For years, SaaS was the obvious answer.
Need a CRM? Buy HubSpot or Salesforce. Need project management? Buy Asana or Monday. Need personal finance tracking? Download Emma, Snoop, YNAB, or whichever app has the least alarming shade of purple this week.
The bargain was simple. You paid every month and someone else handled the messy bits. Hosting. Security. Updates. New features. Backups. Support. The works.
In return, you accepted the tradeoffs. You got a product built for thousands or millions of people, not for you. You paid for features you might never use. Your data lived where the vendor wanted it to live. Your workflows bent around the software, rather than the other way round.
That bargain made sense when bespoke software was expensive.
AI is changing the maths.
Not perfectly. Not magically. And certainly not in the “fire your entire software stack by Friday” way that LinkedIn prophets seem to enjoy. But enough to make the old SaaS model look less inevitable than it did a few years ago.
The market has noticed
There is a reason investors have started asking awkward questions.
Reuters recently reported that Salesforce shares had fallen nearly 33% in 2026, after dropping more than 20% in 2025, as investors worried that AI tools from companies such as Anthropic and OpenAI could pull enterprise customers away from traditional software products. Wall Street has even found a suitably dramatic name for the fear: the “SaaSpocalypse”.
Rightmove gives us a useful UK example. In November 2025, its shares fell as much as 28% after it warned that increased AI investment would slow profit growth in 2026. The company said AI was becoming central to its plans, but investors were not exactly popping champagne. More like quietly looking for the nearest fire exit.
HubSpot tells a slightly different version of the same story. Its problem is not that it has failed to embrace AI. Quite the opposite. It has been adding AI aggressively. The worry is that AI makes users more efficient, and if users become more efficient, companies may need fewer seats. That is awkward when much of your revenue model is built around charging for seats. Barron’s reported that HubSpot has already started moving towards outcome-based pricing and AI credits to reduce its dependence on human headcount.
That is the real pressure point.
AI does not need to replace every SaaS application to damage the SaaS model. It only needs to make customers question what they are paying for.
The SaaS defence is not stupid
The loudest SaaS critics often make the same mistake as the loudest AI hype merchants. They confuse a good demo with a reliable system.
A CRM is not just a few screens and a database. It is permissions. Audit trails. Integrations. Reporting. Workflows. Compliance. Disaster recovery. Billing logic. User management. Export tools. Support processes. A stack of necessary but unglamorous details.
This is where SaaS founders have a decent argument.
Their strongest case is not “our interface is lovely”. It is “our platform is the source of truth”.
If your sales pipeline, customer history, marketing journeys and support tickets already live in HubSpot, then the best AI agent may not be the one that replaces HubSpot. It may be the one that works inside it. Or across it. Or against its API. That is the ecosystem argument. Data gravity still matters.
There is also the reliability problem. Salesforce’s SCUBA benchmark, which tests AI agents on realistic CRM tasks, found that even strong agent systems still struggle with enterprise workflows. Closed-source systems achieved up to 39% task success in a zero-shot setting, improving to 50% with demonstrations. That is promising. It is also not the sort of number you want near your quarterly revenue report without a very sturdy adult in the room. The spreadsheet jazz player, again, cannot be left unsupervised near the investor pack. (SCUBA: Salesforce Computer Use Benchmark)
So no, serious companies are not about to replace every operational system with a vibe-coded spreadsheet and a cheery “looks good to me”.
At least, one hopes not.
But the SaaS critique is not stupid either
SaaS has always had a hidden cost: it is generic by design.
A platform like HubSpot has to be many things to many companies. That is its strength. It is also its burden. Most users only touch a small slice of a large product. SaaSfactor’s piece on feature underuse makes the familiar product point: many SaaS users regularly engage with only a minority of available functionality, while the rest sits there for edge cases, advanced workflows, or “we built this for that enterprise prospect in 2019 and now nobody dares remove it” reasons. (Why 80% of Your SaaS Features Are Underused)
The customer still pays for the bundle.
That used to be fine. Building your own alternative was usually absurd. You did not commission a bespoke CRM because you disliked three tabs in HubSpot. You sighed, paid the invoice, and trained your staff to click around the bits they did not need.
AI changes this at the margins.
It lowers the cost of building small, specific tools. It lowers the cost of data import and export. It lowers the cost of writing glue code. It lowers the cost of making a one-off interface for a one-off problem.
And once those costs fall far enough, the old SaaS tradeoff starts to wobble.
Why pay for a large generic system if the job you actually need done is narrow, temporary, or highly specific to your business?
My personal finance experiment
I ran into this recently with my own finances.
The obvious route was SaaS. Emma, Snoop and similar apps are slick. They give you a polished interface, bank connections, spending categories, trends, alerts, and that satisfying feeling of getting your life in order.
I can see the appeal. The mildly obsessive part of me quite enjoys spending an evening uploading statements, categorising transactions, tidying everything into neat little boxes. It has the emotional texture of spring cleaning. I feel like a responsible adult. A dangerous illusion, I know.
But the problem I wanted to solve was not “which app do I need”. It was:
Where is my money going?
What can I cut back?
What patterns am I missing?
What questions should I be asking?
