Build vs Buy: When Does It Makes Sense to Replace Your SaaS Subscriptions?
When does vibe coding your own software to replace your expensive SaaS subscriptions actually make sense? I've replaced a few apps I subscribed to and have learnt (and saved) a ton. You can too!
Hey there!
When does it make sense to cancel a SaaS subscription and build your own version of it using AI?
I’m always reading posts on X saying it’s a terrible idea. You’re going to waste weeks building it! What about the maintenance?
But then I decided to replace a few of the expensive SaaS tools I was using and I’m converted!
Watch how I built my own Intercom style support system
There are some real reasons you should consider ‘rolling your own’ software to replace pricy subscriptions.
It’s a great way to learn how to build things. You’re already a user of the tool you’re looking to replace (probably a power user!) so you know exactly how you want your new app to work. You know what features it should have. You don’t have to spend weeks searching for the perfect app idea, you’re just copying (and hopefully improving) a tool that already exists. I speak to so many people who get stuck at the step
Learn a new app development stack. I started my first replacement build as a weekend project simply because I wanted to learn how to build an app using Convex as the DB and hosting it on Cloudflare Workers using NextJS. If there’s a new technology you’ve been wanting to try - rebuilding a tool you already use can be a great way to give things a test drive.
Better fit your use case. One huge bonus of building your own tools is that you can customize them for your exact use case and make them totally personal to you. For my support system I was able to give it a deeper integration into my app’s backend to perform admin tasks and fetch data when needed, all things I used to have to do manually. Building the tool myself meant I could integrate it deeply.
Reduce your costs. It goes without saying you can radically reduce your operating costs by replacing expensive SaaS subscriptions with your own vibe coded version. In fact, my support system built on Convex and Cloudflare is effectively free to operate!
There have been some fairly high profile cases of companies replacing expensive SaaS, such as Klarna who replaced expensive Salesforce & Workday subscriptions with custom built software.
Klarna went all-in on this approach back in 2024. They shut down Salesforce and Workday (plus a bunch of other tools) and rebuilt their internal systems with a simpler AI-powered stack.
This was in 2024… before Claude Code and Codex was even released, so the barriers have been lowered even further since then… is there any such thing as a sticky SaaS anymore?
But it’s not just billion-dollar companies doing this. I did the same thing on a much smaller (and more realistic) scale with two tools I was actually paying for every month.
My Crisp/Intercom replacement for customer support.
Old cost: ~£100/month
New stack: NextJS + Convex + Cloudflare Workers (effectively free at my scale)
Live chat widget on the site, full help desk, product updates feed and my favourite feature: new tickets ping me in Slack and I reply straight from there, it syncs back instantly. It handles incoming and outgoing emails using Postmark for email support.
What started as “just learn Convex” turned into a system that fits my exact workflow better than anything off-the-shelf.
Both projects taught me new a stack, gave me deeper integrations I could never buy, and paid for themselves in months.
Most importantly, because I was already a power user, prompting the AI was easy, I knew every feature, every frustration, and every edge case I wanted to fix.
So what should you actually build this on?
If you’re going to try one of these projects, the stack you pick is important. Get it wrong and you’ll spend your weekend fighting deployment issues instead of building. Here’s what I’d use if I was starting from scratch today.
For the app itself, NextJS. Not because it’s trendy, but because every AI coding tool has seen a million NextJS projects and gets it right first time.
Claude Code works incredibly well with it. As you can see in my video tutorial you get a really nice dev environment with claude code below a live preview of your app.
For the database and backend, Convex. This is the one I’ve really fallen in love with. It’s a database, backend functions, file storage, real-time sync, all in one thing, and it’s fully typed, which means the AI writes correct code instead of guessing at your schema.
Real-time sync basically for free is what made the Slack integration on my support tool a 10-minute job instead of a weekend. The free tier is generous enough that most side projects will never pay a penny.
If Convex feels too opinionated for you, Supabase is the sensible alternative which I’ve covered in previous editions. Proper Postgres, good auth, works with anything. Bit more wiring to do yourself, but it has a brilliant MCP server.
For hosting, Cloudflare Workers or Vercel.
Both have free tiers that’ll comfortably cover a personal project, both deploy from a git push, both will stay up without you thinking about them. I’ve been using Cloudflare recently because I wanted to learn how to host full apps on it, but honestly you can’t go wrong either way.
For auth I used the built in Convex Auth with email one time passwords.
And then Claude Code to actually build the thing!
The trick I’ve landed on is to tell it the exact stack upfront… “NextJS, Convex, deploying to Cloudflare Workers” otherwise it’ll waste time picking technologies for you.
Then just work through the features one at a time, the same way you’d use the tool you’re replacing.
That’s it. That’s the whole stack. Nothing exotic, nothing you’ll have to rip out in six months.
So when does it actually make sense?
Here’s the quick decision framework I now run every time a renewal email lands in my inbox:
Cost threshold: Under ~$50/month? The bar is high… it better teach me something big or save me serious time. Over $80–100? Way easier to justify the experiment (the savings can easily pay for a pro Codex or Claude Code sub!)
How well do I know the tool? If I’m a daily power user, I can describe the perfect version in detail. That context makes AI builds shockingly fast.
Learning upside? Every build lets me try a new stack (Convex this time, Cloudflare Workers last time) that I can now use on client work or future products.
Maintenance check: Will this need weekly babysitting? If it’s built on rock-solid platforms like Cloudflare, Vercel, or Convex… usually no. If issues do arise, Claude Code is usually able to pull in the logs directly via the Cloudflare MCP and address the issue without much input from me.
Being able to build so quickly using AI has flipped the economics completely.
A few years ago this would have been weeks of pain and totally unviable. Now it’s a weekend project that levels up your skills and cuts costs.
I still read those X threads saying “don’t do it — maintenance will kill you.” They’re not wrong for every case.
But when the tool is something you use every day, the savings are real, and you’re treating it as a learning investment… why not give it a go?
What SaaS tool are you thinking about replacing?
Hit reply and tell me! I’d love to hear what stack you’re thinking of using.
Happy building! James
PS: If you’re interested in this kind of thing I’d love for you to join us at the Create With Conference in Brighton this June! Ticket are only £49 for a day of learning about how to build with AI.