These are the four I actually pay for. Not the four I signed up for, not the four with the best launch videos, the four that keep earning their spot on my card. A twenty-dollar subscription you open twice a month is still a waste, and I've paid for enough of those to build a rule out of it.
If you came from the video, that's the shortlist. Here's the part that doesn't fit in sixty seconds: why each one stays, the one I'd make prove itself, and the scorecard I run before any new tool gets to bill me. Run it on your own stack and you'll cancel at least one thing this week.
The four I pay for
Each of these earns its keep because it finishes a kind of work I do most weeks, not because it's clever. That's the whole filter.
Claude
My default for anything with a lot of files, research, or code behind it. I'm on the 20x Max plan because I hit the limits on smaller tiers, which is exactly the signal you want before you pay up: you outgrew the cheaper option doing real work, not out of curiosity. If you're setting it up from scratch, my one-day Claude setup gets you there.
ChatGPT Plus
I keep this alongside Claude, not instead of it. Plus includes Codex for shipping code, and a second strong model means I can check one against the other when the stakes are high. Two paid chat tools sounds like waste until the day they disagree and one is right.
Higgsfield
If you make content or run ads, making video without filming changes the math on what you can post. This is the clearest case of a tool paying for itself: one skipped shoot day covers a year of the subscription. If it doesn't replace a real cost like that for you, it's a toy, and you should treat it as one.
Grok
I use it for work that keeps coming back: set the process up once, then review what it brings back on a schedule. You can see a small business built on exactly that in my build a business with Grok walkthrough. It earns its place when the recurring job is real, not when you're just poking at it.
The one I'd make prove itself
Perplexity is the one I'd make audition before I added another line item. Nothing wrong with the product. The honest question is whether a dedicated research tool adds enough on top of the research already built into Claude and ChatGPT, both of which I pay for. When two tools I own already do 80 percent of a third's job, that third has to be clearly better at the part I actually use, not just nice to have. That's a buying decision, not a review score, and it's the trap I fall into most: paying for the specialist when the generalist I own already covers it.
The scorecard I run before I pay
When I'm on the fence about a tool, I don't argue with myself, I score it. Five questions, one point each. Four or five, I pay. Three, I stay on the free tier and set a reminder. Two or below, I close the tab. Writing it down stops me from talking myself into things because the demo was good.
Tool: [ ] Price: [ ]/mo Deciding on: [today's date]
[ ] It finishes work I do most weeks, not once in a while
[ ] The free tier does NOT already cover how I'd use it
[ ] A tool I already pay for does NOT do 80% of this job
[ ] It replaces a real cost (hours, a hire, another tool)
[ ] I'd feel the loss within a week if it disappeared
Score (count the checks): __ / 5
4-5: pay for it. 3: free tier + revisit in 30 days. 0-2: skip.
Cancel reminder set for: [pay date + 30 days]The last line matters more than the score. Most wasted spend isn't the tool you shouldn't have bought, it's the one you should have canceled and forgot about.
A real decision: the wedding photographer
Say you're a solo wedding photographer, maybe 25 weddings a year, staring at three paid plans and wondering which to put on the card. Run each through the scorecard instead of guessing.
A photo-editing AI at 30 dollars a month scores a 5. You edit thousands of images after every wedding, the free tier caps you at a handful of exports, no tool you own touches culling and color at that volume, and it replaces the late nights you lose to it now. You'd feel it gone by Monday. Easy yes.
A general chat tool like Claude or ChatGPT at 20 dollars scores a 4. You'd use it weekly for client emails, timeline questions, and contract wording, one paid tool covers all of that, and it replaces the hour you spend rewording the same "here's what to expect on the day" email. It's a 4 not a 5 only because the free tier half-covers the light weeks. Still a yes, just pick one and don't pay for both yet.
A dedicated AI copywriting tool at 25 dollars scores a 2. You'd touch it twice a month for a caption, and the chat tool you just decided to pay for already writes captions fine. That's the specialist trap in real life: essential in the demo, a feature you already own by Tuesday. Skip it, and revisit only if captions become a real grind.
Ask one thing: does this finish work I actually do, often? If the free tier covers it, or you'd open it twice a month, skip it. Set a reminder 30 days out and cancel anything that hasn't earned its place. The subscription is the easy part. Remembering to leave is the discipline.
What goes wrong
The demo sells you the ceiling, not your reality. Every tool looks incredible when someone who built the workflow drives it. Run your own real task in the free trial before you pay, not the polished example from the landing page. If it can't do your Tuesday, the demo doesn't count.
Free would have been fine. A lot of paid tiers exist to sell headroom you'll never reach. A photographer editing 25 weddings needs the paid plan; someone editing four does not. Match the tier to your actual volume, not the volume you imagine having, and step up only when you hit a wall doing real work.
You forget to cancel. The renewal is silent and the guilt of "I might use it" keeps it alive for months. The only cure that's worked for me is the reminder on the scorecard: 30 days after I pay, I get pinged to justify the next charge or kill it. No reminder, no exceptions, and the graveyard of forgotten subscriptions fills right back up.
If you want the full framework behind these calls, how to choose which AI tool to pay for goes deeper on trading tools off against each other. And once you've picked one, don't let it sit half-used: my one-day setup gets a tool earning its money in the first week instead of the third month.