I have spent roughly $340 on AI subscriptions since January. That is either a research budget or a small problem, and I am still deciding which. Either way, it means this best AI tools review comes with receipts: actual prices, actual cancellations, and the exact moments each tool let me down.
Context matters here. I write and edit for a living, so most of my testing happens against a client deadline, not in a demo sandbox. A tool that feels clever on a Tuesday afternoon gets deleted fast when there is a 2,000-word brief due Friday morning.
What separates the best AI tools from the rest
Most roundups rank tools by feature count. That gets it backwards. Features are cheap to add and easy to list. What is hard is whether you keep opening the thing once the novelty wears off.
Four things separate keepers from novelties for me:
- Time saved per session. If setting up, prompting and cleaning up takes longer than doing the task myself, it is a hobby, not a tool.
- Predictable output. I need to know roughly what I am getting. Dice-rolling on whether a summary invented a statistic is not a workflow.
- Honest pricing. A free tier that caps you at three runs per day is not free. It is a demo with a countdown.
- Output I can hand over. If the result needs so much rewriting that my byline becomes a lie, it has failed.
The best AI tools for writing, research and thinking
Duck.ai: a second opinion without an account
A lot of my early drafting involves unpublished client material, and pasting that into a consumer chatbot always felt slightly grubby. Duck.ai is DuckDuckGo’s private AI chat, and it has become the one I open when I want feedback on a paragraph without leaving a permanent trail attached to a profile. No account, no history piling up. It handles the basics well: rewriting a clumsy intro, testing whether an argument actually holds, outlining a piece I have been avoiding.
What it will not do is remember anything. There is no workspace, no saved project, no memory of the brief you gave it last Thursday. For a one-off question that is fine. For long-running work you want more scaffolding around you.
Napkin AI: turning a dense paragraph into a diagram
I reach for Napkin whenever a client says “can you make this clearer” and the honest answer is a diagram. Napkin AI turns a messy paragraph into a clean diagram without opening a design tool, which matters when you are a writer who finds layers panels mildly threatening. You paste text, pick a visual style, nudge the layout, export. Six weeks on the free plan before I paid anything, and the paid tier costs less than one print magazine subscription.
The best AI tools for video and visual work
Kapwing AI: video editing for people who do not edit video
I made peace years ago with the fact that I am not a video editor. Then a client asked for six short clips from a 40-minute webinar, due in two days. Kapwing AI handled it: automatic subtitles, one-click resizing for vertical, filler-word removal, and text-based editing where deleting a sentence from the transcript deletes it from the timeline. That last part is the actual magic trick. I edited video by editing words.
The catch is scale. Files past the 30-minute mark get sluggish in the browser, and I re-export constantly because I keep changing my mind about the hook. For a single 15-second clip it is overkill. For repurposing anything long, it earns its keep.
The best AI tools for meetings and admin
Otter.ai: the highest-return subscription I own
Otter is the only AI tool where I could point to a specific invoice and say “that came from a meeting summary.” It joins calls, produces a transcript, and pulls out action items. Raw transcripts, though, are close to useless on their own. The value is entirely in configuration, and there is a solid walkthrough on how to set up Otter.ai so meetings actually turn into follow-up emails. Do that once, properly, and the tool goes from pleasant to something you would pay double for.
Two honest caveats. Heavy accents and people talking over each other still trip it up, and if you record a client call without saying so, you deserve the conversation that follows.
Lindy AI: useful, fiddly, occasionally brilliant
Lindy AI builds small agents that sit in your inbox and calendar. I use it for two jobs: sorting inbound pitches into buckets and handling scheduling back-and-forth so I stop writing “does Thursday at 2 work?” eleven times a week. When it works, it saves me about 90 minutes weekly.
When it breaks, it breaks quietly. An agent that stops firing does not send you an alert, it just goes silent, and you notice three days later when a client asks why nobody replied. Budget ten minutes a week to check your automations are still alive.
Where AI tools still let you down
The failures are consistent enough to predict now. A chatbot told me a competitor had “approximately 4,200 employees” with total confidence; the real number was under 300. Otter renders certain surnames phonetically forever, no matter how many times I correct them. Kapwing compresses exports unless I manually override the settings. Lindy’s agents have looped on the same email twice in six months and I had to kill the workflow and rebuild it.
None of that makes the tools worthless. It makes them employees who need supervision rather than magic wands. I check every number a model gives me against a primary source, and I read every summary before it leaves my outbox.
How to judge a tool in a week instead of a quarter
Annual plans are where AI companies make their money, and where you lose yours. My rule is a seven-day trial with one real task attached. Not a test prompt. A genuine deliverable you already owed someone.
Then run the Friday test: at 4pm on a Friday, with no one watching and no deadline pressure, do you voluntarily open it? If the answer is no three weeks running, cancel the subscription. I have killed four tools this year using exactly that standard, and I do not miss any of them.