Agentic AI
Global Business Services
Workforce Transformation
How Agentic AI Affects Your Workforce
Understand how Agentic AI affects your workforce when 90% of work is automated and what it's actually like to work in a blended Agent/Human team.
10 Key Takeaways from Hypatos and Chazey’s Webinar on Designing an Agentic GBS
Most agentic AI conversations focus on the technology. Hypatos founder and CEO Uli Erxleben and Chazey Partners Executive Partner Chas Moore used a recent webinar, hosted by Sally Fletcher, to focus on something closer to home: what happens to the people, roles, and organizational design of GBS as agents take over the work. For anyone who missed it, here are the takeaways that matter when building an agentic GBS strategy.
Erxleben claims that within five years, he expects roughly 90% of today's transactional business support work to be carried out by AI agents. Whether that timeline is five years or seven is beside the point; the direction is settled. The organizations that will manage the transition well are the ones thinking it through now, working backwards from that end state rather than reacting to it later.
Moore walked through a set of statistics that don't line up into a single tidy narrative: half of CEOs now say their own job security is tied to their AI strategy's success this year. Separately, 77% of CEOs believe GenAI is simultaneously overhyped in the short term and underhyped in its long-term potential. IBM has tripled entry-level hiring to bring in digitally native talent, even as broader unemployment projections get invoked as cautionary tales. Perhaps most telling: when ServiceNow surveyed workers on their biggest AI fear, the top answer wasn't job loss. It was that AI is making them stupid, by eroding the everyday problem-solving that builds expertise.
Both speakers converged on a counterintuitive point: despite the discussion of job losses, implementing agentic AI is actually very labor-intensive up front. Your most experienced people (the ones holding undocumented, tacit process knowledge) become the critical resource, not a legacy cost. That knowledge has to be codified into clear, ownable, globally standardized processes before it can be handed to an agent. Skip this step, and there's nothing for the AI to learn from. It's also the moment organizations most need to protect against losing that knowledge to retirement before it's captured.
Moore pushed back on business cases that stop at headcount reduction. He pointed to several underused levers that agentic AI unlocks:
● Non-wage savings — e.g., working capital improvements from automated collections and faster order-to-cash cycles.
● True scalability — the same team absorbing significantly more volume, rather than the old model of adding headcount less than proportionally as you scale.
● New revenue paths — mature GBS functions taking on more business units, or even offering services to external clients.
Erxleben added that this scalability is already changing how GBS leaders think about their own function's ambitions; some are actively exploring taking on new scope and external clients now that efficiency makes it viable.
Chazey's readiness framework assesses 12 parameters across four pillars: data, technology, leadership and people, and governance and process (plus client expectations layered on top). Moore's advice on where to focus first: global process documentation and standardization, and the leadership buy-in required to make binding decisions about how processes should run. Without both, agentic transformation stalls regardless of how strong the underlying technology is.
Erxleben flagged a common mistake: organizations default to small, narrow pilot use cases to de-risk their first agentic project. The better approach is to start with the highest-volume, most labor-intensive process, i.e., accounts payable or order-to-cash, because these are usually also the best-documented and most standardized, and because a credible win there proves real business impact rather than just producing a nice demo.
Erxleben was emphatic that agentic systems must be explainable and auditable i.e. no black boxes. In the live demo of Hypatos AI Agent Workforce, this showed up concretely: every agent decision came with a stated rationale, and a global process owner role emerges as essential for defining standards, making governance calls, and preparing the organization for agent-executed work.
Buy for high-volume, mission-critical, compliance-heavy, end-to-end processes —Erxleben's advice was blunt: don't build that yourself, it's too hard to maintain. Build for narrower, business-specific needs that don't carry that same compliance weight. And run both a proof of concept (fast, to understand what the model can do) and a pilot (slower, to test it against real processes and systems) — they answer different questions.
Erxleben's live walkthrough of an accounts payable agent team was the session's clearest illustration of what "working with agents" actually looks like day-to-day. Multiple specialized agents hand off work in sequence, each governed by detailed work instructions and organization-specific knowledge that cannot come from a generic model's training data. When an agent couldn't resolve a tax-coding decision, Erxleben corrected it in plain natural language, and the agent asked clarifying questions before saving the feedback to its knowledge base for future transactions.
The framing that stuck: this is mentoring a junior colleague, not configuring software. No prompt engineering background required, and the interaction doesn't have to happen in English; any language works, and a well-instructed agent will ask for clarification when something is ambiguous.
Asked how agentic AI change management differs from classic change management, Moore's answer reframed the whole exercise. Traditional shared-services change management focuses on winning over the people affected by centralization. With Agentic AI, the provider doing the bulk of the work is the AI itself, so the change effort shifts toward the client relationship, retraining, and a wholesale redesign of the talent system, because roles change from what someone does to why the role exists and what they oversee. Erxleben added that this is not a single-team change; it's a new way of working that touches the entire organization.
The consistent thread across the session: agentic AI's biggest bottleneck isn't the model, it's the organizational groundwork: that means documented processes, clear ownership, codified expertise, and outcome-based use case design. Get that right, start with your biggest and best-documented process rather than your safest one, and treat change management as a redesign of the entire talent system rather than a rollout communication plan. To view the whole webinar, access it here.
Further stories from our blog
