Opinion
Stop Buying Software.
Start Describing Problems.
Why Software Is No Longer a Product, It's a Thought
Photo by Daniil Komov on Unsplash
Inspired by the Citrini Research scenario “The 2028 Global Intelligence Crisis”
A couple of weeks ago, a piece of speculative macro writing broke the internet a little, in the way things do when they name something people already sense but can't articulate. Citrini Research published a fictional memo from June 2028, a post-mortem on what they called the Global Intelligence Crisis. The premise: what if AI turns out to be completely right on capability and catastrophically wrong on distribution? What if it works exactly as promised, and that's the problem?
Read it. The whole thing. Come back.
One thread in their argument pulled at something I've been trying to press on clients for months. They describe the collapse of mid-market SaaS almost as an aside: a competent developer with agentic coding tools could replicate the core functionality of a $500,000 annual contract in weeks. Not perfectly. But well enough that renewal went from inevitable to optional.
Citrini frames this as disruption; incumbents losing, margins compressing, creative destruction doing its thing, just faster. Classic. And mostly right.
But it misses the more unsettling shift underneath. This isn't about who wins the software market. It's about what software is. And that changes everything.
The Sentence That Can't Be Unread
Software economics have one fundamental rule: development costs are front-loaded, so you amortize them across as many users as possible. More users, better unit economics. Simple, powerful, and responsible for building some of the most valuable companies in history.
But it has a consequence that nobody puts on the pitch deck:
The product must fit everyone, which means it fits no one perfectly.
Stop. Sit with that. You've lived it every single day. I won't elaborate. I'll just ask you to recall your most recent encounter with Microsoft Teams, Salesforce etc ...
Every SaaS product is a negotiation between what you actually need and what you're willing to tolerate. Features averaged. Workflows generalised. Interfaces built for a median user who doesn't exist. Roadmaps driven by the loudest, largest customers while everyone else adapts.
Scale required standardisation. Standardisation required compromise. Compromise, as any NPS score will confirm, delivers a 6 or a 7. (a phrase the kids love these days for anything 'meh').
We accepted this deal for thirty years. Not because we liked it. Because there was no alternative.
The Deal Is Off
Agentic AI coding tools, and I am watching this happen in real time with clients, have demolished the economic logic that made standardisation necessary.
When an AI system can generate working, deployable software from a clear description of requirements, the entire cost structure inverts. The front-loaded fixed cost collapses. The marginal cost of building a new application approaches the cost of describing what you want.
This is not a gradual shift. It is a rupture.
When building becomes cheap, standardisation becomes optional. When standardisation is optional, compromise is optional. When compromise is optional, the entire logic of general-purpose, multi-tenant SaaS stops being inevitable and starts being a choice, and increasingly, a poor one.
Software Is Now Disposable.
That's Not an Insult.
Here is the idea most people, including many building these tools, have not yet absorbed:
The application is disposable.
Not worthless. Not low quality. Disposable in the precise economic sense: its value no longer comes from permanence, from an installed base, from the switching cost it creates. It comes from fit, right now, for this problem, for this organisation.
Three things follow from this that I think are genuinely new:
You use it and move on
Software built for a specific task, a campaign sprint, a procurement cycle, a due diligence process, does its job and you're done. You don't maintain it, migrate it, or protect your investment in it. You made it for a reason. The reason ends. So does the software.
You upgrade the intelligence, not the product
When a better model arrives, a new version of Claude, a leap in capability from Claude Code, a new architecture, you plug it into the back end. The logic you built stays. The intelligence that powers it improves. No value destruction. No migration project. No six-month implementation. Just better.
You need it, you make it
The era of “I built it, now you buy it” is ending. The new logic is simpler and more radical: when you have a problem, you build the exact tool for that problem, at the moment you need it, shaped precisely to your context. Not a close approximation. Not a platform you customize at great expense. The actual thing.
This is not complex. It is just very different from everything we've normalised.
Scale Doesn't Disappear. It Moves.
The obvious objection: if everyone builds their own, where's the leverage? Where does scale come from?
It doesn't disappear. It migrates upstream.
Old Model
Scale lived in the product — the installed base, the switching costs, the accumulated user data.
New Model
Scale lives in the capability to configure intelligence correctly — understanding a business deeply enough to describe what it needs.
Pattern recognition across many such configurations, the ability to translate organisational reality into working systems quickly and without waste.
Scale comes from reconfiguring logic for more use cases, not from replicating the same product to more users.
This looks different. It doesn't look like a SaaS company. It looks like a practice, one with dramatically more leverage than traditional professional services, because the gap between understanding and implementation has collapsed.
The organisations and people who develop that capability will have a durable advantage. Not because they own a product. Because they own a muscle.
The Question Nobody Is Asking
Most organisations are still asking the wrong questions about AI. Which SaaS vendor has the best AI roadmap? How do I add AI features to my current stack? Which copilot should my team use?
These questions assume the stack is fixed and AI is an ingredient you add to it. The actual shift is that AI makes the stack itself the variable. You don't integrate AI into your software. You build your software from intelligence, on demand, for the specific thing you need.
The right question isn't “which tool should I use?”
It's: what does this actually need to do, and what's the fastest way to have something do exactly that?
Sometimes the answer is still off-the-shelf. Compliance, financial controls, security; processes where conforming to a standard is the point. Keep the standard software.
But for every process where your organisation creates its specific value, its relationships, its decisions, its knowledge, its edge, precisely fitted is now affordable. Which means tolerating imprecise fit is now a choice, not a constraint.
Most organisations don't yet realise they've been handed that choice.
The Canary Note
Citrini ends by reminding the reader: you are not reading this in June 2028. It is February 2026. The S&P is near all-time highs. The feedback loops haven't started yet.
Same applies here. The transition is underway but not complete. Most organisations are still in the phase where the old model works well enough to seem permanent. Most software companies are still protected by installed bases large enough to feel safe.
But the economics have already changed. The technology to act on them already exists. The only thing lagging is the mental model.
The shift from software as capital investment to software as disposable expression of applied intelligence isn't coming. It's here. Just unevenly distributed, as futures tend to be.
The organisations that move first won't just survive it. They'll look, in retrospect, like they simply understood something obvious before everyone else did.
Because it is obvious, once you see it.
When the cost of building the exact right thing collapses to the cost of describing it, you stop tolerating approximations.
That's not a technology story. It's a logic story. And the logic is already settled.
Joanna Bakas
Managing Partner at Frontira International, an AI automation consultancy specialising in intelligent systems design for SMEs and enterprise clients across Europe.