SaaS Will (Not) Survive. One Marketer’s Take
A joke of mine did pretty well:
“Actually, vibe coding has been around since programming languages first appeared.
It’s just that back then, the people doing the vibe coding were software clients and project managers.
And instead of neural networks, they used programmers.”
As I’m writing this, it has 28,300 views, 416 likes, 16 reposts, 11 bookmarks, and 14 replies on Twitter. Every joke is only partly a joke. The rest is harsh melodrama — sometimes with a touch of tragic pathos.
If you’ve studied the JTBD approach to marketing and business strategy, especially through Tony Ulwick, you’ll know that there are several growth strategies depending on how well the customer’s Job is currently being served and how much they’re willing to pay for it.
There’s a disruptive strategy: make the solution simpler and cheaper for overserved customers who are already getting TOO MUCH of what they’re being sold.
There’s differentiated: make it better for underserved customers who aren’t getting enough from existing solutions and are willing to pay more.
And then there’s dominant: make it both better AND cheaper.
Ulwick also uses a 20% benchmark: for a new solution to have a real shot in the market, it should deliver a meaningful improvement in getting the Job done — roughly 20% or more.
So:
- If something can do the Job 20% better for THE SAME money, why wouldn’t you buy it?
- If something can do the Job significantly cheaper with THE SAME effectiveness, why wouldn’t you buy it?
- If something can do the Job both cheaper AND 20% better, then WHY HAVEN’T I BOUGHT IT YET?
There’s another important distinction here: customers can be divided into overserved and underserved.
Put simply, there are those who are already getting TOO MUCH performance from a product relative to what it costs them — and those for whom the existing solutions still aren’t good enough to get the Job done properly.
The first are overserved. The second are underserved.
And a product can target either group, depending on which part of the matrix I described above it plays in.
To make this clearer, here’s an example from the same Tony Ulwick JTBD book — technically a lead magnet.
Take Uber. You call a car, get in, and go. But for people who don’t particularly care about having the whole car to themselves, a shared-ride option appeared: pick someone else up along the way. It costs less while still doing the same Job of getting you from point A to point B.
That’s for overserved customers.
At the same time, there’s Uber Black: a premium car, higher requirements for drivers, more comfort. That’s movement toward underserved customers for whom the basic “call a car, get in, arrive” experience isn’t enough.
Now you’re ready for the point about what’s happening to the market for IT work, IT products, and SaaS — as promised in the title.
A huge number of frontend tasks are SIMPLE.
Take WordPress. You either used ready-made themes — and even with FSE, you still had to adapt them — or you needed a programmer to do anything remotely more interesting than what the theme gave you out of the box.
Even if the task itself was simple, you still needed someone who had already spent time learning JSON, PHP, HTML, CSS, JS (Next, Node, or just plain vanilla JS) to give you at least a somewhat more interesting UX and UI and make your packaging look different from your competitors’.
Take design. A huge number of design tasks are SIMPLE too.
Make an ad banner — probably for a split test, meaning it’s basically MADE TO DIE. That banner may never be used again. At best, it gets recycled somewhere in the content plan.
Or make a mockup JUST to get a rough idea of what something might look like. Then the mockup ends up buried in some archive folder — if it isn’t deleted altogether.
You know how much designers HATE revisions?
And they’re right. Plenty of established designers explicitly put it in their contracts: you get, say, four rounds of revisions, submitted in batches. After that, you pay again.
Or take copy.
Sometimes text is just information. Something you need to shove into the same split test, or put somewhere as an instruction. It doesn’t need an authorial voice. It should have exactly zero personality — just clean, information-first copy in the style of Maxim Ilyakhov’s infostyle (a Russian school of concise, utility-first writing).
And there sits a separate copywriter producing all of it. Or you do it yourself.
I think you can see where I’m going with this.
In these situations, neural networks OBVIOUSLY do either the same thing for not just 20% less, but sometimes MUCH less.
Or they do it better for the same cost. Or sometimes BOTH better AND cheaper. And that’s why EVERYONE got shaken up — programmers FIRST. Because in the digital age, code became exactly the kind of service with a huge number of overserved customers.
