B2B software providers in our space should be trembling in our boots, right?!
Earlier this year, a Cloudflare engineer reimplemented most of the Next.js API surface in about a week. Using AI, he targeted the same test suite and observable behavior, without copying a line of source code.
The reaction in most circles focused on the legal angle: can you add contract terms to prevent this?
But far from forcing us to invent contractual reasons for human and organizational expertise, if anything, AI is making this expertise even more deeply essential by creating significantly higher backend complexity.
What does this reveal about how enterprises scale, and whether their systems are ready for it?
As Bas Van den Berg, VP, Amplify Platform at Axway, puts it: building software is becoming easier, but running it reliably, at scale, is not.
“It is one thing to generate a function or product set with AI, and quite another to maintain it on a continuous basis, with everything that comes along with it: follow-the-sun support, platinum-level SLAs, guaranteed compliance…The appreciation is shifting from the vendor’s ability to create, to their ability to operate — and what they shield customers from over time.”
The headline-grabbing story of AI evolution signals a deeper shift in how systems are built, run, and scaled. Here is what it means for enterprise environments that were never designed for that level of operational pressure.
Is SaaS really dead? Please.
Enterprise software selloffs seemed to paint a sobering picture earlier this year among industry analysts. The Next.js reimplementation was one of many converging factors that amplified existing anxieties across the sector:
- Notably, Claude Code and Cowork made agentic AI feel suddenly practical inside enterprise workflows.
- Most recently, the Fable/Mythos shutdown raised the security stakes: if advanced models can be restricted or disabled overnight, enterprises need continuity and control over how agents access and act across systems.
In other words, the problem is not only whether agents can perform useful work. It is whether enterprises can trust them inside governed, business-critical processes. Even OpenAI says there’s a big difference between people using AI individually and enterprises relying on AI at scale: “We have not yet really seen AI penetrate enterprise business processes,” said Brad Lightcap, OpenAI COO.
As enterprises move quickly to embrace agentic AI, my colleague Mustapha Ezzaime rightfully points out that speed without control creates new risks:
“Autonomous agents can access sensitive systems, trigger actions, and make independent decisions at machine speed. Without centralized governance, this speed quickly turns into security gaps, regulatory exposure, unmanaged costs, and loss of trust.”
This right here may be a significant factor in the skepticism around this so-called SaaSpocalypse. “Software is dead…again…for real this time…maybe,” says one expert.
It’s also worth being precise about what agents are actually replacing. The evidence so far is that agents don’t replace SaaS products; they replace SaaS interfaces. The software still exists. The permissions still exist. The workflows, the data model, the audit trail: all still there. What disappears is the human navigating menus. Agents consume APIs directly, while enterprise systems continue providing the underlying records, workflows, approvals, and controls.
In that sense, the future may look less like SaaS extinction and more like what Michael Carden called “application tourism”: foundation models may dabble in vertical workflows, but the durable value remains in the domain expertise, integrations, and operational context that make enterprise software actually useful. The first casualty of agentic AI may not be SaaS products. It may be SaaS user interfaces.
At the very least, this shift offers an opportunity to truly differentiate between whether software vendors are building it as a genuine operational capability, or as managed hosting with a shiny sales story.
Where enterprise systems create real value
If behavior is observable, implementation parity is now cheap. Features, APIs, frameworks, and UI patterns are all behavioral artifacts. They can all be reproduced.
The Next.js reimplementation wasn’t a novel attack. It was a clean-room reimplementation: observe behavior, reproduce it independently. Samba did this to Windows SMB. MariaDB did it to MySQL. React has a dozen clones. The only thing that changed is that AI compresses the timeline from years to a week, and the cost from millions to a few thousand dollars in compute.
Not everything becomes easier to replicate at the same pace. AI may have commoditized some capabilities, but others will become more valuable as systems grow more complex.
