The Vendor Landscape: Joule, Sana, Copilot, and Beyond

Deep Dive | ~10 min read
Track: 02 Architecture and Technology
Module: 02.5 Evaluating Agentic HR Platforms

Every enterprise HR leader faces the same question in 2026: which AI vendor actually delivers autonomous, governed agents — and which is dressing up a chatbot in agent clothing?

The answer depends on where a vendor sits in the HR technology stack, what data it can reason over, and how much it can actually do versus how much it can merely suggest. This article maps the current vendor landscape against six evaluation dimensions, identifies five distinct archetypes, and provides a framework for making selection decisions that will hold up over a three-to-five-year planning horizon.

The Five Vendor Archetypes

Before evaluating individual products, it helps to understand the structural categories. Each archetype carries inherent strengths and constraints that no amount of product iteration can fully overcome — because the constraints are architectural, not feature-level.

1. HCM-Embedded Agents

Examples: SAP Joule, Workday Sana, Oracle HCM AI Agents

These are AI capabilities built directly into — or tightly partnered with — a system of record. They benefit from deep integration with a single HCM platform’s data model, workflows, and permissions. The trade-off is that their world ends at the boundary of that system.

SAP Joule functions as an embedded copilot across the SAP ecosystem. It can surface SuccessFactors data, trigger workflows within SAP modules, and answer questions grounded in SAP’s data model. Its strength is native access to transactional HR data for SAP customers. Its limitation is that it cannot reason across non-SAP systems without custom middleware, and its agent capabilities remain tethered to SAP’s process definitions.

Workday Sana represents Workday’s AI agent strategy, built through a partnership with Sana that layers conversational and agentic capabilities on top of Workday’s platform. Agents can operate within Workday’s Business Process Framework (BPF) and access Workday data natively. However, autonomy is constrained by BPF step definitions — an agent can execute a Workday business process, but it cannot orchestrate a workflow that spans Workday, an external ATS, and a learning platform in a single reasoning chain. The data boundary is Workday’s data boundary.

Oracle HCM AI Agents follow a similar pattern: strong within the Oracle Cloud HCM footprint, limited outside it. Oracle’s advantage is breadth across its own application suite (ERP, SCM, HCM), but cross-vendor orchestration requires significant custom integration work.

Structural limitation: HCM-embedded agents inherit the data gravity of their parent system. In enterprises running heterogeneous HR technology stacks — which is most enterprises — these agents see only a fraction of the relevant context.

2. Horizontal AI Copilots

Examples: Microsoft 365 Copilot, Google Duet AI (Gemini for Workspace)

Horizontal copilots bring AI to the productivity layer: email, documents, chat, calendars. Their reach is enormous — hundreds of millions of users — and their delivery surface is wherever employees already work.

Microsoft 365 Copilot can summarize meetings, draft communications, analyze spreadsheets, and increasingly interact with third-party data through plugins and Microsoft Graph connectors. For HR, this means a manager can ask Copilot to draft a performance summary or pull together team utilization data. The limitation is domain depth: Copilot has no native understanding of skills taxonomies, internal mobility patterns, succession risk, or workforce planning models. It operates on documents and data surfaces, not on talent intelligence.

Google Duet AI (Gemini for Workspace) offers comparable capabilities within the Google ecosystem. The same strengths apply — broad reach, natural integration into daily workflows — and the same limitations hold. Neither platform maintains a skills ontology, models career pathways, or understands organizational talent dynamics at a structural level.

Structural limitation: Horizontal copilots are wide but shallow in HR. They can assist with HR-adjacent tasks (drafting, summarizing, scheduling) but lack the domain-specific context architecture required for genuine talent decisions.

3. Talent Intelligence Platforms

Examples: Eightfold, Beamery, Phenom

Talent intelligence platforms built their value on skills inference, talent matching, and labor market analytics. They brought machine learning to HR before the current agentic wave, and they hold meaningful advantages in specific domains.

Eightfold has invested heavily in skills ontology and talent matching. Its Talent Intelligence Platform ingests workforce data and infers skills from resumes, job histories, and behavioral signals. The skills graph is a genuine asset. Where Eightfold faces challenges is in agent orchestration — moving from “here is an insight” to “here is an action, executed with governance, across multiple systems.” The platform was architected for intelligence, not for autonomous multi-step workflows.

