The architectural assumptions underlying the majority of SaaS products built between 2015 and 2023 are no longer valid. Here is what the research says about what comes next — and how fast.
Not broken. Not deprecated. Still technically functional. But obsolete in the same way that a desktop software product was obsolete the moment SaaS became the dominant delivery model — even if users hadn't noticed yet.
The SaaS era produced a clear set of architectural assumptions: seat-based pricing, human-operated interfaces, workflows designed around human task completion speeds, and integrations built for human-readable data exchange. These assumptions were rational constraints given the technology available. They are no longer the binding constraint.
The software industry is entering a new era that may prove even more disruptive than the SaaS revolution that preceded it. The emergence of generative AI and agentic AI is not just another technology wave — it is a foundational shift redefining what software is, who builds it, who uses it, and how companies operate. McKinsey & Company That is McKinsey's Technology, Media & Telecommunications Practice, not a startup pitch deck.
The companies that understand this earliest will rebuild first. The ones that don't will be rebuilt around.
Every SaaS product built before 2022 was designed around a core assumption: a human sits between the software and the outcome. The software presents options. The human makes decisions. The software executes. Repeat.
This assumption is load-bearing. It determines UI design, database schema, pricing model, integration architecture, and the shape of every workflow inside the product. Remove or replace the human from that loop — which agentic AI now allows — and the entire structure needs reconsidering.
The era of "AI as a feature" is rapidly giving way to native-AI. BetterCloud According to the May 2025 Bond Capital report, the era of the SaaS point solution is approaching its end. BetterCloud
This is not a UI refresh problem. It is not solved by adding a Copilot button to an existing dashboard. The AI revolution is not about bolting new features onto old systems. It is about a fundamental architectural shift. To truly unlock the power of AI, enterprises must move from AI-enhanced to AI-native systems — rebuilding the technology stack from the ground up, with AI at its core. Correct Context
McKinsey's research identifies three distinct models emerging as successors to traditional SaaS architecture. Understanding them is essential to understanding which products survive the transition and which get rebuilt from scratch.
Model 1 — Agent-Augmented SaaSAgents automate the repetitive tasks of human users, acting as software users themselves rather than replacing the software. The SaaS product remains the system of record. The human workflow is compressed but not eliminated. This is the near-term path for incumbents with large installed bases.
Model 2 — Agent-Centric ArchitectureA human primarily interacts with a single agent interface. That agent, in turn, interfaces with multiple back-end agents and APIs directly, without the human ever touching the underlying software. This accelerates the commoditisation of core SaaS elements, with more of the value moving to the agent layer. McKinsey & Company The user experience and the agent's functional intelligence become the product — not the database or the dashboard underneath.
Model 3 — Hybrid Vertical IntelligenceAgents distinguished by deep domain-specific knowledge — a legal agent trained by lawyers, a finance agent trained by analysts. The software and the expertise are inseparable. Vertical AI has the potential to eclipse even the most successful legacy vertical SaaS markets, Bessemer Venture Partners according to Bessemer Venture Partners' State of AI 2025 report. The reason is precise: traditional SaaS failed to solve high-value vertical-specific tasks that were multi-modal or language heavy. Vertical AI is finally meeting these users where they are, with products that feel less like software and more like real leverage. Bessemer Venture Partners
According to Accel's 2025 Globalscape report, AI coding assistant adoption among developers jumped from 36% in 2023 to 90% in 2025. Chargebee The gap between idea and prototype has narrowed to a degree that is structurally changing competitive dynamics.
Consider what this means concretely. A two-person founding team with access to modern AI tooling can now prototype, test, and ship a functional vertical SaaS product in weeks — targeting a workflow that previously required months of engineering and a team of ten. GitHub Copilot users complete 126% more projects per week than manual coders. Index.dev 41% of all code written in 2025 is AI-generated or AI-assisted. Second Talent
The barrier to competitive entry in software has collapsed. The moat that came from simply having shipped a working product — the first-mover advantage of existing code — is effectively gone. What this produces is a market where incumbents with legacy architecture face challengers who can rebuild their core value proposition in a fraction of the time it took to build it originally.
