Build vs buy in the AI era: which SaaS tools should you still pay for?
For fifteen years, the default answer to almost every internal tooling need was the same: find the SaaS that does it, pay per seat, adapt your process to the product. That default is breaking. At KLIMB, we recently replaced part of our Notion setup with a planning tool we built ourselves, without launching a classic software project.
TL;DR Keep SaaS for commodity, compliance-heavy and system-of-record tools. Build what encodes how your team actually works. Treat security as part of the build, not an afterthought. And watch the next two shifts: custom software as a service (CSaaS) and service-as-a-software.
This isn't a story about one tool. It's a shift in who gets to make software, and it forces every company to look at its stack with fresh eyes.
Why is the software market going through a major shift?
Because AI has collapsed the cost of building software, and the market is now questioning what's still worth renting.
In early 2026, investors staged a brutal repricing of enterprise software. According to Stripe Economics, nearly $1 trillion in valuation was wiped out by early February, and perhaps up to $2 trillion in the weeks that followed. Traders nicknamed it the "SaaSpocalypse".
The twist: SaaS revenues didn't collapse. Stripe's own SaaS Index shows that revenue growth actually accelerated through the sell-off, with year-on-year growth above 30%. The panic was about the future, not the present.
That's the real signal. The question isn't whether SaaS disappears. It's which software keeps its value when any team can produce a working tool in days. Generic, seat-based tools that mostly organize information are the most exposed. Tools that hold critical data, handle compliance or sit at the core of a business are far harder to replace.
What happens when a design studio replaces Notion with its own tool?
It stops adapting its process to the software, and starts building software around its process.
KLIMB is a brand and website studio, not a software company. Like most agencies, we ran our planning in Notion for years. It did the job, but it was a generic workspace stretched to fit a very specific reality: brand projects, website builds, designers shared across clients, phases that overlap.
So we built Studio, our own planning and project management tool, and vibe coded it with Lovable. It wasn't a classic software project with a product team and a long roadmap: it was built by the people who run the studio, with AI doing much of the coding. The tool reflects exactly how a brand and website studio plans its work, because the people who use it every day are the ones who designed it.
Studio isn't alone. We now run a small set of internal tools built the same way, including one to curate resources and benchmarks for clients and another to assemble commercial proposals. A year ago, each of them would have meant another SaaS subscription, or a spreadsheet nobody loved.
The lesson isn't "cancel all your subscriptions". It's that the question "should we build this?" is now a reasonable one to ask, even for a team of designers and strategists.
How does this change the roles inside a team?
When everyone can build, job titles stop describing tools and start describing problems to solve.
We saw it first with our developers. KLIMB used to have Webflow Engineers. Their job has since evolved into Solution Engineers. They still build our clients' websites, of course, but no longer tied to a single CMS: the right platform depends on the project, not on a skill set. And they now help build our own internal tools too.
That change goes far beyond the tech team. It's a shift in mindset for every person in the studio. We believe each person should now become the engineer of their own function, augmented by AI:
- Brand Strategist becomes Brand Strategy Engineer
- Website Developer becomes Website Engineer
- Project Manager becomes Project Engineer
- Designer becomes Design Engineer
The word "engineer" isn't about writing code. It's about owning a system instead of a list of tasks: understanding how your work gets produced, spotting what can be automated, and building the tools or workflows that make your craft scale. The strategist still thinks, the designer still designs. They just stop waiting for someone else to build what they need.
Our own roles as founders are evolving too. With execution getting faster and cheaper, our job is less about managing production and more about building a company that genuinely moves its market forward, not one that simply follows the latest trend.
Which tools should you keep in SaaS, and which should you build?
Keep the tools where reliability, compliance and scale matter more than fit. Build the ones where fit is the whole point.
The way we use AI every day changed how we look at our stack. Every subscription now gets the same question: is this tool doing something hard, or just organizing our work in someone else's format?
Keep it in SaaS:
- Systems of record: accounting, payroll, CRM core, legal documents
- Tools with heavy compliance or certification requirements
- Infrastructure you never want to maintain: email, storage, authentication
- Products with strong network effects or integrations you depend on
Consider building it:
- Workflows specific to how your team operates
- Internal dashboards and views on top of your existing data
- Tools used by a small team, where per-seat pricing hurts
- Processes you've been forcing into a generic tool for years
A simple test helps: if the tool disappeared tomorrow, would you rebuild it the same way? If the answer is "no, we'd build it around how we actually work", it's a candidate. If the answer is "we'd buy it again", keep paying.
What about security and data protection?
