As AI agents reshape enterprise software, a deeper question emerges: where will value truly live inside the evolving AI enterprise stack?
Over the past year, most conversations about AI in enterprise software have focused on models, agents, and capabilities.
The more I study the space, the more I find myself thinking about architecture — specifically, where value may ultimately live inside the enterprise AI stack.
In a previous post, I explored why not all SaaS moats are created equal in the agent era. Some companies derive defensibility from the data they hold. Others from the workflows they govern. Others still from the ecosystems they’ve built. And in some cases, the moat has historically been the interface itself.
However, as technology leaders continue thinking about the implications of AI agents, another idea has been forming.
Perhaps the question isn’t only how strong a company’s moat is. It’s also where that company sits in the emerging AI stack.
Because the architecture of enterprise software increasingly appears to be organizing into distinct layers — each playing a different role in how value is created for individuals and organizations alike.
At the foundation sits the layer receiving the most attention today: AI intelligence itself. Large language models. Multimodal reasoning systems. The massive compute infrastructure required to train and operate them.
These systems are extraordinary. They can synthesize information, generate software, analyze complex documents, and interact conversationally with users in ways that were difficult to imagine only a few years ago.
As the ecosystem matures, we’re all noticing that access to intelligence is expanding. Organizations can increasingly access multiple AI models through cloud environments and development platforms, allowing them to experiment and adapt as the technology evolves. In many cases, intelligence is becoming something enterprises can integrate rather than build themselves.
That doesn’t make the intelligence layer less important. However, it does suggest that the companies creating the models may not be the only ones capturing long-term economic value.
Above Intelligence sits a layer that receives far less public attention but may ultimately prove just as important, the control layer. This is where organizations govern how AI actually operates inside the enterprise. AI systems can generate recommendations, insights, or actions. But those actions still need to be coordinated across real business systems, executed within defined permissions, and recorded in ways that meet operational and regulatory requirements.
The control layer provides the structure that makes this possible. It orchestrates workflows, manages permissions, enforces policies, and coordinates activity across multiple systems. In many ways, it acts as the operating environment for enterprise AI, ensuring that intelligent agents operate within the guardrails organizations require.
As AI agents move from experimentation into real operational environments, this layer becomes increasingly important.
Enterprises don’t just need intelligence. They need governed intelligence.
Finally, there is the layer where business value is ultimately realized, the domain execution layer These are the systems where the operational work of the enterprise actually happens. Customer relationships are managed here. Financial transactions are processed here. Supply chains are coordinated here. Employees are paid here. IT services are run here.
When an AI agent resolves a service request, processes a refund, approves an expense report, or schedules a technician, that action ultimately happens inside one of these operational systems.
This is where AI moves from generating insight to executing work. And enterprises don’t invest in software simply because it produces insights. They invest in software because it enables work that affects revenue, cost, and risk.
This three-layer perspective connects directly back to the conversation about SaaS moats. Moats explain defensibility, Layers explain economic role.
A company might have a strong moat but operate in a layer where pricing pressure increases over time. Another company might operate in a strategically powerful layer but struggle to defend its position without a durable moat. Looking at the landscape through both lenses helps clarify where long-term value may accumulate.
Data depth tends to matter most where operational context drives decisions. Workflow complexity becomes particularly important where orchestration and governance determine how systems operate. And ecosystems influence how technologies spread across organizations.
Structural Shift in Enterprise Software: What’s striking right now is how many organizations across the technology industry appear to be converging toward similar architectural ideas.
Across the market we see increasing investment in AI agent platforms, orchestration layers, governance frameworks, and domain-specific AI capabilities.
In other words, many organizations are trying to connect the same three layers I mentioned earlier.
Some will focus on intelligence, Others on orchestration, Others on operational systems where work actually happens.
It’s also possible that the future enterprise stack will not be dominated by a single platform, but rather by a set of coordinated systems operating across different domains of work. I believe that question is still unfolding.
The more I reflect on this transition, the more it seems that the real story may not be that AI replaces enterprise software. Instead, AI may simply rearrange where value lives inside the stack. The companies generating intelligence will remain extraordinarily important. But the systems that govern that intelligence and the systems that apply it where real work happens may ultimately capture just as much — or perhaps even more of the long-term value.
