The Values Agent is a mandatory sensing agent in the Peak Operating Loop that monitors alignment with the firm's values across both people and AI agents. It is the mechanism that turns values from a poster into operating behavior, surfacing positive recognition and misalignment signals through the same triage as any other sensor.
Most firms treat values as aspirational. The Peak Operating Loop treats them as a measurable signal stream. Without that move, values cannot govern an AI-native firm, because half the operating decisions now happen inside agents that don't read posters.
The problem: values without enforcement
In a traditional services firm, values are reinforced through hiring, performance reviews, manager judgment, and culture itself. Imperfect, but real. The mechanisms exist.
An AI-native firm breaks those mechanisms. Decisions get made by agents. Work product gets produced by agents. Client interactions get handled by agents. Values without a sensing and enforcement mechanism stop governing the half of the firm that runs on software. Posters do not bind agents. Performance reviews do not catch an agent that quietly drifts. Manager judgment cannot see what 30 agents produced overnight.
The Two-Layer Operating Model and the Values Agent close that gap.
The Two-Layer Operating Model
Every value in the Peak Operating Loop is rendered twice: as a runtime rule for people, and as a runtime rule for AI agents.
A value with no runtime rule for people is a slogan. A value with no runtime rule for agents is unenforceable in any firm that uses AI at scale. Both layers are required, and both must be specific enough that compliance can be observed.
For example, a value like "integrity" is too abstract to enforce. A runtime rule like "every claim in a deliverable cites its source, and any claim that cannot be sourced gets flagged for review before it ships" is enforceable, observable, and runnable by both a person and an agent.
For agents specifically, the runtime rules sit inside the Two-Layer Operating Model: rules for how each agent behaves, and rules for how it manages any subagents it spawns. An agent that violates either layer is not just off-brand; it is failing its operating contract.
The Values Agent
The Values Agent is the sensor that watches whether the runtime rules are actually being followed.
It reads:
- Status reports and project documentation for recognition mentions, kudos, and explicit references to value-aligned behavior
- Decision and escalation logs for when values were invoked and when they were violated
- Agent operation logs for behavior consistent or inconsistent with each agent's stated operating rules
- Performance review and one-to-one outputs
- Client and partner feedback when it touches behavioral matters
- Internal communications within privacy bounds
It surfaces two kinds of signals:
Positive alignment signals. Recognition events, repeat-recognized individuals or teams, examples of value-driven behavior worth celebrating. Most monitoring systems are alarm-based and only flag the negative. The Values Agent treats reinforcement as equally important, because seeing value-aligned behavior get recognized is what teaches the rest of the firm what the values actually mean in practice.
Misalignment signals. Drift patterns, repeated violations, or single events serious enough to warrant attention. These rise through the same level triage as any other signal.
Signal levels applied to values
The Values Agent uses the standard four-level triage of the Peak Operating Loop:
Level 1, Monitor. A single mention, positive or negative. Logged on the Loop Dashboard, no action required.
Level 2, Investigate. A pattern (positive or negative) that warrants a named owner taking a closer look. Two repeat recognitions of the same team this week, or two minor compliance lapses by the same agent, both belong here.
Level 3, Adapt. Leadership reviews at the next Loop Meeting. Possible outcomes: a recognition program, coaching, agent revision, or an Adapt Workflow that updates the runtime rules themselves.
Level 4, Re-aim. A serious breach by a person or agent that requires immediate action. Removing an agent from production. Changing a role. The Re-Aim Workflow gets triggered.
A signal does not become strategy until it has an owner, an implication, a recommended action, and a decision deadline. Values Agent signals follow the same rule. Misalignment does not get acted on automatically; it gets surfaced to a human owner who decides what to do. Positive signals can auto-surface as recognition. Negative signals about specific people always require human judgment before any action is taken.
Worked example: Jupiter Peak's own values as runtime rules
Jupiter Peak's six values are rendered as runtime rules for both people and agents. The Values Agent monitors each.
Integrity. People rule: name the hard truth before the comfortable version. Agent rule: cite every source; flag anything that cannot be verified; never fabricate a credential or a number. What the Values Agent watches: claims in deliverables that lack sourcing, agent outputs that softened a difficult conclusion, situations where someone took the harder path when the easier path was available.
Client Focus. People rule: every decision traces back to a measurable client outcome. Agent rule: every agent action ties to a specific Move or win condition; agents working outside their assigned scope escalate rather than expand. What the Values Agent watches: work that didn't connect to a stated outcome, decisions made on internal politics rather than client impact, recognition events tied to specific client wins.
Continuous Improvement. People rule: every working session closes with one specific thing to do better next time. Agent rule: agents log their own near-misses and suggest review-cadence adjustments. What the Values Agent watches: working sessions that ended without a next-improvement, agent logs that didn't capture self-corrections, individuals or teams who consistently improve their own metrics.
Adaptability. People rule: when the plan stops working, the plan changes; when the discipline stops working, the discipline holds. Agent rule: agents fail gracefully and escalate rather than persisting on a failing approach. What the Values Agent watches: failures that didn't escalate, plans that should have changed sooner, discipline lapses dressed up as adaptability.
