do Excellent on speech, weak on complex visual editing.
www.descript.com →Synthetic voice and dubbing, across many languages.
What it does not do Cloning a voice without the person's consent is forbidden, and detectable.
elevenlabs.io →Composes a full track, lyrics and vocals included, from a description.
What it does not do Commercial usage rights depend on the plan: read them before publishing.
suno.com →Automatic transcription, open weights, runs on your machine with nothing leaving it.
What it does not do No interface: this is a building block for people who can run a program.
github.com/openai/whisper →Joins the meeting, transcribes, and returns a summary with the decisions.
What it does not do Your meetings go to its servers: check that before discussing clients on it.
otter.ai →Answers with its sources and links, which makes checking fast.
What it does not do Quality depends entirely on the pages it finds. It cites a bad blog as readily as a good paper.
www.perplexity.ai →You give it your own documents, it answers only from them, pointing at the exact passage.
What it does not do Nothing outside your documents. That constraint is what makes it reliable.
notebooklm.google.com →Chains tools and models into visual automations, self-hostable.
What it does not do A long automation breaks at the first failing step: plan retries and a log.
n8n.io →Agent observability and tracing: see what each step actually did.
What it does not do An observation tool, not a fix. It shows the failure, it does not repair it.
www.langchain.com/langsmith →One address to call dozens of models, with switching between providers.
What it does not do One more dependency in the chain, and prices follow the providers'.
openrouter.ai →The central repository of open models, with weights, licences and counters.
What it does not do Everything is published unfiltered: check the licence and the author every time.
huggingface.co →The directory does not rank and does not score. A missing tool is not a bad tool: it has not been checked yet. Prices move, they are given as the vendor displayed them on the day of the check.
Geneva, Switzerland. Write to hello@newsworldai.xyz.