What five conversations at AI Leadership Day taught me about the future of SaaS, and why the lazy “SaaS is dead” narrative misses the point entirely.
Last week, we hosted AI Leadership Day 2026 in Amsterdam. Five speakers answered five questions we’d been wrestling with for months. An audience of SaaS founders, CTOs, and investors who didn’t come for platitudes, they came because the ground is shifting under their feet and they want to know where to step next.
I want to share what I took away from that day, mixed with my own perspective. Because after years of building We Love SaaS and having hundreds of conversations with European SaaS leaders, I’ve landed on a conviction that keeps getting reinforced: the role of software is fundamentally changing. Not dying. Changing. And if you don’t understand the nature of that change, you’ll either overreact or underreact, both of which can be dangerous.
The shift nobody talks about clearly enough
Here’s the thing: for the past (at least) 10 years, software was a tool. Humans used it to get work done quicker, more accurately, more efficiently. The entire SaaS model was built on this premise: give people a better tool, charge them per seat, grow as they hire more people who need access.
That model is breaking down. Not because the tools stopped working, but because the job description of software itself has changed. Software doesn’t just help people do work anymore. Software does the work.
That’s not a subtle change, that’s a profound shift.
When your accounting software doesn’t just display outstanding invoices but actively chases them, processes them, and reconciles them, you’re no longer selling a tool. You’re selling a financial operation. When your customer service platform doesn’t just route tickets but resolves 80% of them autonomously, you’re not selling a helpdesk; you’re delivering customer outcomes.
Sequoia recently published a sharp piece called “Services Are the New Software” that puts numbers behind this intuition. Their argument: for every dollar a company spends on software, six dollars go to the services that make that software work. The next trillion-dollar company, they argue, won’t sell tools, it will sell outcomes. Closed books instead of accounting software. Resolved tickets instead of support platforms. The underlying AI models are increasingly commoditized; what matters is who implements them and delivers the result.
This aligns with what the sharpest VCs in Silicon Valley have been signaling for months. The old SaaS playbook (streamline a workflow, charge per user, expand as the customer’s headcount grows) is structurally challenged. When AI agents do the work, there are fewer seats to sell. The value shifts from enabling humans to replacing human effort. “Software is eating labor,” as some have put it. And that rewrites the economics of the entire industry.
Now; is SaaS dead?
I hear this narrative constantly, and I think it’s lazy.
Shallow SaaS tools that were already commodities with a thin feature layer? Yes, those are in serious trouble. The chatbot-on-the-homepage, the basic if-then automation, the glorified spreadsheet with a login page; these were never defensible, and AI is now their death sentence.
But the world will create and buy more software than ever before. The demand for intelligent, outcome-oriented software is exploding. SaaS companies that adapt, that rethink what they sell, how they build, and who their user actually is, will not just survive. They’ll capture more value than the previous generation of software companies ever did. The pie is getting bigger. The question is whether you’re baking the right kind of pie.
Five questions, five realities
At AI Leadership Day, each presentation tackled a specific question we identified in the run-up to the event. Not generic “what is AI” stuff, pointed questions that European SaaS leaders are losing sleep over right now.
Here’s what we learned.
1. Should you build a new AI-native product or build agents that work with existing tools?
Aart Bijkerk, founder of Nance, had the clearest answer I’ve heard to this question. Nance is building an AI agent for finance. Not a new accounting system, but an autonomous colleague that works inside the accounting systems companies already use.
His reasoning is pragmatic and, I think, correct. Two AI-native bookkeeping startups tried to replace established tools entirely. Both failed; one ran out of money (ironic for a cash flow tool), the other turned out to rely on humans in the background. Meanwhile, the established tools like Exact Online are relatively cheap, deeply embedded in compliance workflows, and frankly good enough at what they do.
The insight that landed hardest: agents replace human labor, not tools. The real cost in finance isn’t the €50/month software subscription, it’s the people spending hours navigating that software to get work done. Nance doesn’t compete with Exact; it uses Exact more productively than any human can.
This reframes the strategic question for every SaaS company in the room. If you’re an existing platform: your next user might not be a person. It might be an AI agent. Are your APIs ready for that? Are your pricing models? Should you create agent-specific accounts with different permissions and different economics? Because if an agent can replace an employee that costs € 8.000 per month, the willingness to pay looks nothing like a traditional SaaS subscription.
And if you’re building something new: don’t rebuild the system of record. Build the intelligence layer on top of it. That’s where the value (and the margin) lives now.
