Logo
Me Writing Reading

Re-evaluating In-House Software

So I’ve been thinking about in-house software lately. You know how it goes - your team needs something specific, you look around at what’s available, and then comes that moment where someone says “maybe we should just build it ourselves?”

Traditionally, this is where everyone gets nervous. And for good reason.

The Old Rules

In-house software used to be the exception, not the rule. The conditions had to be exactly right:

  • The problem space had to be kind of adjacent to your company’s core business (but not too far)
  • The scope had to be somehow limited - you couldn’t be building Salesforce
  • The software you were looking for either didn’t exist, or wasn’t readily available in a nicely packaged offering
  • Or maybe it existed but cost way too much for what you needed

A lot of conditions had to align. Making in-house software that wasn’t part of your core business? That was often seen as a distraction. “Focus on what you do best,” everyone would say. “Buy everything else.”

And honestly? They were usually right. Building software is expensive. Maintaining it is even more expensive. The hidden costs would eat you alive - documentation, training, bug fixes, security updates, scaling issues. I’ve seen teams sink months into building something that could’ve been solved with a $50/month SaaS subscription.

But Here’s the Thing

AI switches things over. Completely.

I can now write software as fast as I can describe what I want. Like, actually write it, not just prototype it. If you’ve got a clear goal and the need is relatively small, why wouldn’t you just build it?

Think about it. That internal tool your team needs? The one where you’ve been comparing three different SaaS products and none of them quite fit? You can probably build exactly what you need in a weekend now. No compromises, no feature bloat, no paying for seats you don’t use.

I’ve been doing this more and more. Need a specific dashboard? Built it. Custom integration between two services? Done. That workflow automation that would require Enterprise pricing on every platform? Knocked it out in an afternoon.

The New Default?

I’m starting to think in-house should be the new default. At least for certain categories of software.

Agents are great at writing code. They understand patterns, they can work with APIs, they can even help you architect things properly. You describe what you need, iterate a few times, and boom - you’ve got working software that does exactly what your team needs. As long as there’s an API, all is possible. I think that with the rise of browser automation technologies, anything on the internet does have a public API, regardless of what they think (small side note but, think for a minute that even tech giants with unlimited money like Meta and X can’t get rid of bots).

The economics have completely changed. What used to take a team of developers months can now be done by one person with AI assistance in days or even hours. The build vs buy calculation looks totally different when “build” suddenly costs 10% of what it used to.

The Reality Check

But (and there’s always a but), there’s still one major stopper: deployments and DevOps.

That part hasn’t been streamlined by AI yet. Not really. You can generate all the code you want, but you still need to:

  • Set up hosting infrastructure
  • Configure CI/CD pipelines
  • Handle monitoring and logging
  • Deal with security updates
  • Manage backups
  • Scale when needed

You’ll still pay a price to operate and maintain that software. Maybe not in development time anymore, but definitely in operational overhead. Your AI assistant can write you a perfect Node.js app, but it’s not going to SSH into your server at 2 AM when something breaks.

So What’s the Play?

I think the sweet spot right now is internal tools and utilities. Things where:

  • The requirements are clear and relatively stable
  • You don’t need five-nines uptime
  • Security isn’t a show-stopper, no one is gonna sell your Jira tickets on the dark-web
  • You can tolerate some rough edges

Start there. Build the things that would make your team’s life easier but weren’t worth the old cost of development. That data transformer, that custom admin panel, that integration glue between your systems.

Then see how it goes. Maybe your organization realizes investments on developer experience and infrastructure are worth more than they thought. Tell them you can code it on 1 week, but deploy it on 1 month.

Until then? Next meeting you have about new software, propose building it in-house. Even if it doesn’t get approved it’ll get the cranks moving for your teammates. One step at a time.

What do you think? Have you been building more in-house lately? What’s working and what’s not?

Related thread below that sparked this post