Skip to content
Best AI Agents for HR

Agents, rank 1 of 14, read 8 Nov 2026

Shapes's agents: onboarding, the comp cycle and employee questions, with every control published

Read the matrixRead the method

The matrix row is the fullest on the card: yes on ten of twelve, partial on offboarding (workflows run for leavers; no agent named) and on governance (ISO 42001 in progress). The approval sentence is on the agents page; the audit log, the switches and the field-level permissions are dated release notes from October and November 2026, which is why they score as facts rather than claims.

The one thing the row does not do is the one thing the weakness line names: the official MCP server at mcp.shapes.co exposes 35 read-only tools over OAuth 2.0 with PKCE and an official Claude connector, and an outside agent cannot write through it. A company that runs its AI stack outside the HRIS should read the HiBob, Humaans and Workday pages for the writing servers and this page for the controls.

Acts on
onboarding a hiring wave: yes; a compensation cycle: yes; employee questions from policy: yes
Approval model
yes: 'Nothing goes live until you say so.' Directory built with an agent writes nothing until Approve.
Log
yes: 'See who did what in the audit log' release note.
Off switch
yes: Separate switches for AI chat, AI-built Vibes and MCP access, on by default.
Sensitive fields
yes: Field-level permissions: edit, view or hidden per permission level (help center), and a release note on restricting fields.
MCP
live for all customers, read: mcp.shapes.co: 35 read-only tools, OAuth 2.0 with PKCE, official Claude connector.
AI governance
partial: ISO 42001 in progress on the security page.

The 12 capabilities for Shapes, with the evidence

Read 8 Nov 2026. A cell says "none found" about the pages read on that date, never about the product.
CapabilityFoundEvidenceSource
Answers employee questions from policyyesAI chat open to everyone reads the Knowledge Base; agents 'know your org, your policies and your people'.shapes.co/releases/ai-chat-now-reads-the-help-ce
Drafts changesyes'Ask for a change, they draft it, you apply it.'shapes.co/agents
Runs onboarding end to endyesOnboarding named as a process agents handle; the Workflows agent drafts the flow and runs on new hires.shapes.co/
Runs a compensation cycleyes'Onboarding, comp cycle, employee questions' named on the home page; Compensation app with cycles.shapes.co/
Runs offboardingpartialWorkflows run for leavers (release note); offboarding is not named as an agent process.shapes.co/releases/run-workflows-for-new-starter
Acts in other systems (Slack, Teams, email, ATS, payroll)yesWorkflow steps include emails, Slack messages and webhooks; agents open next to the work.shapes.co/releases/workflows-2-0-now-live-for-ev
Approval before anything goes liveyes'Nothing goes live until you say so.' Directory built with an agent writes nothing until Approve.shapes.co/agents
Audit log of agent actionsyes'See who did what in the audit log' release note.shapes.co/releases/see-who-did-what-in-the-audit
Off switch for AI featuresyesSeparate switches for AI chat, AI-built Vibes and MCP access, on by default.shapes.co/releases/turn-ai-features-on-or-off
Sensitive-field controlsyesField-level permissions: edit, view or hidden per permission level (help center), and a release note on restricting fields.help.shapes.co/en/articles/9569339-permission-le
Official MCP server or connector, with access levelyesmcp.shapes.co: 35 read-only tools, OAuth 2.0 with PKCE, official Claude connector.mcp.shapes.co/
AI governance certification held or in progress (ISO 42001)partialISO 42001 in progress on the security page.shapes.co/security-and-compliance

