June 25, 2026

Pegasystems CEO: Vendors Releasing Unpredictable AI Agents Shows a ‘Philosophy of Madness’

The CX and enterprise tech space more broadly is currently gripped by an obsession with scale. Major vendors are envisioning a future populated by tens of thousands of autonomous digital workers. Yet behind the promotional push for independent AI agents, a more practical debate is emerging around the baseline rules of corporate governance.

Alan Trefler, the chief executive of Pegasystems, recently challenged this uncritical race toward total automation. He suggested that the industry’s current path introduces significant operational liabilities.

In an interview with Computer Weekly, Trefler argued that enterprises are inadvertently structuring long-term architectural problems by allowing non-deterministic software agents to execute critical network decisions. While platform providers are introducing centralised monitoring systems to manage these new digital workforces, Trefler suggests that the fundamental unpredictability of large language models makes them poorly suited for solo deployment within core business functions.

Trefler said:

“I think they have already acknowledged they’re going to have thousands of agents running, maybe over 10,000, and I think that philosophy is madness.”

The friction points become highly visible when isolated AI initiatives interact directly with client bases. Unmanaged autonomy can lead to rapid reputational friction. In highly regulated sectors, giving unguided algorithms the authority to interpret rules dynamically creates an unacceptable compliance variance. Trefler points out that businesses requiring absolute consistency of process cannot rely on disassociated software entities that might handle clients inconsistently or inadvertently breach compliance standards.

Enterprise Risk and the Shift Toward Hybrid Architecture

Macroeconomic research affirms that this cautionary stance is shifting from an isolated opinion to a broader consensus. Gartner projects that while 40% of business platforms will feature task-specific AI agents by the end of 2026, a quarter of organisations will suffer major business disruptions by 2028. This has been chalked up to inadequate AI governance and flawed guardrail practices. The risk is driven by broken interoperability and a lack of baseline structural boundaries when independent agents trigger multi-system workflows.

To counter these structural vulnerabilities, segments of the market are shifting towards a hybrid architecture. This will balance generative capabilities with deterministic execution. McKinsey’s global corporate tracking reveals that security and operational risks remain the primary barriers to scaling agentic AI. Eighty percent of early adopters are already encountering unexpected or risky agent behaviours. Consequently, leading developers are designing systems where generative models serve exclusively as a semantic interface to parse intent, while leaving the execution of the actual transaction to rigid, pre-audited rules engines.

“You don’t want these disaggregated, disassociated initiatives trying to run important things in the business where it might treat customers differently in ways that it shouldn’t,” Trefler said in the Computer Weekly interview.

This hybrid model addresses the growing financial scrutiny occurring across corporate tech boards. Forrester reports that a quarter of planned AI investment is on track to be deferred due to mounting return-on-investment concerns and escalating API token overheads. The initial phase of unconstrained prototyping is giving way. Increasingly, there is a strict environment where tech spend must be directly tied to measurable gains in operational efficiency rather than pure consumption.

What the Autonomous AI Agents Debate Means for CX and Enterprise IT Leaders

This evolving debate changes how software assets must be evaluated for procurement teams and systems architects. The core criterion for selecting an AI vendor should be the structural boundaries built around those agents. Corporate buyers should evaluate how easily an AI framework integrates with their existing, audited business logic.

IT departments are shifting their resources away from maintaining basic, out-of-the-box integrations. Instead, they are focusing heavily on creating semantic firewalls that prevent runtime models from altering core business parameters. Ultimately, the maturity of an enterprise AI deployment is now revealed by the control the organisation maintains over its outcomes.