Agent Studio

Configure and monitor autonomous agents that take moves to improve your product maturity.

Agent Studio is the surface for managing autonomous workflows. Agents run on schedule or trigger, analyze your product data, and either surface findings or dispatch moves into Linear, Slack, or the re-score queue. Pro plans and above.

Agent Studio

Agent Studio is the surface for managing autonomous agents. Agents run on schedule or event triggers, analyze your product data, and produce structured artifacts or dispatch moves to connected tools (Linear, Slack, re-score queue).

Agent Studio is available on Pro plans and above.

What agents do

Agents are purpose-built workflows that monitor specific aspects of your product maturity:

Strategic intelligence

Reads scoring trajectory, surfaces cross-product patterns, and surfaces strategic moves for your portfolio.

Anomaly detection

Monitors score moves and adapter signal for unusual patterns, alerting you to regressions or unexpected lifts.

Competitive monitor

Tracks competitor product changes and scoring moves, surfacing competitive intelligence in your dashboard.

Coaching digest

Surfaces periodic coaching summaries grounded in your scores, highlighting the highest-impact moves for your current stage.

Lifecycle tracker

Monitors your lifecycle placement and flags when team behavior diverges from expected patterns for your stage.

Outcome tracker

Closes the feedback loop by linking dispatched moves to measured score lifts over time.

Getting started

1. Navigate to Agents

   Open Agents from the main navigation. Your account will be provisioned with default agent definitions on first visit.

2. Review default agents

   Each agent shows its type, description, current status (active or paused), and last run time. Default agents start at Notify autonomy.

3. Activate an agent

   Toggle an agent from Paused to Active to start its scheduled runs.

4. Promote autonomy when ready

   After validating an agent’s output quality over several runs, promote it to Suggest or Auto to enable action dispatch.

Autonomy levels

Every agent has a configurable autonomy level that controls how much it can do without human approval:

Level	Behavior	Best for
Notify	Surfaces findings in the dashboard only. No automated actions.	New agents you are still evaluating
Suggest	Creates draft actions that go into your approval queue for review before execution.	Agents you trust but want to oversee
Auto	Executes configured actions immediately when conditions are met.	High-confidence, well-validated agents

Approval queue (Suggest mode)

When an agent is set to Suggest, its actions are queued for human review instead of executing immediately. Each pending action shows:

• The agent that produced it and the run that triggered it
• The proposed action (Linear issue title, description, priority)
• An Approve or Reject button with an optional reason field

Approved actions execute immediately. Rejected actions are recorded in the decision history so the system learns from your judgment over time.

New agents default to Notify. Only promote to Auto after validating output quality over several runs.

Review before dispatch

Agents with Review required enabled hold their artifacts (strategic briefs, anomaly reports) for human review before dispatching to Slack or email. Enable this under Settings > Review on any agent to ensure a human sees every insight before it reaches the team.

Set the autonomy level from the agent configuration panel under Settings > Autonomy.

Configuring triggers

Each agent can be triggered in multiple ways:

Trigger type	Description
Schedule	Runs on a recurring schedule (daily, weekly, or custom cron expression)
Event	Fires when a specific condition is met (score threshold, integration sync, delta breach)
Manual	Run on demand from the Agent Studio UI

Trigger conditions

Event-based triggers support conditions that refine when they fire:

• Score threshold trigger when a dimension score drops below a specified value
• Score delta trigger when a composite score changes by more than N points
• Sync completed trigger after a specific integration syncs new data
• Product added trigger when a new product is created in the account

Each trigger has a configurable cooldown (minimum time between firings) to prevent noise.

Start with the default weekly schedule. Once you are comfortable with the agent’s output, add event-based triggers for real-time alerts.

Actions

Agents set to Suggest or Auto autonomy can dispatch real actions when triggered:

Action type	What it does
Create Linear issue	Opens a Linear issue with the finding, suggested owner, and priority
Send Slack message	Posts a structured summary to a configured channel
Trigger re-score	Queues a fresh score of the affected product
Queue coaching recommendation	Surfaces a “Do This Next” item in the next DAC session
Refresh documentation	Creates a Linear ticket flagging documentation that has drifted from implementation. Auto-generated by the Spec Quality agent when dimensions score 2 or below.