So I tried something different. I gave ChatGPT redacted CSV exports from my bank and credit card statements and asked it to build me a spreadsheet tailored to the analysis I wanted.
The first useful version took about twenty minutes.
That is the bit that feels new. Not that AI produced a perfect financial product. It did not. The follow-up revisions took longer. A few things needed correcting. Some analysis needed tightening. Versioning was fiddly. I had to push it, check it, and occasionally stop it wandering off into spreadsheet jazz.
But even with that friction, the experience was different from SaaS.
I was not learning someone else’s product. I was not conforming to someone else’s categories. I was not paying another monthly subscription for one small job.
I had a temporary, custom tool that answered my actual question.
That matters.
The SaaS app would have given me polish. The AI-generated spreadsheet gave me fit.
The wrong problem
This is where SaaS can quietly nudge us into solving the wrong problem.
I did not really want to manage a personal finance app. I wanted to understand my finances.
Those are not the same thing.
A lot of software turns the user into a clerk. Upload this. Categorise that. Approve this rule. Fix that sync. Reconcile this account. Admire our dashboard. Upgrade to Pro.
It feels productive. It may even be productive. But a lot of the time you are doing administrative labour so the software can eventually tell you something useful.
AI flips part of that around.
If it can ingest the messy data, normalise it, categorise it, generate the analysis and let you ask follow-up questions, then the product does not need to be a permanent destination. It can be an ephemeral working surface.
Not an app you live inside.
A tool you summon when needed.
The middle ground: durable data, flexible intelligence
This does not mean every business should replace SaaS with disposable AI scripts. The better model is a middle ground: keep the parts that need to be deterministic. Data models. Permissions. Audit logs. APIs. Version history. Backups. Validation. Security boundaries.
Then let AI handle the flexible layer. Importing awkward files. Mapping columns. Generating summaries. Producing charts. Answering follow-up questions. Drafting reports. Surfacing anomalies. Translating business questions into structured queries.
For the finance example, that might mean a small bespoke persistence layer for transactions. Something auditable, fast and boring. It defines what a transaction is, how categories work, where source files came from, and what changed over time.
Then AI sits above it.
Not as the source of truth. As the analyst, importer, explainer and interface.
That shape feels much more durable than either extreme.
Pure SaaS says: use our entire platform forever.
Pure AI hype says: replace everything with a prompt.
The useful answer says: keep a reliable core and make the interface more adaptive, tailored to your needs.
This is where bespoke AI starts to make commercial sense. Not as a grand replacement for every system in the business, but as a targeted layer around the work that never quite fits the tools you already pay for. The awkward imports. The recurring reports. The hand-built spreadsheets. The data trapped between departments. The small internal workflows that are too specific for SaaS, but too valuable to leave manual.
The real threat to SaaS
The threat is not that SaaS disappears.
Software is not going anywhere. Web apps are not going anywhere. Databases, APIs, user permissions and audit logs are not going anywhere. The boring infrastructure has survived every hype cycle so far. It will survive this one too.
The threat is that the old SaaS bundle becomes harder to justify.
Per-seat pricing looks odd when AI reduces the number of seats.
Siloed data looks worse when agents need context across systems.
Bloated products look expensive when bespoke tools become cheaper.
Closed ecosystems look less attractive when interoperability becomes the main value.
That last point is especially important.
For years, centralised data was a moat. If all your customer data lived in a SaaS platform, that made the platform sticky. Painful to leave. Easy to expand. Lovely for the vendor.
In an agentic world, the moat can become a wall.
If an AI system needs to reason across your CRM, finance system, email, support desk, product analytics and internal documents, then every silo becomes friction. The more your SaaS tools trap context inside their own worlds, the less useful they are as part of a wider intelligent workflow.
Data gravity still matters.
But data captivity starts to look suspicious.
So, is SaaS dead?
No. The lazy version of “SaaS is dead” is wrong. Enterprises still need trusted systems of record. Regulated workflows still need guardrails. Mission-critical processes still need vendors who can be sued, audited and phoned at 3am when everything catches fire.
A recent paper on the AI buy-or-build decision makes this point well. It argues that the “SaaSocalypse” thesis is overstated for most enterprise software, but that building becomes more attractive for commodity utilities and differentiating custom applications. In other words: do not rebuild your bank. Do consider building the weird internal workflow that makes your business different. (The Buy-or-Build Decision, Revisited)
It means less buying a generic subscription because building was too expensive. Less accepting bloated products because custom tools were out of reach. Less pretending that seat-based pricing maps neatly to value. Less letting data sit in silos when the next wave of AI needs context to be useful.
The old SaaS model was built for a world where software was expensive, slow to develop and painful to maintain.
AI makes software cheaper to shape around the user.
That does not kill SaaS overnight. But it does change what businesses should expect from it.
SaaS used to win by being available, polished and cheaper than custom development.
Now it has to prove something harder.
It has to prove it is better than the thing you could build for yourself.
At Boring.ai, this is the kind of work we care about: not replacing your business with a chatbot, but building small, reliable AI systems around the dull, expensive, repetitive work that slows people down. If your team is paying for software and still doing manual work around it, that is probably where to start.