Insight Partners Managing Director Ryan Hinkle framed the stakes directly: "The question is: which companies lose to the kids in a garage rebuilding their product AI-first, and which ones unlock new revenue streams with AI?" Insight Partners
For years, the prevailing assumption in enterprise software was that companies would build differentiated AI solutions internally. In 2024, that assumption held: 47% of AI solutions were built in-house, 53% purchased. In 2025, 76% of AI use cases are purchased rather than built. Menlo Ventures
This inversion is significant. It means enterprises have concluded — through expensive and visible failures — that the competitive advantage lies not in owning model infrastructure, but in deploying the right AI-native architecture across their actual business workflows as fast as possible. The build-it-yourself approach lost to the speed advantage of working with specialist teams who already know the patterns.
Enterprise AI has surged from $1.7B to $37B since 2023, now capturing 6% of the global SaaS market and growing faster than any software category in history. Menlo Ventures The capital is moving toward AI-native solutions at a rate that will force the hand of every incumbent.
There is a countervailing force in this transition that deserves honest treatment. Not all AI-accelerated software development produces better software.
A METR randomised controlled trial published in July 2025 — 16 experienced developers, 246 real tasks across mature open-source codebases — found that when developers use AI tools, they take 19% longer than without. METR The study noted this contradicted both developer self-assessments and expert predictions, which forecast time savings of 20–39%.
More materially, code churn — defined as the percentage of code discarded less than two weeks after being written — is projected to double, creating substantial risks for teams deploying to production environments. DevOps AI-generated code, when handled without architectural discipline, produces what researchers are calling AI-induced technical debt: fast to write, expensive to maintain, and difficult to extend.
This is the critical distinction between AI-accelerated rebuilding done well and the wave of rushed "AI-native" rewrites that will fail quietly in 2026 and 2027. By 2025, 75% of engineers use AI tools, yet most organisations see no measurable performance gains in delivery — what Faros Research calls the "AI Productivity Paradox." Substack
Speed without architectural discipline does not produce better software. It produces more software, faster, with higher maintenance costs and shorter effective lifespan. The rebuild cycle accelerates, not for competitive reasons, but because the code itself degrades faster.
Insight Partners draws a useful distinction between "systems of record" and "systems of action." Systems of record — essentially digitised filing cabinets — are acutely exposed. Systems of action, where knowledge workers cannot do their jobs without the software, are positioned very differently. Insight Partners
The products most likely to survive the next three years without a fundamental rebuild are those that sit on proprietary data moats — where the value is in the accumulated, structured data the product has collected, not in the workflow interface built on top of it. CRMs with years of customer interaction data. Analytics platforms with proprietary benchmarks. Financial tools with historical transaction records. The interface will be rebuilt. The data is the defensible asset.
Products with no data moat — workflow tools, productivity utilities, point solutions built to solve a single human process step — are the most exposed. These are the categories where AI-native challengers can replicate core functionality fastest and where the switching cost for users is lowest.
Avenir's January 2026 report found that 63% of enterprise buyers expect their existing software vendors to benefit from generative AI, while only 8% expect them to lose. Chargebee Customers prefer evolution over replacement — but that preference is contingent on execution. Incumbents that move quickly to rebuild on AI-native architecture retain that goodwill. Those that ship cosmetic AI features while leaving the underlying architecture unchanged will lose it.
McKinsey's research is precise about timelines: operational shifts could drive a step change in velocity and productivity within a year, while value proposition shifts will likely be felt in one to two years when AI disruption to SaaS business becomes more pronounced. Fundamental shifts to workforce and workplace will likely need two to three years to fully manifest. McKinsey & Company
This is not a ten-year transformation. The products that will define the next era of SaaS are being designed and built now. The architectural decisions being made in 2025 and 2026 will determine competitive positioning for the rest of the decade, for the same reason that the SaaS architecture decisions made between 2008 and 2012 determined the competitive landscape of the 2010s.
The window for rebuilding on the right foundation — before AI-native competitors reach distribution scale and before legacy technical debt compounds to the point of structural disadvantage — is three years, at most. Probably less for the most exposed product categories.
Sources: McKinsey Technology, Media & Telecommunications Practice 2025 · Bessemer Venture Partners State of AI 2025 · Menlo Ventures State of Generative AI in the Enterprise 2025 · Insight Partners 2026 Investor Predictions · Bain & Company Technology Report 2025 · METR Randomised Controlled Trial, July 2025 (arXiv:2507.09089) · Bond Capital AI Report May 2025 (Mary Meeker) · Accel Globalscape Report 2025 · Stack Overflow Developer Survey 2025 (49,000+ respondents) · Faros Research AI Productivity Paradox 2025 · Avenir State of Enterprise Software January 2026
.png)
.png)
.png)