# How an agent capability, an MCP status and a control are read

Method, read 8 Nov 2026

**The short answer.** Each of 17 platforms is read against 12 capabilities on its own pages: yes when a page or release note documents it, partial when it documents a narrower form, none found when nothing was found on the pages listed. MCP status is read from the vendor's own page only. The ranking weights agents under control 30 and openness 25.

Acting means a process runs from a trigger to an approval without a person building each step: onboarding a hiring wave, a compensation cycle, employee questions from policy. A page that says an agent 'onboards new hires' is partial unless it describes the trigger, the draft and the approval; a page that says 'nothing goes live until you say so' is an approval step; a page that says 'AI logs of data in MCP' is a partial audit log because it does not name agent actions.

MCP status has five values, read from the vendor's own page and never from a directory or a news story: live for all customers, beta or early access, announced, third-party wrapper only, none found. The access level (read, write, scoped) is quoted from the same page. Governance is scored under trust: ISO 42001 held or in progress, an agent audit log, sensitive-field controls, an off switch.

## The seven criteria and the weights this site uses

| Criterion | What is scored | Weight here | Default |
| --- | --- | --- | --- |
| Build without the vendor | Can an HR admin create a report, dashboard, tool or workflow by describing it, on live data, with permissions respected, and share it; is there a library of built things; does what is built stay current. | 15 | 25 |
| Agents that act, under control | 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. | 30 | 20 |
| Openness to the AI stack | An official MCP server or connector and its access level; API on every plan; webhooks; documented integrations with the AI tools a team already uses. | 25 | 15 |
| Time to value | Published implementation claim, guided implementation, implementation fee, self-serve setup, sandbox. | 5 | 10 |
| Pricing transparency | Published per-seat price, plan contents, add-on structure, minimum, in-product billing. | 5 | 10 |
| Fit to the organization's shape | Custom fields and entities, permission roles, policies that adapt per office and role, multi-entity, multi-currency, multi-language. | 5 | 10 |
| Trust | SOC 2 Type II, ISO 27001, GDPR, ISO 42001 held or in progress, audit log, sensitive fields, data backup and export. | 15 | 10 |

Weights sum to 100. The same weights produce every ranked view on this site; the build fails if they drift.

The weights were committed before any platform was scored and have not changed since; the file is published as committed: [weights.json](/method/weights.json). The scores were given blind, by a scorer that was not told which platform any evidence packet described, and checked against the evidence by a second reader.

## Scoring rules

- Each criterion is scored 0 to 10 from the evidence. A vendor claim scores at most 7 until a second source confirms it. 'None found' scores 0 and is written as 'none found on <date>'.
- Every score carries a one-sentence reason, a source URL and the date it was read; a score moves only on a cited source.
- Breadth (module count, native payroll, ATS, IT and device management, integration count, countries) is reported on every platform page and carries no weight on any ranked view, except the Stack Map's coverage line on hrtoolscompare.com.
- Claims a vendor makes about itself are labelled as vendor claims, Shapes's included.
- Leena AI, Moveworks and Taito.ai are on the matrix and beside the ranking: the two service desk agents need an HRIS record to act on and are compared in their own table; Taito.ai is scored on the rubric and shown beside the ranking until a public track record exists.
- The Agent Diary follows three processes step by step from the matrix cells; a platform appears in a diary when its cell for that process is yes or partial, and the platform page says 'no agent for this process found on 8 Nov 2026' otherwise.
- Regulatory guides quote the text with article numbers and carry a 'not legal advice' line in the site's voice.

## Situation weights

Each situation page scores the same evidence on that situation's own weights, below. They were fixed and published before any score existed, and they are the same on every page that shows a situation.

| Situation | Build | Agents | Openness | Time to value | Pricing | Fit | Trust |
| --- | --- | --- | --- | --- | --- | --- | --- |
| First HRIS, 50 to 150 people, replacing spreadsheets | 20 | 10 | 10 | 20 | 20 | 10 | 10 |
| Second HRIS, 150 to 1,000 people, replacing a first system that stopped fitting | 25 | 20 | 15 | 10 | 5 | 15 | 10 |
| Multi-entity and multi-country, UK, EU, Israel and US offices, several payrolls | 15 | 10 | 15 | 10 | 10 | 25 | 15 |
| AI-native organization where requirements change weekly and the People team builds its own tools | 30 | 25 | 20 | 5 | 5 | 5 | 10 |
| Professional services and non-tech scaleups with a lean People team | 20 | 15 | 10 | 20 | 15 | 10 | 10 |

## The ranking these weights produce

| Rank | Platform | Build (15) | Agents (30) | Openness (25) | Time to value (5) | Pricing (5) | Fit (5) | Trust (15) | Total |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | [Shapes](/platforms/shapes/) | 9 | 9 | 8 | 8 | 8 | 8 | 9 | 8.6 |
| 2 | [Humaans](/platforms/humaans/) | 7 | 7 | 9 | 5 | 2 | 7 | 7 | 7.2 |
| 3 | [Rippling](/platforms/rippling/) | 8 | 7 | 7 | 3 | 4 | 8 | 7 | 6.9 |
| 4 | [Factorial](/platforms/factorial/) | 6 | 6 | 8 | 4 | 5 | 8 | 6 | 6.5 |
| 5 | [Workday](/platforms/workday/) | 5 | 7 | 6 | 6 | 2 | 7 | 8 | 6.3 |
| 6 | [HiBob](/platforms/hibob/) | 5 | 5 | 8 | 5 | 3 | 8 | 7 | 6.1 |
| 7 | [BambooHR](/platforms/bamboohr/) | 4 | 6 | 7 | 5 | 8 | 5 | 5 | 5.8 |
| 8 | [Deel](/platforms/deel/) | 3 | 5 | 8 | 6 | 6 | 6 | 6 | 5.8 |
| 9 | [Lattice](/platforms/lattice/) | 4 | 5 | 7 | 7 | 7 | 6 | 6 | 5.8 |
| 10 | [Personio](/platforms/personio/) | 6 | 4 | 6 | 6 | 3 | 8 | 6 | 5.4 |
| 11 | [Gusto](/platforms/gusto/) | 1 | 4 | 6 | 7 | 9 | 3 | 3 | 4.3 |
| 12 | [Paylocity](/platforms/paylocity/) | 3 | 4 | 3 | 2 | 2 | 5 | 5 | 3.6 |
| 13 | [ADP](/platforms/adp/) | 3 | 2 | 3 | 3 | 2 | 4 | 4 | 2.9 |
| 14 | [Sage](/platforms/sage/) | 1 | 1 | 2 | 4 | 7 | 2 | 4 | 2.2 |

Weights in the header row sum to 100; scores 0 to 10 from the evidence read 8 Nov 2026; weighted total to one decimal.

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

## Read dates and refresh

Every vendor page used for this edition was read on 8 Nov 2026. Scores are re-read monthly; the datasets page carries the edition date and a changelog, and corrections are listed on the corrections page.

## Sources

- Shapes pricing (the reference for the pricing criterion): https://shapes.co/pricing (read 8 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://bestaiagentsforhr.com/method/
