Embedded Files
AI makes building and switching cheap. The moat shifts to domain expertise, proprietary data, and distribution.
Copyright 2026, Bernd Schoner
February 18, 2026
Six months ago, I wrote an AI impact strategy paper for my (then) employer—a SaaS company. I thought I was being pretty harsh and pessimistic about the challenges AI would bring to our specific business and to our industry.
I was wrong.
The trends I identified last year have accelerated dramatically—and a few more have emerged. Each one alone could subvert the software industry as we know it. Together, they’re rewriting the rules of how software gets built, sold, supported, and replaced.
1) Development cost → ~0
By last fall, I had convinced myself that the productivity of writing code had increased by 10× to 100×. I got a lot of pushback—interestingly, much of it from professional developers—arguing that real savings for production code were “only” 30–40%.
Since then, the tooling has shifted again—multiple times.
In a recent New York Times piece, software entrepreneur Paul Ford described how a software project that used to cost $350k can now be implemented by one person in an afternoon with the help of a $200 Claude Code license. Whether you believe the exact multiplier is 10×, 100×, or even more, the directional change is undeniable: the marginal cost of producing competent software is collapsing.
Of course, there are counterarguments:
“AI doesn’t produce production-quality code.”
“This code isn’t secure.”
“It can’t handle our edge cases.”
Maybe not today, but soon (for many domains literally tomorrow) those objections will not matter. Much like self-driving systems can outperform humans in aggregate safety metrics, AI-generated software will increasingly outperform human-written code across a broad classe of problems—both in quality and in speed. Development cost will effectively approach zero.
2) Enterprise support → ~0
Enterprise support and inflated consulting offerings—historically key drivers of premium SaaS pricing and large integration fees—are heading the same way.
Customers expect products to work out of the box more than ever. And when software does break, they increasingly consult an AI to diagnose, patch, and work around issues immediately.
They won’t wait for an overseas call center to pick up a ticket 24 hours later—then take weeks to resolve it.
Enterprise support won’t disappear entirely (regulated industries and complex deployments will still need humans). But as a default business model, “support as the justification for margins” erodes rapidly.
3) Stickiness / switching costs → ~0
AI is particularly good at recreating software that already exists in some form:
complex open-source systems
porting an iOS app to Android
re-implementing internal tools that match common patterns
Unfortunately for incumbents, this also includes:
replacing a vendor with a cheaper vendor
rebuilding a subset of functionality from scratch
migrating data and workflows with far less effort
The act of migration—dreaded by risk and IT departments alike—is becoming something you do without giving it much second thought. Much like IT replaces aging hardware, organizations will increasingly choose the software that serves them best right now—then switch again.
This is amplified by the accelerating rate of innovation. Software built last year often can’t compete with what’s built today. Incumbents either bake rapid innovation into how they design and ship—or they lose to the latest upstart.
What doesn’t go to zero
If development cost, support cost, and switching costs compress, what’s left? The answer is differentiation—but not the kind most SaaS companies are optimized for.
4) Differentiation through data, domain knowledge, and customer access → ∞
The saying “data is the new gold” comes from an era when AI was mostly about analytics and data mining. But the core point still holds—especially when your data is proprietary and competitors can’t access it.
At a time when the cost of processing trends toward zero and capability increases dramatically, the value of proprietary data as raw material for value generation can explode.
A similar effect happens with domain knowledge. Builders who combine strong engineering with deep vertical understanding keep an edge. Everyone else produces increasingly commoditized, generic output.
These forces push in a direction that’s uncomfortable for many SaaS vendors:
more software becomes contextual and bespoke
more value accrues to whoever owns the data
winners have direct customer access and tight feedback loops
Which leads to the next point.
5) In-house bespoke software development and maintenance → ∞
Many former software customers will choose to build and maintain solutions internally—because they can.
The operating business typically
understands the customer problem best;
has access to vast proprietary data;
can iterate faster when the cost of building falls;
can tailor solutions precisely to workflows (instead of adapting workflows to a vendor).
In-house development won’t replace every SaaS category, but the center of gravity shifts. “Buy vs. build” changes when building becomes radically cheaper, faster, and safer.
6) Product strategy, product management, and UX → the new bottleneck
Here’s the twist: for all its ability to execute on a plan, AI is still not great at choosing the right plan.
It can generate options, drafts, and variations. But it struggles with strategy grounded in:
real market research
customer insight and segmentation
sharp product tradeoffs
coherent positioning
state-of-the-art UX that users actually love
These tasks become the new bottleneck.
Companies that employ the best people in product strategy, product management, and UX will thrive. Everyone else will simply implement more failing products—faster.
Sustained advantage will come from doing the hard work of defining what a winning product means now, then executing quickly. AI will make these product stakeholders dramatically more productive, but humans will remain in the driver’s seat—and become more impactful, not less.
A bimodal future
The picture that emerges is bimodal:
On one hand, software vendors and customers alike can’t continue to do business as usual. Heavy processes, large teams, legacy architectures, and obsolete specs become existential liabilities.
On the other hand, there is real opportunity. Capturing it will require our best minds, precise day-to-day execution, and product processes that are state-of-the-art—and constantly evolving.
Recent market turmoil around public software companies is only one expression of a deeper shift and uncertainty shared by all stakeholders: investors, entrepreneurs, individual developers, and customers.
The SaaS era isn’t ending because software stops mattering.
It’s ending because software becomes too easy to produce, and value migrates to what’s hardest to copy: subject-matter insight, proprietary data, distribution, and trust.
Page updated
Report abuse