Early access
Job Architecture Agent
Benchmarks every role and pay band against the market. Flags drift.
Hours1 (modeled outcome) to re-benchmark a job family, down from 4 to 6 weeks
HR agent
Early accessPerformance Agent
The Performance Agent is a governed AI agent that drafts evidence-based reviews. Managers edit and own the result.
How it works
What the Performance Agent does
Assembles each employee's goals, feedback, and delivered work into a timeline, then drafts the review against your competency framework. The manager sets the rating.
- 01
Connect the record
The agent reads goals, feedback, and prior reviews from your performance system, and delivered work from project, ticketing, and code systems the employee's team uses.
- 02
Build the evidence timeline
For each employee it assembles a dated timeline of goals, outcomes, and feedback, and flags periods with little evidence.
- 03
Draft against your framework
The draft follows your competency model and rating scale. Each statement cites its evidence. A language check flags phrasing associated with bias for the manager to revise.
- 04
Manager edits, rates, submits
The manager owns the final text and the rating. Calibration packs go to HRBPs. The Registry reports cycle completion, time spent, and evidence coverage.
app.meridian.example/agents/performance/draft
Review draft · D. Okafor · H2 2026
Draft
Summary
Delivered the multi-state payroll consolidation two weeks ahead of plan and reduced off-cycle payments by 38% across three quarters.
Strengths, with evidence
- Led the Colorado and Washington rate migrations [goal 3]
- Cut month-end payroll reconciliation from 6 days to 2 [metric]
- Mentored two analysts through certification [peer feedback ×3]
Evidence used
- Goals 7
- Peer feedback 14
- Delivered work items 62
- Compensation data used None
The manager edits and submits. Meridian never files a review.
The agent contract
Six terms, applied to this agent
Every Meridian agent runs under the same six terms, enforced at the Gateway below the agent. Here is what each one means for the Performance Agent.
Scoped to one workflow
Drafts one review from the record. The manager edits, rates, and submits.
Reads only permitted data
Reads only the goals, feedback, and delivered work the manager could already see.
Every action logged
All reads logged. Employees can see which evidence a shared review cites.
Consequential actions require human approval
Never assigns or recommends a rating. Drafts stay private until the manager submits.
Measurable outcome
Drafting time, on-time submission, and evidence coverage reported per cycle.
Runs on your chosen model
Run on your chosen model. The bias-language check runs the same way on every model.
Scope
Data, actions, and approvals
What the Performance Agent reads, what it does on its own, and where a person signs.
Data it reads
- Goals and OKRs with status history
- Peer, upward, and manager feedback
- Delivered work from project management, ticketing, and code hosting
- Prior reviews and development plans
- Competency framework and rating scale
- Role, level, and tenure from the HRIS
Actions it takes
- Assemble a dated evidence timeline per employee
- Draft the review in your competency framework with evidence links
- Flag thin-evidence periods and suggest questions for the manager
- Run a bias-language check and propose neutral phrasing
- Draft development goals from gaps identified in the review
- Prepare calibration packs per team with rating distributions and rationale
What needs approval
- The agent never assigns or recommends a final rating; the manager sets it
- Drafts are private to the manager until the manager submits
- Protected characteristics are excluded; comparisons are tenure- and level-adjusted
- Employees can see which evidence the review cites once it is shared with them
- 1:1 notes are used only if the manager explicitly shares them; all reads are logged
Outcomes
Numbers the Performance Agent is accountable for
Modeled outcomes from design-partner deployments. The Registry reports your actuals from the first week an agent is live.
−60%
manager time to draft a review1
100%
of draft statements linked to a goal, feedback item, or delivered work
+22pts
reviews submitted on time1
- 1.Modeled outcome from design-partner deployments. ↩
Integrations
Systems it works with
Connected through prebuilt connectors and zero-copy warehouse access in Data Fabric. The agent acts as the user and inherits each system’s permissions on every read.
- HRIS and performance management
- Project management and ticketing
- Code hosting
- Document collaboration
- Slack
- Microsoft Teams
Credits
What the Performance Agent costs
In early access, available on Enterprise plans with custom credit pools. Credits are consumed per completed action, never per message. Usage beyond the plan is billed at $0.12 per credit.
| Action | Credits |
|---|---|
| Review draft with evidence timeline | 4 |
| Feedback synthesis, per employee | 2 |
| Calibration pack, per team | 6 |
Questions
Questions about the Performance Agent
Answered the way buyers ask them. Each answer is also published as structured data.
Related terms
AI can draft them from evidence. It should not rate people. Meridian's Performance Agent assembles the record and writes a first draft where every sentence links to a goal, a piece of feedback, or delivered work. The manager edits the draft, decides the rating, and owns the result.
By starting from the full record rather than recent memory, by flagging phrasing that research associates with biased reviews, and by giving HRBPs calibration packs with tenure- and level-adjusted comparisons. It does not eliminate bias. It makes the evidence visible so people can check their own judgment.
Employees see the review their manager submits, which is the manager's text. They can also see which evidence items the review cites. They never see unsubmitted drafts.
Your performance and goals system, plus the systems where work is visible: project management, ticketing, code hosting, and document collaboration. You choose which sources each team connects. The agent reads only what the manager could already see.
Related agents
Agents most often deployed alongside the Performance Agent, on the same data and the same approval policy.
Get started
Put the Performance Agent to work.
Start with one workflow, one approver, and one number to move. Most design partners were live in five weeks.