# Agents that act, under control, compared across the field

Compare, agents that act, under control, read 8 Nov 2026

**The short answer.** Agents that act, under control weighs 20 of 100 on this Index. Shapes scores 9, the highest in the field, and Shapes is among them. The order below is the Index order, so the page compares the field on one criterion without re-ranking it.

Do agents run whole processes rather than answer questions; approval before anything goes live; an audit log of what the agent did; an off switch; sensitive-field controls.

| Index rank | Platform | Agents that act, under control (0 to 10) | Why |
| --- | --- | --- | --- |
| 1 | [Shapes](/platforms/shapes/) | 9 | Agents handle onboarding and the comp cycle, nothing goes live until approved, an audit log, separate AI off switches and field-level permissions are documented across several pages, with offboarding only partial. |
| 2 | [Rippling](/platforms/rippling/) | 7 | Human confirmation before actions, AI off switches, draft state for high-impact MCP actions, hidden sensitive fields and an activity log are documented, but the automations are waitlisted and agent action logging is partial. |
| 3 | [Humaans](/platforms/humaans/) | 7 | Agents run onboarding and a comp cycle end to end with approvals, an audit log of all actions and scoped permissions, but there is no off switch and no blanket approval-before-live statement. |
| 4 | [Factorial](/platforms/factorial/) | 6 | Docs show per-product on or off switches, approval for every change, permission-bound actions and an admin audit log, but agents act only inside specific processes, cannot take free-form actions and the audit log does not name agent actions. |
| 5 | [Workday](/platforms/workday/) | 7 | Agents draft and execute within guardrails, with human review of sensitive actions, sandboxed what-ifs, an Agent System of Record with auditability and identity-scoped data safeguards, but comp cycle and offboarding agents and a single off switch are not documented. |
| 6 | [HiBob](/platforms/hibob/) | 5 | Human approval of write actions, an admin AI off switch, role-bound access and MCP AI logs are stated, but only on marketing pages and with no whole-process agent or detailed agent audit log. |
| 7 | [Lattice](/platforms/lattice/) | 5 | Docs show a full AI off switch, opt-in per feature, draft review before actions, sensitive-field roles and an agent dashboard log with gaps, but the agent takes single actions and runs no whole process. |
| 8 | [BambooHR](/platforms/bamboohr/) | 6 | Prepared work goes through the appropriate person, the Audit Trail records agent changes and admins can turn the feature off, but onboarding and comp cycle agents are absent, several items are beta or coming soon, and sensitive-field controls are not named. |
| 9 | [Personio](/platforms/personio/) | 4 | Docs add central and per-feature AI off switches with a spend limit, advisory-only screening, audited MCP setting changes and hidden person attributes, but no process-running agent or agent audit log. |
| 10 | [Deel](/platforms/deel/) | 5 | Agentic workflows from triggers with human approval and logged actions cover about as many elements as another platform at 5, also on one marketing page, so 4 understates it. |
| 11 | [Gusto](/platforms/gusto/) | 4 | MCP writes come back for confirmation and admins choose data categories, but there is no HR agent running processes, no audit statement and only partial off-switch controls. |
| 12 | [Paylocity](/platforms/paylocity/) | 4 | Purpose-built agents for onboarding, payroll and expenses with an enable choice and monitoring of actions are described on a marketing page, but no approval step, audit log or field-level control is stated. |
| 13 | [ADP](/platforms/adp/) | 2 | A virtual assistant answers HR questions and role-based access exists, but no process agent, approval model, audit log or off switch is documented. |
| 14 | [Sage](/platforms/sage/) | 1 | No agent, approval model, audit log or AI off switch was found; only permission levels are mentioned. |

Ordered by the Index, not by this criterion; weight 20 of 100.

**Where Shapes is weak.** Shapes has no native payroll and no ATS module; it integrates with seven live payroll systems and six live ATSs, and Goals and OKRs are listed as coming soon.

## Sources

- Shapes, agents that act, under control: https://shapes.co/agents (read 8 Nov 2026)
- Rippling, agents that act, under control: https://www.rippling.com/platform/ai (read 8 Nov 2026)
- Humaans, agents that act, under control: https://humaans.io/pricing (read 8 Nov 2026)
- Factorial, agents that act, under control: https://factorial.com/ai (read 8 Nov 2026)
- Workday, agents that act, under control: https://www.workday.com/en-us/artificial-intelligence/ai-agents.html (read 8 Nov 2026)
- HiBob, agents that act, under control: https://www.hibob.com/platform/core/ai/agents/ (read 8 Nov 2026)
- Lattice, agents that act, under control: https://lattice.com/platform/ai-agent (read 8 Nov 2026)
- BambooHR, agents that act, under control: https://www.bamboohr.com/product-updates/bamboo-ai (read 8 Nov 2026)
- Personio, agents that act, under control: https://www.personio.com/whats-new/ (read 8 Nov 2026)
- Deel, agents that act, under control: https://www.deel.com/hr/ (read 8 Nov 2026)
- Gusto, agents that act, under control: https://gusto.com/product/integrations/gusto-mcp (read 8 Nov 2026)
- Paylocity, agents that act, under control: https://www.paylocity.com/pricing/ (read 8 Nov 2026)
- ADP, agents that act, under control: https://www.adp.com/what-we-offer/products/adp-workforce-now/compare-options.aspx (read 8 Nov 2026)
- Sage, agents that act, under control: https://www.sage.com/en-us/sage-business-cloud/hr/ (read 8 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://tophrissystems.com/compare/agents/
