The future of SaaS

How can the AI stack be unbundled?

Agentic AI is changing how work gets done, while headless architecture and natural language processing are changing where work gets done. Treating these as a single phenomenon obscures the question of which risks organizations are ready to take back, and when doing so is worth it.

The reality is: the future of enterprise applications is not all customized or all SaaS. It is about selectively keeping vendor-managed tools where they reduce risk and allowing the enterprise to take control where a custom experience, orchestration layer, or workflow creates a competitive edge.

What role does headless architecture play?

Headless architecture separates the interface (where humans show up to work) from the back end. The “head” comprises the front end, including the screen, the form, the chat window, and the workflow with which users, and increasingly AI agents, interact. The “body” is the back end: the database, the business logic, the controls, the audit trail, and the system of record.

For decades, enterprise applications fused the two; now headless is pulling them apart.

The headless architecture model reflects the fact that no enterprise runs on one vendor’s stack alone, and that the experience layer can live above all of them. Headless unlocks two distinct advantages. The first is that humans do the work they already do, but faster and with less friction. The second is that work humans used to do is no longer done by humans at all. The agent handles it end-to-end, and the human is involved only by exception. 

How are decisions about humans in the loop made?

Organizations struggling with the degree to which work should be fully automated can focus on calibrated autonomy. This involves matching the level of human involvement to the risk, reversibility, and judgment intensity of the work itself. Some decisions warrant a human in every loop, while most do not. And a handful can be handed to agents entirely with humans intervening only when an exception is flagged.

Specifically, work can be allocated as follows:

• Full human oversight: High-risk, high-judgment decisions that require a human in every loop, such as fraud claims, regulated decisions, and actions with material financial exposure.
• Shared oversight: Moderate-risk processes where agents and humans collaborate, such as billing disputes and complex customer escalations.
• Exception-only: Low-risk, high-volume processes handled entirely by agents with humans intervening only on flagged anomalies, such as password resets and routine triage.

What new roles are needed to manage the unbundled enterprise stack?

A converged stack owner is a new kind of leader who will own calibration decisions across an entire value stream. Although in most enterprises this role doesn’t yet exist, it will be a necessary addition so that a senior leader is aligned to oversight of the stack’s risk profile; redesign of human roles around the new automation envelope; budgeting for the orchestration layer; decisions on automation calibration for each value stream; and, most important, cross-functional arbitration when CRM, ERP, and AI teams disagree.