I remember working on some UI for my website and not being able to decide which approach I liked better. So I asked Codex to build three branches, one solution in each. I switched between them, picked the one I liked — or rather, made a fourth one that combined solutions 1 and 3 — and that was it.
I feel sick imagining how much money and how many nerves that would have cost me with a human programmer. Imagine being asked to write three versions of the same thing, knowing there’s a good chance two out of three will be thrown straight into the trash.
Or that after all three, you’ll be asked to make a fourth.
In most cases, you’d get some grumbling along the lines of: can we PLEASE sit down first and figure out exactly what we need, because I AM NOT writing code just so it can go straight into the bin.
Because I am a highly paid programmer. Even if I’m a junior — it’s still IT, and rates are higher here than in most places.
Or, as a compromise, let’s order some concept mockups from a designer first, so at least we can throw the pictures — and the designer’s work — into the trash instead of the code — and the programmer’s work.
With AI, it goes differently.
Ask. Generate the code. Look at it. Figure out what you like. And you figure it out by looking at the actual implementation. Once you’ve made the decision, ship the good version to production and delete the rest. So what does any of this have to do with SaaS — and whether it lives or dies?
Everything.
If something can be done more easily and cheaply through a neural network — imperfections, rough edges and all — then that SaaS will die.
Flowchart and mind-mapping tools may be among the first to take the hit. Or, more precisely, their users will reassess what those tools are worth. Need to throw together a block diagram as a one-off?
That’s a job for AI generation.
Need to sit with it, move things around, think through it? Then sure, you might still want a mind-mapping service. Although, hypothetically, you could just ask the AI to regenerate the thing from several different angles and think through it that way.
Notion, on the other hand, has become ONLY MORE CONVENIENT for me personally because of MCP with GPT. Because sometimes you genuinely need to store data on separate shelves.
Notion gives you a nice sandbox for creating databases. But then you create the shelves you actually need — and suddenly there are ten of them. Now you need at least ten clicks just to put the data where it belongs. And you still have to type the text manually — text that is just information, with no style and no TOV.
Once you have more than two databases like that, you start looking at Notion as a headache. They’re all necessary. Every one of them emerged as a separate entity for a reason. But pretty quickly, working with them becomes a burden. Notion handles your Core Functional Job well.
Its Consumption Chain Job, though… And then MCP walks onto the stage. You just dictate everything to the neural network in plain language, and it runs off, fills everything into Notion, and doesn’t complain. Two or three minutes later, you can have ten databases populated, connected, and properly linked to one another.
Notion becomes a context repository and external memory.
A place you can open, reflect on a specific snapshot of data yourself, click around manually, write something, and bring that context into the exact state you need.
Notion became a prompt.
It makes what a neural network can already do 20% cheaper and/or more effective, because Notion is a strict software product. Its features and functions aren’t regenerated by the neural network from scratch every time you ask for something. Because neural networks aren’t perfect either. That’s why prompt engineering emerged as a discipline in the first place.
You have to learn how to write prompts in such a way that dynamic chaos theory — the butterfly effect — somehow gives you stable generations. Anyone who has created AI characters knows that this is intellectual work too: tuning a prompt until the visualization remains stable and recognizable despite all the generative variability.
So:
The SaaS products that survive will be the ones that stabilize a Job AI can already do and make that Job more efficient.
The SaaS products where that stability itself is overserved will die.
The specialists who survive will be the ones who figure out how, with AI, they can get a particular Job done 20% cheaper and/or more effectively than the user could by using AI alone, without that specialist.
And remember: cheaper is a flexible concept.
Some stakeholders have already learned the lesson that being cheap can make you pay twice. They’ll happily hire a specialist who uses AI rather than wave them off with, “I can do it myself with AI.”
Well… Do it. If you get lucky, you get lucky.
By the way, I’ve started building my own marketing SaaS.
That one is going to survive.
I think I’ll be ready to talk about it pretty soon.
It’s being built in TypeScript, even though literally all I know about TS is that it’s “typed JavaScript.” And, to be fair, I haven’t even googled it or asked the same neural network to explain it to me in more detail…
You can catch a hint of what that SaaS is in this article:
Value Exchange Ecosystem — https://thequietorbit.com/value-exchange-ecosystem/
Stay tuned, basically.
Datalinks record closed. Before Planetfall: 0073-03-06