You know that feeling when you spend 45 minutes trying to trick Jira into doing what a simple spreadsheet could do in 10 seconds? That sinking moment when you realize you’re a prisoner of somebody else’s idea of a perfect workflow?
I spent years believing the problem was me. I wasn’t organized enough. I didn’t understand “agile” properly. I needed more training. Then I realized the truth: mass-market software is designed for nobody you’ve ever met.
Every SaaS tool you use is a compromise. A negotiated settlement between what you need and what 10,000 other customers need. The result? A bloated, one-size-fits-nobody product that makes you feel like you’re the problem.
But here’s what’s changed: Large Language Models just made bespoke software economically viable for the first time since the 1980s. We’re entering the age of situated software — hyper-specific tools built for a team of five, a single department, even a two-person startup. Tools that are designed to be thrown away after six months because they’ve already done their job.
This isn’t a theory. A friend of mine, running a small logistics company, generated a custom inventory dashboard in 45 minutes using GPT-4. It’s ugly, it’s brittle, and it wouldn’t scale to a hundred users — but it perfectly matches his mental model. He saved 12 hours of busywork per week. That’s the power of software that’s not trying to conquer the world, just serve your corner of it.
The old SaaS playbook — build generic, raise VC, grow to 10,000 customers — is running on fumes. The next wave isn’t another Salesforce. It’s a thousand tiny tools, each one perfectly tailored to a context that will never be replicated.
You want to know the real twist? The most valuable software in the future will be deliberately unscalable. It will break the moment you try to sell it to a second company. And that’s not a bug — it’s the whole point.
Stop bending your workflow to fit someone else’s template. The LLM era lets you build your own. Start small. Start ugly. Start specific. The mass-market is dead. Long live situated software.
FAQ
Q: Doesn't this mean we'll have a million one-off tools with no maintenance?
A: Yes, and that's exactly the point. Many problems don't need long-term maintenance — they need a six-month solution that disappears when the context changes. Think of it as software as a consumable, not a capital asset.
Q: How does this actually work for non-technical people?
A: LLMs with natural language interfaces let you describe your workflow in plain English and get a working prototype. The barrier isn't coding anymore — it's knowing what you actually need. The hardest part is clarity, not syntax.
Q: Isn't this just another form of low-code/no-code hype?
A: No-code tools still require you to adapt to their templates. LLMs generate code from scratch, so the output can be truly idiosyncratic. The difference is the difference between playing with Lego blocks and having a 3D printer that makes exactly the shape you draw.