LoginSign Up
← Back to all sparks
L

Lattice

HR
Velocity0.0

People management platform for performance reviews, engagement, and compensation

lattice.com ↗

Lattice built an AI agent worth visiting, then handed its data to everyone else's.

ai-agentmcpperformance-reviewscompensationmeeting-capturehr-analytics
◆Current state
Lattice runs AI Agent as its own changelog category alongside Performance, Compensation, and Analytics, and the June work built it into something worth opening: the Agent attends 1:1s and captures summaries, action items, and coaching insights, answers data questions with charts rather than prose, and takes voice input. July then pointed the same capability outward. An MCP server exposes Lattice to Claude, OpenAI, and Slack, and AI review drafts are assembled from a person's own 1:1s, feedback, goals, and updates with cited sources. Underneath both, the performance and compensation machinery keeps absorbing manual setup — calibration groups generated from the org chart, approved promotions carried from a review cycle into a comp cycle.
◆Where it's heading
The interesting move is that Lattice stopped assuming it wins by being the app you open. June made the Agent a destination; July made Lattice reachable from Claude, ChatGPT, and Slack instead, which is a bet that being the HR data layer beats owning the chat surface. Review drafts show what that data layer is being accumulated for: the meeting capture, feedback, and goals the Agent has been collecting are now the raw material for the review itself, with citations as the trust mechanism. Compensation is on a slower and more procedural track, closing loops — sending recommendations back for revision, exporting with HRIS IDs attached — rather than adding AI.
◆Prediction
Expect the review-drafting work and the MCP surface to converge, so a manager can draft an evidence-cited review from Claude or Slack without opening Lattice. Whether compensation gets the same treatment is genuinely unclear from these entries — that area is still being hardened procedurally, and approval chains carry consequences that make an AI draft a harder sell.

◆Recent moves

  1. 2mo ago

    AI Agent in 1:1s

    ⚡ SPARK

    The point where the Agent stopped waiting to be asked. Sitting in the 1:1 and capturing summaries, action items, next-meeting agendas, and coaching insights for both participants is what produced the corpus July's review drafts are built from.

  2. 2mo ago

    Revamped Homepage

    The Home rebuild makes Agent output the default entry point: 1:1 prep, task follow-ups, and AI-summarized team updates lead the page. It is the packaging decision that follows from the Agent work rather than a capability of its own.

  3. 2mo ago

    Analytics Charts in Lattice Agent

    Agent answers to data questions come back as charts or tables with a link to the full report instead of a text summary. It narrows the gap between asking the Agent and using Analytics, which the July scheduling and sharing work widens further.

  4. 2mo ago

    Voice Modality

    Voice input with automatic transcription into the Agent's draft field. A modality add on an established surface, consistent with the June push to make the Agent something used in the flow of a meeting rather than typed at.

  5. 2mo ago

    Auto-Create Calibration Groups

    Calibration groups generated from employee attributes or the org chart in one step, with a preview before finalizing. Part of the steady inference of review-cycle setup from data Lattice already holds.

  6. 2mo ago

    Calibration Group Conflict Avoidance

    Participant conflicts in calibration groups are flagged at assignment time with a confirmation modal before a conflicting change saves. A guardrail on the auto-creation work shipped alongside it.