All dispatched actions are logged in the agent run history with a full audit trail. In Suggest mode, actions appear as drafts in your approval queue before being sent.

Compound flywheel

The compound flywheel tracks whether agent actions are producing measurable score improvements over time. For each dispatched action:

Field	Description
Linked outcome	Which dimension the action targeted
Before score	Dimension score at the time the action was dispatched
After score	Dimension score on the next re-score
Loop closed	Whether the improvement has been verified

Learning cycles

The flywheel counts completed learning cycles (Score, Connect, Correlate, Act, Learn). Each complete revolution makes the next revolution more accurate. The dashboard shows your cycle count, accuracy trend over time, and intelligence depth percentage.

Outcome attribution

When an action closes its loop (score improves), you can annotate why the score changed:

Attribution	Meaning
Recommendation followed	The team implemented the recommendation as suggested
External factor	Score changed due to hiring, reorg, or market shift
Coincidence	No causal relationship between the action and outcome
Partial	The recommendation was one of several contributing factors

These annotations feed back into the intelligence engine so future recommendations become more accurate.

Access the flywheel from the Agent Studio dashboard or from any agent’s detail page.

Product vitals

Product vitals surfaces real-time operational health signals for your active products. The vitals panel appears on the Intelligence dashboard and updates on every agent sync:

Metric	What it measures
Agent runs	Total agent executions in the current period
Insights generated	Distinct findings surfaced across all agent runs
Actions dispatched	Actions sent to Linear, Slack, re-score queue, etc.
Loops closed	Actions verified to have improved a score
Score velocity	Composite score change per week over the last 30 days

LLM-grounded narratives

Strategic Intelligence and Anomaly Detection agents use Claude to surface human-quality narratives grounded in your scoring data. The agent engine runs a structured analysis first (identifying patterns, anomalies, and priorities), then enriches the output with an LLM call that produces readable strategic briefs with cited evidence.

Other agents (Voice of Customer, Strategy Brief, Spec Quality, Code Quality) run as pure heuristic analyzers by default. You can enable LLM narratives for any agent under Settings > Model by selecting a model.

LLM-grounded runs consume more credits than heuristic runs but produce richer narratives. Start with Haiku for cost efficiency.

Browsing artifacts

Every agent run produces one or more artifacts, structured outputs stored in your account:

• Reports Markdown-formatted analysis with data and recommendations
• Alerts Short notifications about anomalies or threshold breaches
• Recommendations Specific “Do This Next” actions ranked by impact

Navigate to the agent detail page and select the Runs tab. Click any artifact to view its full content. Artifacts are stored permanently and can be referenced in DAC coaching conversations.

Performance monitoring

Metric	Description
Run history	Timeline of all agent executions with status
Success rate	Percentage of runs that completed without errors
Average duration	Typical run time for planning capacity
Artifacts generated	Count of outputs produced per run
Token usage	LLM tokens consumed per run for cost tracking

Troubleshooting failed runs

1. Check the run detail page for error messages
2. Verify that connected integrations are still authorized
3. Confirm the product being analyzed still exists
4. Review your credit balance (agent runs consume credits)

Agent runs consume credits. A typical strategic intelligence run uses 1-3 credits. Monitor usage under Settings > Limits.

Managing agents

Action	Description
Pause	Stop scheduled runs without losing configuration
Resume	Re-enable a paused agent
Edit	Change the agent’s name, description, or configuration
Delete	Permanently remove the agent and its trigger configurations (artifacts are retained)

Next steps

Connect adapters

Agents work best with rich signal from connected tools.

DAC

Reference agent artifacts in coaching for deeper analysis.

Leader: using agents

A guide to configuring agents as a team lead or VP.

API reference

Manage agents programmatically via the REST API.