This is the real cost of building: when you own the tool, you own the risk.
A SaaS vendor spends years and serious budget on security, certifications and audits. When you build your own tool, that responsibility moves to you, whether you planned for it or not. And AI-generated code doesn't solve the problem on its own.
In its 2025 GenAI Code Security Report, Veracode tested more than 100 language models on 80 coding tasks. In 45% of cases, the models picked an insecure way to write the code. The report also notes that models are getting better at writing working code, but not at writing secure code.
Vibe coding platforms have had their own wake-up calls. In 2025, a security researcher scanned 1,645 apps built with Lovable and found that 170 of them left their databases readable by anyone, mostly because access rules were never configured (CVE-2025-48757). The platform wasn't hacked; the apps were simply shipped without the right safeguards.
For a European company, the stakes go beyond embarrassment. An internal tool that stores client names, contracts or project data falls under GDPR like any other system. Before building, a few rules are worth setting:
- Keep sensitive data where it already lives. Let your custom tool read from your systems of record rather than duplicating client or financial data.
- Configure access control from day one. Authentication, roles and database rules are part of the build, not a later sprint.
- Get a technical review before going live. Someone who reads code should check what the AI produced, especially permissions and API keys.
- Decide who owns maintenance. Security patches don't stop once the tool works.
What is custom software as a service (CSaaS)?
CSaaS is software built around one company's way of working, then hosted, secured and improved by a provider on a subscription.
It sits between the two classic options. With SaaS, you get reliability but adapt to a generic product. With an internal build, you get a perfect fit but carry the maintenance and the security risk. CSaaS tries to keep the fit and hand off the burden.
The model usually combines a one-off build fee with a monthly subscription that covers hosting, support and continuous improvement. Providers such as CURA already describe the market as moving "from SaaS to CSaaS".
Why does it reinforce the shift? Because it removes the strongest argument against building. Most teams hesitate not because they can't get a tool made, but because they don't want to own servers, patches and access rules. CSaaS lets a company commission software that matches its process exactly, without becoming a software company. For a mid-sized business, the comparison is no longer "custom vs off-the-shelf". It's "which provider maintains the version that fits us".
What is service-as-a-software, and how is it different?
With service-as-a-software, you don't buy a tool to do the work. You buy the work, done by software.
The idea was popularized by the VC firm Foundation Capital. Classic SaaS sells access to a tool: your team still does the job, just faster. Service-as-a-software flips the model. AI agents carry out the task end to end, and the provider is paid for the outcome, such as a resolved support ticket or a processed invoice.
The stakes are much bigger than software budgets. Foundation Capital estimates a $4.6 trillion opportunity, because AI starts competing not with software subscriptions but with salaries and outsourced services.
For a company reviewing its stack, this adds a third question. Beyond "build or buy?", it's now "should this be a tool at all, or a service delivered by agents?". We'll go deeper into service-as-a-software, and what it means for agencies and their clients, in a dedicated article.
Frequently asked questions
Is SaaS dying because of AI?
No. Stripe's data shows SaaS revenues kept growing through the 2026 sell-off. What's changing is which SaaS is worth paying for: generic tools are under pressure, critical systems are not.
Can a non-technical team really build its own internal tools?
Yes, for focused internal use cases. Tools like Lovable make it possible to ship a working app without full-time developers. Security, data access and maintenance still need someone technical in the loop.
What's the difference between CSaaS and service-as-a-software?
CSaaS delivers a custom tool your team uses, maintained by a provider. Service-as-a-software delivers the result itself, with AI agents doing the work and the provider paid on outcomes.
Which tools should never be rebuilt in-house?
Systems of record such as accounting, payroll and core CRM data, plus anything with heavy compliance requirements. The risk and the maintenance cost rarely justify the gain in fit.
Is vibe coding safe for business tools?
It can be, if security is treated as part of the build. Configure access rules from day one, avoid duplicating sensitive data, and have the code reviewed before real users rely on it.
What should you do next?
List your current subscriptions and sort them into three groups: keep, rethink, build. Start with the tools your team complains about most, or the ones where you pay for fifty features and use five.
Then pick one low-risk internal workflow and build it. Not your accounting, not your client data: a planning view, an internal dashboard, a proposal builder. You'll learn more about what's worth building in two weeks of practice than in two months of debate.
That's how we approached it at KLIMB, and why we now look at every tool in our stack differently. The best software for your team might be the one that doesn't exist yet.
Sources: Stripe Economics (September 2026), Veracode (July 2025), Matt Palmer, CVE-2025-48757 (2025), Foundation Capital.
