Redefining Digital

For the Autonomous Enterprise, a Chatham House Summary

As digital transformation reaches a new inflection point with the mainstreaming of predictive, generative and now agentic AI, enterprises are reassessing both their past progress as well as future direction. We’ve already invested significantly in digitizing operations, modernizing infrastructure, and experimenting with emerging technologies – and yet the journey has only really begun. The definition of “digital” is also evolving - from discrete, tech-led initiatives to embedded strategies that are inseparable from core business and growth priorities. And this redefinition of Digital now means managing significantly higher complexity, balancing incredible speed and agility with stability and leverage, integrating sustainability, ethics and responsible AI governance.

Against this backdrop, a subset of Executive Technology Board members convened at a historic winery in Champagne, France this summer. As a group we reflected upon our learnings across industries and geographies, and worked to capture the collective intelligence across our real world experiences. This document is produced under Chatham House rules from that one day conversation.

The Autonomous Enterprise

Enterprise digital strategy is undergoing a decisive shift: from legacy IT modernization to building AI-native, agentic operating models - fundamentally redefining how work gets done in an enterprise. This “autonomous enterprise” is defined by software agents acting on behalf of teams, orchestrating workflows, and driving value creation across functions.

A foundational framework for building multi-agent ecosystems can rest on three principles: interoperability, decision-making transparency, and cost transparency, backed by defined internal guardrails. Organizations must now wrestle with new governance questions—especially around the full agent lifecycle: entitlements, supervision, certification, and decommissioning. This includes defining who grants access to agent capabilities, how agents are monitored and aligned with business objectives, and how obsolete or underperforming agents are retired securely and ethically - particularly as agent behavior increasingly reflects proprietary process logic rather than static code.

To prepare for this transformation, companies are beginning to model agent/human ratios and reframe workforce cost structures. Many are targeting a shift from a 1:10 human-to-agent ratio or more, aiming to reduce white-collar footprints while deploying 20–25 enterprise-wide agents to replace hundreds of micro-bots.

Agent-Centric Architectures

Core transaction platforms are also being “hollowed out.” Instead of being reengineered, they are reduced to systems of record while orchestration logic is offloaded to AI agents. As one board member said, “ERP used to be the heart of the enterprise — now it’s just a ledger. The real action is happening outside.” Another added, “We’re not modernizing ERP; we’re neutralizing it — stripping it to essentials and externalizing intelligence.”

The architectural implications are significant. Orchestrators—rather than core platforms—are becoming the new control surface: “I want to route around ERP, not through it — orchestration agents are our new controllers.” One model envisions layers of agents: operational agents that execute tasks, SME agents with domain expertise, orchestration agents for process coordination, and audit agents that monitor compliance. A thin ERP core is complemented by modular agents for HR, CRM, and supply chain. Instead of customizing ERP, the goal is to keep it standard and lean, while deploying agents for bespoke outcomes.

Architectures must also manage the lifecycle of agents: registration, authorization, training, and decommissioning. Questions arise about IP ownership, especially when agent logic is encoded in prompts or outcomes, not traditional code. Enterprises must define how agents are certified, governed, and integrated with security and compliance workflows. “If you’re not accountable for security, but you’re responsible for the agent — you’re in trouble.”

The new Agentic Stack

To succeed in an agentic future, enterprises must rethink the foundations of their platforms—moving from monolithic stacks to modular, composable, and orchestration-ready environments. This shift is also driven by growing concerns about platform security and control. SaaS models, while once accelerants, are increasingly questioned due to rising costs, limited customization, and potential vulnerabilities—especially as many open-source models hosted on platforms like Hugging Face carry inherent security risks.

Enterprises are also reassessing their cloud strategies, with some repatriating workloads and deploying on-premise GPU clusters to protect intellectual property, improve performance, and reduce cost. One board member noted that colocation yielded a 30% cost reduction even when factoring in the required talent and infrastructure. Others are considering hybrid models where select workloads are returned to private environments while maintaining the flexibility of public cloud when needed. This evolving platform posture reflects a deeper strategic imperative: maximizing autonomy, cost efficiency, and security as enterprises scale AI-driven operations. “We’re breaking the monolith. I don’t want a big rock. I want Lego blocks — composable, modular, and agent-ready.”

Criticality of Design and Guardrails

“Scale as a way to win is pretty much over. It’s about proprietary data and orchestration now.” A core challenge is data proximity. Agents must act near data. Pulling data across layers, systems, or clouds creates latency and cost problems. “LLMs aren’t smart if they’re far from the data. You can’t have latency or token cost kill the agent’s reasoning.” Embedding vector data closer to source or deploying agents near key systems is a necessary shift.