In the Agent Diary

The seven criteria for Shapes

Weighted total 8.6, rank 1 of 14 on Best AI Agents for HR.
CriterionWeightScoreWhy
Build without the vendor159Dashboards, tools and workflows described in plain language on live permission-aware data, a community hub of named built views, and documented Workflows agent, any-field and date triggers, conditional tasks and task views, with only a template gallery missing.
Agents that act, under control309Agents 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.
Openness to the AI stack258An official read-only MCP server and Claude connector with 35 tools, OAuth 2.0, admin off switch, API on Core, webhooks on Plus and a long integrations list are documented, but MCP cannot write.
Time to value58Days-not-weeks claim, step-by-step guide, spreadsheet or agent directory build, guided implementation and sandbox on Plus are documented across two sources, but no fee amount.
Pricing transparency58Per-seat prices ($8 and $12 billed annually), plan contents, seat counting, cancellation terms and in-product billing are documented, but add-ons are talk-to-sales and the minimum amount is not stated.
Fit to the organization's shape58Custom fields and tables, permission roles with filters, unlimited entities on Plus, workflows assigning policies per office and role, and per-policy approvers are documented with two sources, with plan limits and no currency or language detail.
Trust159SOC 2 Type II, ISO 27001, GDPR, ISO 42001 in progress, audit log, field-level permissions, backups, export and 60 day deletion or return, across several pages; minor gap on some cert detail.

Breadth, reported and not weighted

Modules
9: People Directory, Time Off, Workflows, Attendance, Surveys, Docs and eSign, Performance and Growth, People Analytics, Compensation
Native payroll
none; integrates with ADP, Moorepay, PayFit, Sage, Staffology, Papaya Global, Deel, Remote, Bright
ATS
none; integrates with Greenhouse, Lever, Ashby, Workable, Teamtailor, JazzHR, Spark Hire
IT and devices
none; identity via Okta, OneLogin, JumpCloud, Microsoft Entra
Integrations
35 named on the home page wall
Countries
one country or ten (About page); UK, Israel, EU and US payroll integrations

Where it is strongest, and where it is weak

Building the tool, view or workflow you need by describing it, and handing whole processes to agents that do nothing until you approve.

Shapes's MCP server is documented as read access on the date read; write actions run inside Shapes, not through the connector.

Best for

A People team that wants agents to run onboarding and the comp cycle under an approval it can see, with the switches and the log published, and its outside AI tools reading rather than writing.

Sources

  1. Shapes: shapes.co/vibe-views, read 10 Nov 2026
  2. Shapes: shapes.co/releases/workflows-2-0-now-live-for-everyone, read 10 Nov 2026
  3. Shapes: shapes.co/agents, read 10 Nov 2026
  4. Shapes: shapes.co/releases/turn-ai-features-on-or-off, read 10 Nov 2026
  5. Shapes: mcp.shapes.co/, read 10 Nov 2026
  6. Shapes: shapes.co/pricing, read 10 Nov 2026
  7. Shapes: shapes.co/, read 10 Nov 2026
  8. Shapes: shapes.co/releases/build-your-directory-with-an-agent, read 10 Nov 2026
  9. Shapes: help.shapes.co/en/articles/9569350-shapes-pricing, read 10 Nov 2026
  10. Shapes: shapes.co/releases/workflows-assign-time-and-payroll-policies, read 10 Nov 2026
  11. Shapes: shapes.co/security-and-compliance, read 10 Nov 2026
  12. Shapes: shapes.co/legal/dpa, read 10 Nov 2026
  13. Shapes: shapes.co/releases/ai-chat-now-reads-the-help-center, read 8 Nov 2026
  14. Shapes: shapes.co/agents, read 8 Nov 2026
  15. Shapes: shapes.co/, read 8 Nov 2026
  16. Shapes: shapes.co/releases/run-workflows-for-new-starters-and-leavers, read 8 Nov 2026
  17. Shapes: shapes.co/releases/workflows-2-0-now-live-for-everyone, read 8 Nov 2026
  18. Shapes: shapes.co/releases/see-who-did-what-in-the-audit-log, read 8 Nov 2026
  19. Shapes: shapes.co/releases/turn-ai-features-on-or-off, read 8 Nov 2026
  20. Shapes: help.shapes.co/en/articles/9569339-permission-levels, read 8 Nov 2026
  21. Shapes: mcp.shapes.co/, read 8 Nov 2026
  22. Shapes: shapes.co/security-and-compliance, read 8 Nov 2026

Last reviewed 10 Nov 2026.