In my first post I said I’d cover the AI tools I’m actually using and what they cost. So here it is — no affiliate links yet (I’ll be transparent when that changes), no sponsored content, just what’s on my screen.
There are roughly four hundred thousand articles telling you the “best AI tools for side income.” Most of them are written by people who tested each tool for twenty minutes and slapped an affiliate link on it. This isn’t that.
I’m using these tools actively, right now, to build Dictare — a voice-to-text and translation app for Mac and Windows, aimed at non-native English speakers who need to produce clean professional English at their keyboard. It’s a real product. I’m building it alongside a full-time day job, with no co-founder, no investment, and a budget I can account for line by line.
Here’s what’s in the stack.
The Stack
Claude (Anthropic) — £18/month
This is the tool doing the most work. I use it for:
- Architecture decisions — when I was deciding whether to rewrite the app from Python to Flutter, I ran the decision through Claude across multiple sessions. It didn’t tell me what I wanted to hear. It told me Python had a GIL issue on macOS Tahoe that was architecturally unfixable. I rewrote the app. It was the right call.
- Strategy — competitive analysis, pricing decisions, launch sequencing, legal risk framing. Not as a replacement for actual advice (I used a solicitor to clear the IP question with my employer), but as a first-pass thinking partner.
- Content — this post included, though everything gets edited and checked before it goes anywhere.
- Code review and debugging — I’m not a professional developer. Claude Code has been involved in every phase of the Dictare build, from the Flutter scaffold to the Supabase backend. It’s the closest thing I have to a technical co-founder.
I’m on Claude Pro at £18/month. That gets you priority access to the most capable models, which matters when you’re using the tool for extended technical sessions rather than quick questions.
Is it worth £18/month? At this stage of the build, clearly yes. One Dictare subscriber at $4.99 covers the cost. Two subscribers and I’m in profit on the tool that built the product.
Flutter — Free
Dictare is built in Flutter. It runs on Mac and Windows from a single codebase — which for a solo builder with no budget is the only architecture that makes sense.
Flutter is open source, maintained by Google, and free to use. The cost is learning time, not money. I had no Flutter experience before this build. That’s fine — see Claude Pro above.
The decision to go with Flutter came after I scrapped the original app entirely. Dictare started life as a Python/Tkinter desktop app. It worked — up to a point. Then macOS Tahoe introduced a change that triggered Python’s Global Interpreter Lock in a way that caused random crashes during audio processing. Not intermittent crashes. Crashes that were architecturally unfixable without replacing the threading model from the ground up. I had 1,600 lines of working Python code and a fundamental problem I couldn’t solve inside the language.
So I rewrote it. Flutter gave me a single codebase for Mac and Windows — the original Python version was Mac only — plus proper access to the native platform APIs that Dictare needs: global hotkeys, paste-at-cursor, and menu bar integration. These aren’t optional features. They’re what makes the app useful. A framework that fights you on them isn’t a framework, it’s a liability.
The rewrite took nine days from blank project to a working backend with a live paywall. That timeline wouldn’t have been possible without AI tooling. It also wouldn’t have been possible if I’d tried to salvage the Python version rather than replace it cleanly.
Supabase — Free tier (for now)
Supabase handles the backend: user authentication, subscription state, and the API proxy that routes requests to Google STT, DeepL, and OpenAI. Users don’t supply their own API keys — they download the app, subscribe, and it works. That’s the model. Supabase makes it possible without running your own server infrastructure.
The free tier is sufficient while I’m in development and early launch. It gives me a Postgres database, auth, and Edge Functions — the core of the backend architecture. I’ll move to a paid plan once there’s real user load, at around $25/month. Until then, the bill is zero.
Stripe — Per-transaction only
Stripe processes the subscriptions. There’s no monthly fee — Stripe takes a percentage per successful payment (roughly 1.5% + 20p in the UK for European cards). At $7.99/month per subscriber, that’s around 30–40p per payment, which is acceptable.
I set up the full Stripe integration as part of Phase 5a of the build: products, pricing objects, webhooks, and a checkout flow that hands off to Supabase to grant Pro access. It took a day to get working end-to-end. It now works.
Claude Design — Free tier
Claude Design handles mockups and simple graphics — any visual assets needed for the blog or for the Dictare marketing website. I’m on the free tier. It does what I need without a separate Canva subscription.
The paid tier unlocks more options. I’ll revisit it when the free tier becomes a constraint. It hasn’t yet.
WordPress + Hostinger — ~£2.50/month
This blog runs on WordPress, hosted on Hostinger’s starter plan. I paid for 12 months upfront at around £30 total — roughly £2.50/month.
WordPress itself is free. The plugins I’m running are all free: Rank Math for SEO, UpdraftPlus for backups, LiteSpeed Cache for speed.
This is the cheapest part of the stack. A self-hosted blog is an asset you own and can eventually sell. A free Substack isn’t. The decision is obvious.
Ollama + Open Source Models — Tried, Dropped
I initially included this in the stack as a “watch this space” item. I’ve since dropped it.
Ollama lets you run open source models locally — no subscription, no API costs, no data leaving your machine. The economics look good on paper. In practice, the quality gap was too significant for the kind of work I’m doing. These models are improving quickly, but for now I’d rather pay £18 and have it done than spend twice as long fixing output from a free model. Time is the real cost.
What’s Not In The Stack (And Why)
Midjourney or other image generation tools — I don’t need them. Claude Design handles the basic visual work. If I were building something image-heavy this would change.
Jasper, Copy.ai, or dedicated copywriting tools — Claude handles everything. Adding a separate subscription for a narrower tool makes no sense at this stage.
Automation tools (Zapier, Make) — Not yet. When Dictare has enough users to justify automated onboarding sequences, I’ll add them. Right now it would be complexity for its own sake. I plan to email every early trial user personally — at launch numbers, that’s actually possible.
GitHub Copilot — I evaluated it. At this build stage, Claude Code does more useful work and has better context management for multi-file architectural changes. Copilot might make sense later for faster line-by-line completion. It’s not where the bottleneck is right now.
The Honest Total
| Tool | Monthly Cost |
|---|---|
| Claude Pro | £18.00 |
| Flutter | £0 |
| Supabase (current) | £0 |
| Stripe | ~£0 (% per transaction only) |
| Claude Design | £0 |
| WordPress/Hostinger | £2.50 |
| Total | £20.50/month |
That’s my operational cost to build and run both the app and the blog. My domain costs (theaiincomelab.co.uk and dictare.co) sit outside this and total around £33 for the year.
To break even on tools alone, I need roughly 5 Dictare subscribers at the founding member price of $4.99. At the Pro price of $7.99, it’s 3. That is not a high bar. The question is whether I can clear it in the first 30 days of launch.
I think I can. I’ll tell you in the income report.
What I’d Tell Someone Starting Today
Don’t subscribe to everything at once. Start with one AI tool and one hosting plan. Add complexity only when you’ve genuinely outgrown the simple version.
The temptation is to build the perfect stack before you build anything. That’s displacement activity dressed up as preparation. The stack above took maybe two hours to assemble across the whole project. Everything since has been building with it.
Dictare is in the final pre-launch phase. The backend is live, the paywall works, and the app runs on Mac and Windows. The next posts will cover the build journey in more detail — what actually happened across the nine build phases, what broke, and what I’d do differently.
— Badger