Visibility is equally essential. Enterprises need telemetry on what agents did, what context they acted on, and how confidently they operated. “We’re designing agent observability like we used to do for microservices — tracing, logging, cost per call, failure rates.” This new discipline - Agent Ops - enables responsible scale.

Unchecked, these risks are significant. “The agent economy will collapse if we don’t fix observability and data proximity.” A reliable control plane for orchestration is essential. LLMs hallucinate; agents need runtime guardrails and visibility into business state, and costs can compound if not tightly managed. Enterprises are investing in a Agent Ops governance layer that sits atop orchestration logic, ensuring policy enforcement and accountability.

Infrastructure for Scale

Enterprise infrastructure must now evolve to support persistent, intelligent agents that are always-on, secure, and cost-efficient. And current pricing models don’t help: “Cloud cost models weren’t built for continuous agent activity. We’re now paying for idle thinking.” Emerging trends—including GPU repatriation, colocation, and hybrid infrastructure strategies—are redefining cost-performance tradeoffs.

Several enterprises are shifting away from public cloud for select workloads, citing 30% or greater cost savings through colocation and self-managed GPU clusters. These shifts also reflect growing concerns over IP protection, sustainability, and performance under AI-intensive loads. Increasingly more and more infrastructure leaders are rethinking infrastructure design - compute placement, network proximity to data, and sustainable architecture principles to better align with agentic operating models.

Rethinking Operating Models

The best platform teams increasingly operate as outcome-bound P&Ls, focusing on lifecycle governance including the sunset of platform capabilities as the move to agentic gets underway. This shift introduces shared services that often challenge the vertical execution of product-aligned teams, requiring careful coordination and modular architecture to maintain agility and accountability.

Many are using two-in-a-box leadership (business and tech paired equally), and aiming for 80% autonomy in platform operation. Product planning is managed through epics, and each feature must be released within a three-month cycle. PTR (Persist, Transform, Retire) is emerging as a common method to guide platform lifecycle decisions.

Agentic architectures prefer, if not require, horizontal integration across vertical silos in order to drive end to end outcome assurance. This spans from core capabilities to customer experience orchestration, with agents acting across silos. “Every function thinks in isolation. It’s time to bring horizontal back.”

Agentic systems are already proving valuable. Some companies have reduced factory onboarding from months to weeks using training agents. Others are deploying agents in customer service, account management, and even product R&D. Overall there is recognition this is “here and now” but requires a new approach and design. “This isn’t prompt engineering anymore. It’s systems design — identity, memory, autonomy, and oversight.”

Future of Workforce

Workforce dynamics are shifting as enterprises model future agent-to-human ratios. Forecasts range from one-to-one (each employee with a digital twin) to one-to-many (20–30 enterprise agents operating across functions). One member recounted a recent meeting of their firms Board of Directors, where he was asked how many employees and how many agents will the firm employ in five years.

These changes are prompting reevaluation of job families, career paths, and skill requirements. Many functions - from contact centers to claims operations - are seeing dramatic reductions in headcount. Equally many others are pivoting toward advisor roles supported by AI - changing the very nature of how work is done.

The role of the CIO/CHRO is evolving to include responsibility for non-human agents. Lifecycle management, credentials and entitlements, and performance oversight and commissioning / termination extend to digital coworkers. Members debated whether a new function—business architecture—should govern these transformations, bridging business and technology. “Every company is now a tech company. We’re heading toward a world of fewer workers and more agents — and that requires a new organizational blueprint.”

These architectural shifts are deeply organizational. Reintroducing a business architecture function to govern end-to-end agentic flows and bridge technical and business governance layers may be necessary. Equally important is managing sequencing: As one board member remarked, “Reduce people first, then give them tools — not the other way around.” This sequencing drove better adoption and less internal resistance. And investing holistically in team composition and driving change and adoption is critical “We hired a few ‘social butterflies’ — they didn’t ship code but helped us attract real talent.”

Finally, a larger question looms: who benefits from these changes, and who is excluded? Concerns abound around marginalization - stemming from unequal access to advanced technologies and training, particularly for those already on the socioeconomic margins. Generational anxieties are surfacing, too - especially among younger professionals questioning their place in a rapidly transforming workforce.

As the agentic shift accelerates, these sociological dynamics must be acknowledged and addressed as part of responsible enterprise leadership. As enterprises push toward the autonomous future, the cornerstones must remain transparency, trust, and governance.

An AI-generated podcast is also available for alternate consumption, with large language models picking what to focus on and how to express the takeaways from this summary.

You can listen to it here.

Executive Technology Board (c)