Beamery approaches talent through a CRM lens, excelling at candidate relationship management, talent pipeline nurturing, and employer branding analytics. Its strength is the top-of-funnel talent lifecycle. Its agent capabilities are oriented toward recruiting workflows rather than the full employee lifecycle — internal mobility, succession, workforce planning, and skills development sit outside its core architecture.

Phenom delivers talent experience across recruiting, onboarding, and career site personalization. Like Beamery, its center of gravity is the acquisition side of talent management. Phenom’s AI capabilities are meaningful within that scope but do not extend to the cross-lifecycle, multi-system orchestration that defines the agentic HR category.

Structural limitation: Talent intelligence platforms were built to inform decisions, not to execute them. The gap between a recommendation and a governed, auditable action — write-back to an HCM, trigger a learning assignment, update a succession plan — is an architectural gap, not a feature gap.

4. Purpose-Built Agent Platforms

Example: Gloat (with Loomra)

Purpose-built agent platforms are designed from the ground up for autonomous, governed, multi-system HR action. This is the newest archetype and the smallest category, because the requirements are severe: you need cross-system data access, a deep skills and talent context layer, agent orchestration infrastructure, governance architecture, and integration into the flow of work — all at enterprise scale.

Gloat occupies this category with its Loomra AI Context Engine and a fleet of 29 agents organized across four categories. What distinguishes the purpose-built approach is architectural:

• Cross-system reasoning. Loomra’s Knowledge Graph (2.4M entities, 18.7M edges) ingests and reconciles data from Workday, SuccessFactors, Oracle HCM, applicant tracking systems, learning management systems, and other enterprise sources simultaneously. The context layer is not locked to any single system of record.
• Model-agnostic AI. Gloat operates across Anthropic, Google, and IBM WatsonX models — and supports customer-provided models. This prevents vendor lock-in at the model layer, allows optimization by task type, and accommodates enterprise procurement requirements around sovereign AI and data residency.
• Intelligent tooling at scale. Loomra’s 14 Intelligent Tools have processed over 200 million matches. The Personalization Engine draws on 5M+ employee behavioral patterns. The Retrieval and Embedding layer runs six models with 90% accuracy, and the Business Logic Engine applies 24 rule categories at sub-15ms latency.
• HCM write-back. Gloat agents do not just surface recommendations — they execute actions back into systems of record with full audit trails. This is the difference between a system that tells a manager “this employee should be considered for this role” and a system that initiates the transfer process, updates the learning plan, and notifies stakeholders — all within governed guardrails.
• Delivery in the flow of work. Agents are accessible through Microsoft Teams, Slack, Google Chat, and Microsoft 365 Copilot — meeting employees and managers where they already work rather than requiring them to navigate to a separate application.

With 100+ enterprise customers, 5M+ employees served across 112 countries, and 200+ deployments backed by 8+ years of ML research, the platform has the production maturity that distinguishes architectural claims from operational reality.

Structural advantage: Purpose-built agent platforms are not retrofitting agent capabilities onto a system designed for something else. The architecture was designed for cross-system, governed, autonomous action from the start.

5. Point Solution Agents

Examples: Textio (writing optimization), Visier (people analytics), various recruiting automation tools

Point solution agents deliver high-quality AI within a narrow functional scope. Textio improves job posting and feedback language. Visier surfaces workforce analytics and increasingly offers agentic query capabilities over people data. Recruiting automation tools handle interview scheduling, candidate screening, and offer management.

Structural limitation: Point solutions excel within their lane but cannot orchestrate across the talent lifecycle. An enterprise deploying point solutions for each HR function faces integration complexity, governance fragmentation, and the absence of a unified context layer that connects insights across domains.

The Six-Dimension Evaluation Framework

Selecting an agentic HR platform requires evaluating vendors against dimensions that reveal architectural fitness, not just feature checklists. The following six dimensions separate platforms that can deliver sustained value from those that will hit structural ceilings.

Dimension 1: Context Breadth

How much of the enterprise talent picture can the system reason over?

Context breadth measures the range of data sources, entity types, and relationships an agent can access when making decisions. A system limited to a single HCM’s data sees employee records, org charts, and compensation. A system that also ingests ATS data, learning completions, project assignments, skills assessments, and labor market signals can reason about talent holistically.

Dimension 2: Autonomy Spectrum

How much can the agent do without human intervention — and how granularly can you control that boundary?

