Lattice unveiled Pace on 12 November, a business performance platform that gives every AI agent a job, a goal and a human manager, and keeps a record of each agent’s manager, goal, cost and results. It is the clearest attempt yet to put AI agents inside the same management system a company uses for its people. Workday and BambooHR, in material published over the past month, set up agent accountability in other ways: Workday ties each agent to the customer’s existing policies and approval rules, and BambooHR runs every AI request as the signed-in employee. Each design gives a different answer to who owns an agent’s work once it starts doing real tasks.
What Lattice announced
Lattice revealed Pace at its Innovation Day in San Francisco. According to the company’s release, Pace is built in Slack and combines more than 60 ready-made “Rhythms” with continuous performance management for people and agents. Lattice describes Rhythms as recipes for how people and AI work together, covering routines such as onboarding, reorganisations, project kickoffs and manager check-ins. Pace drafts, schedules and follows up on those routines, and brings people in when judgment or approval is needed.
The product is still in development. Lattice says it expects early access in December 2026 and that features, functionality and availability are subject to change. Lattice says it serves nearly 5,000 customers.
Lattice frames the launch around spending. Its release cites an IDC projection of $3.3 trillion in cumulative investment in digital labor technologies by 2030, and a KPMG finding that only 7% of senior leaders report having established ROI from AI. Lattice’s chief executive put the company’s position in plain terms.
“We need to invest in people’s skills, give agents clear jobs and human managers, and keep human judgment at the center as we drive business performance,” said Sarah Franklin, CEO of Lattice.
How Pace models an agent
The working part of the design is what Lattice calls a Trusted Agent Record. Each agent in Pace gets defined responsibilities, a human manager and performance goals, and the record holds the agent’s manager, goal, cost and results. Lattice says this gives leaders accountability for what agents deliver in addition to visibility into what they do.
Pace Checks run the loop on top of that record. Per the release, they track progress against goals, spot work that is off track, recommend next steps and measure whether a change produced a better result. Lattice applies the same loop to people and agents.
Pace is designed to work across existing tools and agents, so a customer does not have to replace the agents it already runs. And Lattice says Pace respects existing permissions and security controls, with clear boundaries on what it can see and do.
In our reading, the Trusted Agent Record works as an employee-style file for software, with a reporting line and a cost line, held in a people platform. That places the agent inside HR’s system of record for performance, which is a different home from IT’s asset inventory or a vendor’s admin console.
Workday’s model: the agent inherits the business process
Workday announced autopilot capabilities for its agents at Workday Rising on 13 October, which this site covered when it reported Workday’s autopilot agents for hiring and payroll with customer-set limits. The accountability design there starts from the process. Workday says its agents inherit each company’s policies, permissions and approval rules from day one.
The settings a customer controls
According to Workday’s release, customers choose which tasks run automatically and how far each agent can go. They decide which actions an agent can take on its own, which need review, and when a person makes the final call. Every action an agent takes is logged, so a customer can see what it did, what information it used and who approved it. Customers can also set how often agents run, which Workday presents as a way to manage both activity and cost as agent use grows.
The recruiting example shows where Workday draws the line. The Talent Acquisition Agent checks the queue, finds qualified candidates once a job is posted and can notify recruiters in tools such as Slack with draft outreach. Sourcing runs in the background, and outreach to candidates still requires recruiter approval before anything is sent. Autopilot in the Talent Acquisition Agent was available to early adopter customers at launch, with general availability planned for early 2027. Autopilot for the Payroll, Accounting and Deployment Agents is planned for early adopters in early calendar 2027.
“Our agents can work in the background because they follow the same policies, permissions, and approvals that already run the business,” said Gerrit Kazmaier, president, product and technology, Workday.
Partner agents that advise
On 12 November, the same day as the Lattice launch, Workday and The Josh Bersin Company said they will make Galileo, the firm’s AI research and advisory agent, available to Workday customers in Sana, Workday’s agentic experience and orchestration engine. Galileo draws on more than 50,000 pages of research covering hiring, employee development, workforce planning, leadership and organisational design.
The release sets two limits on the agent. Galileo will operate within the security policies, permissions and access controls organisations already use, and it is designed to advise on workforce strategy, including hiring approaches, skills development and organisational design, rather than make employment decisions. Workday HCM customers can buy Galileo now, and the integrated agent is expected to reach Sana Enterprise customers in February 2027.
“By bringing specialized partner agents like Galileo into Sana, we are giving organizations a trusted advisory resource right in their daily workflow, helping them evaluate workforce strategies with greater speed and confidence,” said Nikki Helmer, vice president, talent management and insights, Workday.
Josh Bersin, CEO of The Josh Bersin Company, described the aim in terms of HR’s workload. “Galileo brings our research and experience directly into their workflow, so they can learn from what other organizations have done rather than starting from scratch,” he said.
Across both Workday releases, the accountability sits in the configuration: the policy, the permission set, the approval step and the log.
