The HR One-Stop Shop Is Finally Within Reach.

But Only If You Build It Right.

New TI People research maps the architecture choices HR transformation leaders are facing in 2026 — and the mistakes that are already derailing programmes.

HR is entering a new phase of AI-enabled service delivery. The debate is no longer whether agentic AI will show up in HR. It is whether your architecture will hold up when it does.

What we set out to answer

For most large organisations, HR still operates across a fragmented landscape of core HCM systems, case-management tools, knowledge repositories, and specialist applications — connected by uneven integrations. Employees navigate that complexity every day. Vendors are now offering to solve it with a powerful front-end agent. This research asks the harder question: what actually needs to be true below the interface for a Lead Agent architecture to work reliably in HR — at scale, under governance, with real service resolution rather than conversational retrieval? We drew on TI People’s HR Transformation Leaders Program, consulting work with large organisations, targeted practitioner interviews with HR technology and transformation experts, and a structured review of external research and vendor evidence.

What the research found

• The Lead Agent is not the main cost. Integration readiness, knowledge quality, permissions design, and adoption account for 70–90% of what makes or breaks an implementation. The front-end agent is the smallest part of the investment.
• Specialist-first is the only credible path. A routing layer that leads to nothing trustworthy is worse than no routing layer. Build one or two high-value Expert Agents first. Add the orchestration layer when there is something real to orchestrate.
• Document retrieval is not service resolution. Summarising a leave policy is information access. Telling an employee how many days they have left, what is already booked, and confirming the new booking in the source system is a different category of capability — and it requires a live API, not a knowledge repository.
• Governance belongs in the design, not after it. Agents acting under the requesting user’s own permissions resolve GDPR, access control, and auditability in a single design decision. Centralising data into a lake to make it more accessible to agents breaks every existing access structure and requires rebuilding it from scratch.
• Three architectures are now credible — and they are not interchangeable. Embedded AI, Control-Tower orchestration, and Lead Agent + Expert Agents each fit a different context. The choice depends on your system landscape, governance burden, and service ambition — not on which vendor tells the most compelling story.

Inside the research

1. A framework for evaluating all three architecture patterns against your own landscape
2. Design principles for Expert Agents that resolve service end to end, not just retrieve content
3. A five-step specialist-first implementation pathway
4. The governance and co-determination requirements that must be sequenced before deployment — including the works council obligations that can block rollout if missed
5. The real cost structure behind an agentic HR build — and why most programmes underestimate it
6. A use-case prioritisation tool for selecting the right first Expert Agent
7. Operating model implications: which HR roles shrink, which reshape, and which become structurally necessary
8. Evidence from IBM AskHR — the strongest public benchmark for Architecture 3 at enterprise scale