2. How do we stay sovereign and compliant while moving fast with AI?
Arjé Cahn (Aimable) brought the most uncomfortable talk of the day, and probably the most necessary one. Arjé is a former CPO of a Silicon Valley double-unicorn, studied AI at the University of Amsterdam in 1997, and has lived on both sides of the Atlantic. He’s seen what happens when you move fast without thinking about where the data goes.
His provocation was sharp: if your biggest client calls you tomorrow and asks, “Where does my data go? What AI has access to it?”, can you answer that question honestly? Most SaaS founders cannot.
The reality is sobering. Even if you’ve “turned off training” with your AI vendor, your data is still sitting on servers across the Atlantic. Anthropic now stores prompt logs for five years. Those logs can be subpoenaed. And 40% of employee AI inputs contain sensitive data (names, bank accounts, personal details that end up in log files you don’t control).
Arjé made three commitments that I think every European SaaS founder should consider.
First, embed sovereignty into your product architecture, not just your contracts. A GDPR checkbox doesn’t prevent data from crossing the ocean; architectural choices do.
Second, make sure the AI in your product follows the same rules as your customer’s employees, the EU AI Act, which goes into enforcement this August, requires exactly this for high-risk use cases.
Third, abstract your AI layer so you’re never locked into a single vendor. When OpenAI triples their price next quarter, can you switch to Mistral or a local model by Friday?
The business case is real. Healthcare, financial services, legal, private equity; they all want AI, but they have to be compliant. They will buy from the companies that can prove where the data lives and who controls it. Sovereignty isn’t a drag on innovation. It’s a competitive advantage hiding in plain sight.
3. Has AI-powered engineering actually arrived, and what does it mean for SaaS teams?
Julien Simon, AI Operating Partner at Fortino Capital, delivered the talk that made the room squirm. He wasn’t interested in debating whether AI can write code. That debate, he said, is over, and anyone still having it is wasting time.
The evidence is settled. Donald Knuth, arguably the greatest computer scientist alive, recently revised his skepticism about generative AI after Claude cracked a math problem he’d been working on for weeks. Google’s DORA research group which said in 2024 that they saw no productivity uptake from AI coding published a 2025 update that said: yes, it works, but only if your engineering processes work in the first place. AI, they found, is an amplifier. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones.
That insight landed like a hammer. If your engineering team is already drowning in tech debt, if your processes are held together with duct tape, AI won’t save you. It will drown you faster. This is exactly what happened at Coinbase: they gave Cursor licenses to everyone, generated a flood of AI-written code, and immediately hit a bottleneck; nobody could review it. Output exploded, quality didn’t.
Compare that with Stripe, which redesigned its entire engineering process before scaling AI coding. They now ship over 1,300 pull requests per week with zero human-written code, running through automated test suites and review pipelines. Three engineers at Google migrated 500 million lines of code in a year, AI generated 74% of it. But the system, not the tool, made it possible.
Julien’s framework is clean: there are three layers in modern software development. The generation layer (where AI writes code) is the least interesting and getting cheaper by the day. The specification layer (where you define what to build, with what constraints, to what standard) is where most organizations are dangerously weak. And the judgment layer (where you decide whether to ship) is where senior engineers find their new, more valuable role. Plan ten times, generate once. That’s the new discipline.
The uncomfortable truth: your biggest resistance won’t come from juniors. It’ll come from senior engineers whose identity is built on being the best coder in the room. AI commoditizes that mastery. The seniors who thrive will be the ones who shift from “writes the best code” to “defines how code should be written.” That’s a leadership transition, not a technical one.
4. What does it actually take to transform a SaaS company from the inside?
Marili ‘t Hooft-Bolle, CEO of Trengo, told the most honest story of the day. No frameworks, just the raw, messy reality of transforming a customer service platform into an AI-first company.
When she took over, Trengo was a multi-channel inbox. Chat, WhatsApp, Facebook, email; all funneled into one place, with if-then automation rules that looked like a forest of branching paths. A solid product, but in a market that was about to be blown apart by LLMs.
Her first move was strategic: she made “bring an AI proposition to market” a company objective and embedded “we are future-proof” as a core value. Not as a slogan, but as an operational framework. Cross-functional hackathons replaced Friday afternoon experiments. Performance reviews were restructured: the bar didn’t just go higher, it moved to a completely different place. Recruitment criteria changed; every new hire had to demonstrate curiosity and willingness to reimagine, not just competence in the current stack.
Then came the hardest part. After 15 months of pushing, her engineering team was still dragging their feet. Not just slow; actively resisting. And this is where Marili’s story gets uncomfortably real: she replaced 85% of her engineering team. Not overnight, not recklessly, but deliberately, through performance frameworks, through candid conversations about where the industry was heading, and yes, through parting ways with talented people who simply didn’t want to make the shift. It cost Trengo two quarters of near-standstill on product development. Sales teams had no one to lean on. Customers felt it.
