Every few weeks I read my card statement line by line, because the fastest way to waste money on software is to stop noticing it. Subscriptions are designed to become invisible. Mine run two content brands, and here's the whole list — what stayed, what got cancelled, and the reason for each, including the embarrassing one.
No affiliate padding. Two of these tools pay me if you buy them and I'll flag those inline; the rest I'm listing because they're what's actually on the bill.
The one that's basically the whole bill
Claude Max — $100/month. This isn't really a tool, it's the worker. It drafts, it writes the code that renders my videos, it runs the automations. If I shut every brand down I'd keep it, because it's what I use to do everything else too. I've written before that this single line dwarfs every other cost combined, and that the interesting fact isn't the number — it's that the same subscription runs one brand or four for the same price.
Is $100 a lot? For a hobby, yes. For the thing that does the majority of the actual labour across four brands, it's the cheapest full-time worker I've ever hired. I judge it against what I'd pay a freelancer for the same output, not against a Netflix subscription.
The tools that earn their keep
Blotato (Starter plan) — affiliate link. The scheduler that publishes to ten accounts across four brands while I sleep. It's the one paid tool that exists because of the brands, and the reason it stuck is an API a script can call, not the calendar. I reviewed it in full here, defaults-that-bite and all. I won't quote a price because tiers move; it earns its line by removing the single most repetitive job I have.
ElevenLabs — affiliate link. The best AI voice I've used, and I only pay for it for one channel — the one where the voice is the product. Everything else runs on Amazon Polly's free tier, and I explained exactly why the free tool wins the volume work. If I did more hero-voice work I'd pay for a bigger tier without blinking. For what I do, the smallest one is right.
Canva Pro. Started as a free trial, converted to paid when the trial ended, and I let it — which is exactly the kind of quiet conversion I'm usually suspicious of. It stays because I actually open it, for thumbnails and quick graphics that aren't worth doing in code. The day I stop opening it, it goes. That's the whole test.
Domains — about $11 a year, each. Two brand names on Namecheap, roughly a dollar a month apiece. The cheapest permanent thing you can own on the internet, and the one cost I never think twice about.
The free tier is load-bearing
Two things do enormous work for zero dollars, and pretending they're free forever would be dishonest, so:
Amazon Polly narrates the daily videos on a free tier — one million characters a month for the first twelve months. My longest news short was 630 characters. I will not come close to the ceiling this year. Next year the meter starts, and I'll re-evaluate then rather than pre-worry now.
Vercel (Hobby) hosts both sites for free and has never once been the reason something broke. The day I outgrow it is a good-problems day.
Half my "stack" is free tiers doing paid-tool work. The skill isn't finding free tools — it's knowing which free tier is generous enough to be load-bearing, and which is a trap that bills you at scale.
The six I dropped — and why
This is the more useful list, because a stack is defined as much by what you said no to. Some of these I paid for and cancelled; some I trialled and declined. I'll be clear which.
1. MiniMax Max — $44/month. Cancelled, and still fighting it off my card. I asked to cancel and the charge kept coming. As I write this it's a line I'm disputing, not one I chose. I'm naming the number because the honest version of "what do you pay for" includes the thing you're trying to stop paying for. Check your statement against your intentions; the gap is usually a tool like this one.
2. Blazr (managed WordPress). Cancelled to zero. An earlier version of this blog ran on it. When I rebuilt on plain HTML and Vercel, I didn't just stop posting — I revoked the access credential, deleted the integration, and confirmed it was dead. A cancelled subscription with a live API key is not cancelled. Kill the connection, not just the billing.
3. Azure's Arabic voice. Trialled, rejected by ear. Before Polly I tested Azure's Arabic neural voice. On paper it was fine. Listening to it next to Polly's, it wasn't, and voice was the weakest link in my videos at the time. I switched. No spreadsheet decided this; my ear did, in about ten seconds.
4. Cursor. Evaluated, declined — never paid. I looked hard at Cursor and stayed on the tool I already live in. I wrote the honest comparison without pretending I'd run Cursor for a year, because I hadn't. "Dropped" here means "decided against after a fair look," which is a real decision even when no money changed hands.
5. Higgsfield. Stayed on the free trial, never upgraded. I generate images and video on its free tier and it's genuinely good. Every time I've come close to paying, the free allowance was enough for what I actually needed that week. The moment it isn't, I'll pay — but I won't pay ahead of the need. Trials are for finding out whether the paid tier is real; this one keeps passing as free.
6. edge-tts and the other free voices. Free, and still not worth it. Worth saying that "free" didn't automatically win. I tried the free command-line voices and they sounded like free command-line voices — fine for a robot, wrong for a brand. Polly's free tier beat them not on price but on quality, which is the only comparison that matters once something clears the "free enough" bar.
What the whole thing actually costs
The honest monthly total is dominated by one $100 line that isn't really a brand cost, plus a scheduler, a voice tier I keep small, a design tool I actually open, and two dollars of domains — sitting on top of free tiers doing the heavy lifting. If you stripped out the disputed $44 that shouldn't be there, it's a lean stack for what it produces.
And that's the point of reading the statement line by line. The disputed charge, the trial that converted, the tool you kept "just in case" — those are where the waste hides, not in the one big subscription you think about every month precisely because it's big.
If you're building your own stack
Three rules I'd hand my past self:
- Price the worker, not the tools. Whatever does the actual labour — an AI subscription, or your own hours — is the real cost. Everything else is rounding.
- Free tiers are a strategy, not a fallback. Half a good stack can be free if you know which free tier is load-bearing and which is bait. Test where the meter starts before you depend on it.
- Cancel loudly. Read the statement, kill the credential not just the billing, and treat a trial as a question — "is the paid tier worth it?" — that you're allowed to answer no.
The tools I pay for earned it by being opened, called, or listened to every week. The six I dropped failed one of those tests. That's the entire filter, and it's cheaper than any of the subscriptions.