Show Don't Tell. People rule: every recommendation ships with a working artifact. Agent rule: agents produce concrete deliverables, not advice about what the deliverable should contain. What the Values Agent watches: outputs that described work instead of doing it, sessions where an artifact was promised and not delivered, examples of agents producing real artifacts rather than meta-commentary.
Technology Leadership. People rule: AI leverage gets pulled forward, anchored to a win condition. Agent rule: agents propose their own augmentation candidates and flag work that should be automated. What the Values Agent watches: manual work that should have been agent-supported, opportunities where AI leverage was visible and not pulled forward, agent outputs that suggested their own improvements.
The full pattern: values rendered as operating rules, monitored by the Values Agent, acted on through Loop Meeting workflows, and updated as the firm and its agents evolve. That is the full closed loop on values.
What this looks like in a Monday Loop Meeting
In a typical week, the Values Agent contributes both kinds of signals to the Loop Dashboard. The leadership team sees them alongside the other sensors.
A positive signal might read: "Civic Insight pilot team received two team-wide recognitions this week, both citing 'earned trust compounds' on the Treasury OIG handoff." The team acknowledges it and moves on. No action required; the recognition itself is the value.
A misalignment signal might read: "Proposal orchestrator agent cited unverified industry statistics in three of its last thirty drafts. QA caught each one, but the pattern indicates drift on 'truth on the page.'" That's a Level 3. It goes on the Adapt list. An owner takes it (in this case, the agent's business owner). The owner's win condition is to tighten the agent's source-citation rule so unverified statistics fail the orchestrator's own review before reaching QA. Next Monday, the team closes the loop on whether the rule held.
The pattern is symmetric. The Values Agent treats people and agents the same way. A human team that drifted on a value would surface through the same triage and produce the same kind of Adapt item.
How the Values Agent compares to traditional culture mechanisms
Culture surveys ask people periodically whether they feel the values are being lived. Useful but lagging. By the time the survey shows drift, the drift has been happening for months. Surveys also can't see what agents are doing.
Performance reviews evaluate individuals annually against value-aligned behaviors. Useful but slow, and dependent on manager attention.
The Values Agent is continuous, observable, and bidirectional. It runs every day, on every artifact, across both people and agents. It surfaces recognition as readily as misalignment. And it produces signals that flow into the same operating cadence the rest of the firm uses, rather than living in a separate HR system that nobody opens between reviews.
This does not replace surveys or reviews. Both still have a role. The Values Agent adds the continuous, AI-readable layer that those traditional mechanisms cannot reach.
Governance of the Values Agent itself
Because the Values Agent touches sensitive ground, its own eight agent-seat elements matter more than most.
The owner is typically the CEO or COO, not a delegate. Permissions are scoped to what the agent can ethically observe; personal communications are out of bounds by default. Escalation rules are strict: positive signals can auto-surface as recognition; misalignment signals about specific individuals always require a human owner to decide what action, if any, to take. Quality standards include the requirement that any misalignment signal includes the evidence trail (which artifacts, which logs, which patterns), not just a verdict. The retirement condition is also defined, because even the Values Agent can be retired if a better mechanism replaces it.
The Two-Layer Operating Model recursing on itself: an agent governed by values, monitoring people and agents for values. That is the methodology's clearest demonstration of its own claim.
Frequently asked questions
What is the Values Agent? A mandatory sensing agent in the Peak Operating Loop that monitors alignment with the firm's values across both people and AI agents. It surfaces positive recognition signals and misalignment signals through the same level triage as any other sensor in the Sense Layer.
What is the Two-Layer Operating Model? The Peak Operating Loop's framework for governing AI agents. Each value the firm holds is rendered twice: as a runtime rule for people, and as a runtime rule for AI agents. The agent layer specifies how each agent behaves and how it manages any subagents it spawns. Values that don't run at runtime are decoration.
Why is the Values Agent mandatory? Because values must be sensed, not just stated. In an AI-native firm, half the operating decisions happen inside agents that cannot read posters or attend culture meetings. Without a sensing mechanism, the firm has no way to know whether its values are governing the half of its operations that runs on software.
How does the Values Agent handle privacy? Strictly. Permissions are scoped to what the agent can ethically observe. Personal communications are out of bounds by default. Misalignment signals about specific people require a human owner to decide what action, if any, to take; the agent never makes automated judgments about individuals.
Can the Values Agent be wrong? Yes, like any sensor. Signals it surfaces are starting points for human review, not verdicts. The methodology requires that misalignment signals come with the evidence trail (which artifacts, which logs, which patterns), so the human owner can validate or dismiss before acting.
How does this differ from culture surveys? Surveys are periodic, lagging, and only watch people. The Values Agent is continuous, leading, and watches both people and agents. The two are complementary, not competing. Surveys still have a role in capturing how people feel; the Values Agent captures what is actually happening in artifacts and operations.
Is the Values Agent watching everyone all the time? It watches the artifacts and logs the firm produces in the normal course of work, within explicit privacy bounds. It is not surveillance. It is the same kind of sensing the firm already does informally, made continuous, AI-readable, and bidirectional.
Who owns the Values Agent at our firm? Typically the CEO or COO. Not delegated to HR alone, because the agent's signals feed leadership decisions through the Loop Meeting, not personnel processes.