Council Post: AI-Native Vs. AI-Bolted: The Architectural Divide That Will Restructure Enterprise Software
Rafa Flores, Chief Product Officer at Treasure Data, is redefining market-agnostic SaaS with breakthrough solutions at the edge of data & AIgettyWhen Anthropic filed for its IPO, the headlines focused on va...
Rafa Flores, Chief Product Officer at Treasure Data, is redefining market-agnostic SaaS with breakthrough solutions at the edge of data & AI
getty
When Anthropic filed for its IPO, the headlines focused on valuation. I focused on the verdict. Not a verdict on Anthropic, but a verdict on a decade-long debate that has paralyzed enterprise software: Can you make an existing platform AI-native, or do you have to build from scratch? We now know the answer. Now every SaaS company that spent the last three years bolting AI features onto legacy infrastructure just got very expensive news.
The Shift Under SaaS
The SaaS model was built on a great premise: Rent software, update it continuously and never ask a customer to manage servers. That worked for decades, and I saw firsthand how frictionless deployment was.
But SaaS was designed for a human user. The entire interaction model assumed a person would log in, navigate a dashboard, configure a workflow and make decisions that led to campaigns. The platform’s job was to hold data and respond to commands, not to do the strategizing. The human in charge did all the thinking, but that assumption is now structurally wrong.
The marketer’s most capable colleague today is an AI agent running in the background. This teammate executes decisions at a speed no human workflow can match. Now, we’re seeing platforms designed for human use buckling under the weight of needs for which they were never designed.
The Bolt-On Problem, Not A Bug
When a legacy SaaS vendor adds an “AI layer,” they are adding intelligence on top of a system built to process human instructions. The rub: These data pipelines were designed for batch processing, and the permission models were built for user roles rather than autonomous agents. The event triggers assume deliberate human action as the initiating condition.
You can put a large language model on top of that stack. Vendors do this every day, but they’re creating a sophisticated wrapper around a fundamentally human-paced system. The AI can suggest, summarize and even generate a first draft. What it can’t do, without enormous friction, is execute, learn and adapt inside a loop that closes in milliseconds rather than days. AI-native operates much differently.
The architectural difference between AI-augmented and AI-native is all in the system assumptions. Legacy SaaS assumes a human will return to the interface, and an AI-native platform assumes an agent is always running.
What Anthropic’s IPO Actually Proves
Anthropic’s path to IPO represents something specific: a foundation model moving from research asset to production infrastructure.
For years, enterprise AI adoption was slowed by a legitimate question: Is this production-ready? There were real concerns like unpredictable latency and audit trails that legal would never approve. The “AI-first” pitch was exciting, but the risk to the enterprise made the C-suite hesitate.
Now we’re seeing a full embrace of AI from these decision makers. Anthropic’s model is now embedded across industries with real compliance requirements and real revenue consequences. The Anthropic IPO isn’t a signal that AI is coming, but that it’s arrived, been tested and met the bar.
For the SaaS market, this is a major shakeup. Enterprise buyers are adopting AI-native platforms—and asking whether the platforms they’re already paying for will become AI-native.
My opinion is no, they cannot become AI-native, not without rebuilding from the foundation.
The Honest Math On Legacy Transformation
I deeply respect the teams at legacy SaaS companies trying to execute this transformation. The engineering challenge is steep, but I’ve orchestrated enough platform evolutions to be honest about what the numbers look like.
Rebuilding a data architecture for real-time, agent-ready infrastructure—while maintaining compatibility for existing enterprise contracts—takes years. Every month that timeline extends, the AI-native alternatives accumulate more production usage, more trained models and more customer data in the right shape for agentic workflows. At some point it’s going to be impossible to catch up.
There’s a concept in physics called path dependence: Where you end up isn’t just a function of where you want to go, but of every step you’ve taken. Legacy platforms carry the path dependence of decades of human-first behavior baked into their product. That path doesn’t disappear because the road map says “AI-focused.”
What Enterprise Buyers Need To Ask
If you’re evaluating your martech or enterprise software stack in light of everything happening in the AI market, the questions are about the architecture instead of the feature checklists.
Questions to consider: When you deploy an AI agent on this platform, does the platform know the difference between a human making a decision and an agent making one? How does the system handle real-time data, not yesterday’s data? What happens when the agent needs to act before a human can review? Is the governance model designed for autonomous execution, or does it assume a human will catch errors?
Unfortunately, these questions don’t have comfortable answers for most incumbent vendors. The honest answer requires acknowledging that the platform was built for a world that no longer exists.
The CPO’s Responsibility In This Moment
This is something that doesn’t get said enough in product circles: The leaders building software right now have an obligation to not sell the road map as if it was the product. The gap between “we’re building toward AI-native” and “we ARE AI-native” is where customer trust dies.
Enterprise buyers are making multiyear commitments right now. They’re committing in an environment in which Anthropic’s IPO and the rapid maturation of production AI say that the AI-first moment is here. If your platform isn’t ready to meet that moment architecturally, the most valuable thing you can offer your customers is honesty about the timeline and a clear migration path toward AI infrastructure. In other words: Don’t give them a road map slide.
The vendors who earn trust in this window will define the next decade of enterprise software. The ones who oversell their AI transformation will lose deals to platforms that were built for this moment. While some martech vendors will scramble to bolt AI onto legacy platforms, the ones who will likely conquer have been purpose-built for the AI-first era.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?