Autonomy is not binary. The relevant question is whether the platform supports a configurable spectrum: fully assisted (agent suggests, human acts), semi-autonomous (agent acts within pre-approved boundaries), and fully autonomous (agent acts and reports). The governance architecture must allow different autonomy levels for different action types, roles, and risk profiles.

Dimension 3: Skills Intelligence

How deep and dynamic is the skills understanding?

Skills intelligence encompasses taxonomy completeness, inference accuracy, decay modeling, adjacency mapping, and labor market integration. A system that maps 50,000 skills statically is fundamentally different from one that models skills as a living graph with relationships, proficiency levels, and market demand signals.

Dimension 4: Delivery Model

Where do agents meet employees and managers?

Delivery model evaluates whether agents are confined to a proprietary UI or available in the tools people already use — Teams, Slack, Google Chat, email, HRIS portals. Flow-of-work delivery is not a convenience feature; it determines adoption rates and, ultimately, whether the investment generates returns.

Dimension 5: Governance Architecture

How does the system ensure agents act within policy, with transparency, and with accountability?

Governance architecture includes audit trails, explainability, role-based permissions, action approval workflows, bias monitoring, and compliance reporting. In regulated industries and global enterprises, governance is not optional — it is the prerequisite for any autonomy beyond basic assistance.

Dimension 6: Integration Philosophy

Does the platform treat integrations as add-ons or as a core architectural principle?

Integration philosophy distinguishes platforms that bolt on connectors from those that architect for multi-system data ingestion, reconciliation, and write-back as foundational capabilities. The difference becomes apparent when an organization needs an agent to read a requisition from an ATS, match it against skills data from an LMS, check headcount approval in the HCM, and initiate an offer — all in a single workflow.

Vendor Comparison Matrix

Reading the Matrix: Where Each Archetype Wins and Where It Stalls

No vendor is weak across every dimension — if they were, they would not have enterprise customers. The strategic question is which dimensions matter most for your organization and which constraints you can live with.

If your enterprise is standardized on a single HCM and your agent ambitions are limited to optimizing processes within that system, an HCM-embedded agent (Joule or Sana) delivers value with minimal integration work. The ceiling is cross-system orchestration and the autonomy constraints imposed by the parent platform’s process architecture.

If your priority is employee-facing AI in productivity tools and you are not asking agents to make talent decisions, a horizontal copilot (Microsoft 365 Copilot, Google Gemini) meets the need. The ceiling is domain depth — these platforms will not build you a skills-aware internal marketplace or run a succession planning workflow.

If your primary challenge is talent intelligence for acquisition, and your agent requirements are centered on recruiting and candidate management, a talent intelligence platform (Eightfold, Beamery, Phenom) brings genuine depth. The ceiling is lifecycle breadth and the shift from recommendation to governed action.

If you need agents that reason across systems, act autonomously within governed boundaries, and operate across the full talent lifecycle, the architecture requirements point to a purpose-built agent platform. Gloat’s positioning here — cross-system context via Loomra, model-agnostic AI, 29 production agents, flow-of-work delivery, and HCM write-back — addresses the dimensions where other archetypes encounter structural limits.

This is not a matter of better engineering. It is a matter of architectural origin. Systems of record tell you what is. Systems of action decide what happens next. The vendor you choose reflects which problem you are solving.

Selection Guidance: Five Questions for Your Evaluation

1. How many HR systems does the agent need to reason across? If the answer is more than one, cross-system context architecture is non-negotiable. Evaluate how the vendor ingests, reconciles, and maintains data from heterogeneous sources.
2. What actions do you need agents to take — and in which systems? Read-only intelligence is valuable but bounded. If you need agents that write back to your HCM, trigger learning assignments, or initiate workforce planning actions, evaluate the vendor’s write-back capabilities and governance model.
3. Where do your employees and managers work? An agent that lives in a proprietary UI competes for attention. An agent embedded in Teams, Slack, or Google Chat meets people where they already are. Ask vendors to demonstrate their delivery surfaces in your environment.
4. What is your model strategy? Single-model dependency creates risk. Evaluate whether the platform supports multiple foundation models and whether you can bring your own model for sensitive workloads.
5. How does the vendor handle governance at the action level? Ask for audit trail demonstrations, not slide decks. Request examples of configurable autonomy — where a manager can approve certain agent actions while others execute automatically within policy bounds.