Legal terms are a reasonable, minor response. Commercial licenses can be extended to prohibit AI-assisted reimplementation and behavioral analysis for competitive purposes. That clause helps with customers. It does not stop a competitor from observing your product’s behavior and building their own version.
Data, operations, compliance, and ecosystem relationships are the remaining durable advantages, and they compound.
The assumption most software companies are operating on is that their competitive advantage comes from the feature set. Build differentiated capabilities, ship them ahead of competitors, maintain the gap.
That model is eroding, and not gradually. If behavioral parity can be achieved in a week, the time advantage that justifies multi-quarter roadmaps shrinks to near-zero. Feature moats built on implementation effort are not durable moats anymore.
The companies that navigate the next five years well will be operational capability companies competing on how well they run the system for their customers. (See also: SaaSpocalypse? Or SaaSolidation?)
To be more specific: it’s not operational capability alone but governed operational capability. The biggest enterprise AI announcements lately aren’t about smarter agents—they’re about controls, approvals, auditability, and policy enforcement. What gets AI from pilot to production isn’t a better model. It’s the governance layer that organizations can trust.
Generative AI is great at synthesizing syntax, mimicking public design patterns, and generating boilerplate interfaces. What it lacks is contextual, historical, and runtime reality.
| Domain | AI Commodity | Operational Moat |
| API & data | Feature behavior and API surface | Production telemetry accumulated at scale |
| Infrastructure | UI patterns and SDKs | Deployment automation and SLO history |
| Governance | Documentation and code | Compliance frameworks / certifications and audit trails |
| Operations | Standard integration and orchestration logic | Operational playbooks built from real failures |
Consider managed cloud operators:
- more deployment data
- more failure patterns observed
- better upgrade automation
- tighter compliance guarantees
A SaaS competitor can reimplement the API; they cannot reimplement the operational intelligence a trusted partner accumulates over years of running your operations in a managed cloud at scale.
There’s one more category worth calling out here: context. Not proprietary code, not even proprietary data—context. Business processes, approval chains, organizational structure, workflow history, customer-specific policies, integration relationships. AI can reproduce software behavior; it struggles to reproduce years of accumulated organizational context.
This is where the conversation shifts from theory to execution. Running systems at scale means orchestrating multiple integration patterns, governing interactions across environments, and maintaining control as complexity grows.
What actually differentiates at scale
In practice, the difference shows up not in feature sets, but in how systems behave under real operating conditions as complexity grows, usage increases, and risk accumulates.
Three areas matter most:
- Operational intelligence: production telemetry, deployment history, and operational experience that make systems easier to run as they scale, with faster upgrades, fewer incidents, and more predictable performance under changing load.
- AI-native operations: using AI to detect anomalies, predict upgrade risks, optimize performance and cost, and automate remediation before issues impact users.
- Compliance as infrastructure: certifications, controls, and governance capabilities that reduce audit burden, lower operational risk, and help regulated organizations scale without rebuilding compliance from scratch.
These advantages compound over time. A competitor may be able to reproduce functionality, but not years of operational learning, compliance investment, and real-world execution.
From building software to running systems: the new rules of scale
As AI compresses the cost of implementation, value shifts toward outcomes. Organizations are increasingly evaluating technology not by the features it offers, but by its ability to deliver reliable results under real-world conditions and growing complexity.
AI commoditizes implementation. But enterprise-scale software is genuinely harder to replace at the operational layer.
The Cloudflare engineer who reimplemented Next.js in a week did everyone in enterprise software a favor: they exposed just how quickly implementation advantage is disappearing.
They also made something else clear: the real challenge has become running systems reliably, continuously, and at scale as complexity grows and expectations shift from features to outcomes.
This is why organizations are shifting toward platforms and partners that can provide both technology and operational continuity: because scalability is maintained every day. How your systems are run matters as much as what they do. Make sure you have the operational foundation in place to scale before growth exposes the cracks.
Explore how organizations are rethinking integration and scalability in an AI-driven world