BambooHR’s model: the agent runs as the person
BambooHR’s documentation for its Claude connector, still in beta, describes a third design, which this site covered when BambooHR launched a Claude connector that follows existing permissions. The connector runs on the BambooHR MCP server. Every request runs as the signed-in user, against the same permissions BambooHR applies when that person is logged in, and the documentation states that the connector never grants access the user does not already have.
An admin turns on the AI Connectors app and chooses who can connect: admins only, which is the default, specific access levels, or all employees. Time off balances require the Time Off view permission for that employee, and hiring data requires access to applicant tracking settings. Being someone’s manager does not by itself grant either. BambooHR notes that two people can ask the same question and get different numbers, because each answer is limited to what that person can read.
Write actions get their own guidance. Claude asks before it runs a tool, and BambooHR advises users to read what it is about to do, especially for anything that creates a time off request, updates an employee record or changes a goal. Choosing “always allow” for a write tool means future runs will not ask. An admin can turn off the AI Connectors app, which revokes every connection at the company at once.
In this model the AI assistant has no separate identity in the HR system. Its accountability is the accountability of the employee who asked, and its reach is that employee’s access level.
Three places to put accountability
- Lattice Pace: the agent has its own record, a named human manager, a goal and a cost line. Accountability runs through a reporting line, like an employee’s.
- Workday: the agent inherits the company’s policies, permissions and approval rules, and the customer sets what it may do alone. Accountability runs through configuration and an action log.
- BambooHR connector: the assistant acts as the signed-in user with that user’s permissions. Accountability runs through the person who asked.
Each design leaves different questions open. A reporting line tells a company who answers for an agent’s results, but the release does not say how a manager’s authority over an agent maps to that agent’s system permissions in other tools. A configuration model records what an agent did and who approved it, and still needs someone to own the configuration. A run-as-user model keeps permissions tidy, and BambooHR itself warns that an answer can be incomplete because of permissions without saying so.
These models will also meet inside one company. A Workday customer that adopts Pace, and lets employees connect an AI assistant to another HR system, would have agents governed three ways at once. Lattice says Pace is designed to work across existing tools and agents, which suggests the company expects that mix.
The skills behind agent oversight sit outside HR
The people expected to run this oversight are being defined on the IT side. CompTIA said on 22 October that it plans a credential, CompTIA AgentOps+, for professionals who deploy, secure, monitor and govern AI agents. The planned skills include securing agent identities, permissions and data access, monitoring performance and behaviour, and supporting governance, accountability, auditability and human oversight. The certification is aimed at DevOps, cloud operations, platform engineering, systems administration and cybersecurity staff, and is planned for release in Q2 2027. We argued in an earlier opinion piece that HR should staff agent operations before the credential exists.
“The focus has been on what AI agents can do, but the more urgent question is who will secure, govern and be accountable for them,” said Carl Bowman, senior vice president, exam services, CompTIA.
CompTIA’s release cites KPMG’s first-quarter 2026 AI Quarterly Pulse Survey, which found 54% of organisations actively deploying AI agents, and Deloitte research finding that only 21% of surveyed companies have a mature governance model for autonomous agents.
What it means for the HR leader
Decide where the agent record lives
Before an agent touches people data, HR and IT should agree which system holds the record of each agent: its owner, its scope, its permissions and its results. Lattice offers one answer inside a people platform. Workday’s answer is the agent’s configuration and log inside the HCM. If both exist, the company needs one of them named as the record so an audit or an employee complaint has a single place to start.
Ask vendors for the cost and outcome fields
Lattice puts cost on the agent record, and Workday lets customers set how often agents run to manage cost. In a demo, ask each vendor to show where the per-agent cost appears, who can see it, and whether results are measured against a goal the customer sets. A cost that only appears on the vendor’s invoice cannot be managed by the person named as the agent’s manager.
Match permissions before turning on write actions
BambooHR’s guidance to leave create, update and delete tools off when users only need to ask questions is a sensible default for any connector. Workday’s requirement that a recruiter approve candidate outreach is an example of a write action held for a person. List every action an agent can take that changes an employee record, a pay figure or a message to a candidate, and assign a human approver to each before the pilot starts.
Check the dates against the budget cycle
Pace is in development with early access expected in December 2026. Galileo’s integrated agent is expected in February 2027. Workday’s payroll autopilot is planned for early adopters in early 2027, and CompTIA’s credential for Q2 2027. A 2027 plan that assumes any of these will be generally available on a given date should carry that assumption in writing, with the vendor’s stated stage next to it.
A practical first step this quarter is a one-page register of every AI agent and assistant that can read or change HR data, with four columns: owner, permissions, approval steps and cost. The vendors above each fill some of those columns already, and the empty cells show where the HR team still has to decide.
Source: Lattice, Workday, Workday and The Josh Bersin Company, BambooHR, CompTIA