But the result? A 200% acceleration in both speed and quality of delivery. A team that’s smaller (they went from over 100 people to about 75) but that ships dramatically more. Engineers who don’t just build features but reimagine what’s possible.
Two things from Marili’s talk stuck with me most.
First: from creativity to reimagination. Creativity usually gets you a v5 of what you already know. Reimagination starts with a blank sheet and asks “what could this be?” That’s a fundamentally different muscle, and not everyone has it.
Second: from agile to hyper-agile. Sometimes you ship something you know will be obsolete in two months, because the learning is worth more than the longevity. To double your rate of success, you have to double your rate of failure.
5. How do you actually ship AI features without breaking your product?
Sohrab Hosseini, co-founder of Orq.ai, brought the most operational talk and it was exactly what the room needed. After 21 years of building SaaS and tech solutions, Sohrab has seen every hype cycle. What sets this one apart, he says, is the maturity gap.
He mapped a spectrum that resonated immediately. Most SaaS companies are still at stage one: a support chatbot answering FAQs. A few have moved to stage two: a chat assistant that can interact with APIs and take actions. The real value starts at stage three: fundamentally changing user workflows, using agents to let users skip entire steps. And the frontier (stage four) is where the SaaS UI disappears entirely into the background, with users interacting through ChatGPT, Claude or other assistants via MCP servers and CLIs.
That last point deserves emphasis. Imagine HubSpot; but you never open HubSpot. You ask your AI assistant what you discussed with a client in your last meeting, and it pulls the answer from HubSpot’s backend. The software still owns the domain knowledge, the data, the logic. But it no longer needs a UI. That’s a massive architectural and business model shift.
To get there, Sohrab argued, teams need to treat agents like products with their own lifecycle. Plan, design, experiment, deploy, operate; just like the software development lifecycle, but with different nuances at every step. What does a unit test mean for an agent? How do you do regression testing on something probabilistic? You need cross-functional ownership, engineers alone can’t do it because they’re rarely the domain experts. Product managers and subject matter experts need to be in the loop at every stage.
He introduced a concept I haven’t heard elsewhere: Agent Lifetime Value. Just as a CFO manages a portfolio of capital expenditures, you manage a portfolio of agents. Each one has build costs, ongoing API and inference costs, and (critically) value attribution. Money generated, money saved, time reclaimed. Roll that up, and you have a real-time view of where you’re getting returns and where you should double down or shut down. It’s the kind of operational rigor that separates companies playing with AI from companies running on it.
What this means for SaaS (and what comes next)
If I step back from these five conversations and combine them with what we’re hearing from the sharpest minds in venture capital and the broader market, I see a few things clearly.
First: the role of software is changing from tool to performer. Your software should be designed not to enable a human user to complete a task, but to execute that task itself. Call it AI agents, digital employees, autonomous workflows; the label matters less than the design principle. If you’re still building for a person clicking through screens, you’re building for yesterday.
Second: building and shipping software is being transformed at the same pace. AI-powered engineering is not a nice-to-have; it’s a survival requirement. The teams in a garage with a laptop and an API key are already replicating your product on a $500 token budget. Autonomous task completion is doubling every seven months. Your legacy competitors are the least of your worries.
Third: European SaaS companies have a unique card to play. Data sovereignty, compliance by design, embedded trust; these aren’t bureaucratic obstacles. They’re product differentiators in a world where enterprises are increasingly nervous about where their data ends up. The EU AI Act, whatever you think of regulation, creates a market for companies that take it seriously.
Fourth: transformation is not optional, and it’s not painless. Marili replaced 85% of her engineering team. Julien told the room their current setup won’t last long. Aart showed that the money will shift from the tool to the work the tool performs. Sohrab demonstrated that shipping AI at scale requires entirely new infrastructure and governance. Arjé reminded us that moving fast means nothing if you lose control of your customers’ trust.
The “SaaS is dead” narrative is the kind of reductive take that gets clicks but misleads founders. SaaS is not dead. But SaaS as we knew it (the per-seat, screen-based, human-operated tool) is reaching the end of its dominance. What replaces it is more ambitious: software that doesn’t wait to be used, but goes to work on its own. Software that sells outcomes, not access. Software that earns its keep not by being opened, but by delivering results while nobody’s looking.
The companies that figure this out will build something bigger than what came before. The ones that don’t will wonder what happened.
It’s not a death. It’s a metamorphosis. It’s normal for every industry, dnd it’s happening right now.
