For Agents
Agent Context
LEADERS STANDARD WORK — COMPRESSED CONTEXT
PURPOSE: Daily control rhythm for human+agent production systems.
Ensures quality, surfaces problems, enables improvement.
FRAMEWORK: Signal / Learn / Operate
- SIGNAL: Review metrics and alerts. What does the data say?
- LEARN: Analyze root causes. What does it mean?
- OPERATE: Take action. What do we do about it?
DAILY RHYTHM (3 control windows):
1. AM SHIFT (~8:00 AM)
- Signal: Review overnight runs, check quality scores
- Learn: Identify any failures, note patterns
- Operate: Set today's run plan, assign priorities
2. MIDDAY SHIFT (~12:00 PM)
- Signal: Check morning run progress
- Learn: Are we on track? Any blockers?
- Operate: Adjust plan, escalate if needed
3. PM SHIFT (~5:00 PM)
- Signal: Review day's results, finalize metrics
- Learn: What worked? What didn't?
- Operate: Close shift log, queue improvements
REQUIRED ARTIFACTS (every shift):
- Shift Log: timestamped record of observations and actions
- Run Plan: what will be produced, by whom, by when
- Improvement Ticket(s): captured learnings for system updates
SCOREBOARD METRICS:
- Speed: cycle time vs target
- Quality: QC pass rate (first-pass yield)
- Cost: resources consumed vs budget
ANDON TRIGGERS (stop and escalate):
- QC score <75% after 2 attempts
- Agent produces unexpected/off-topic output
- Context file missing or corrupted
- Run exceeds 2x expected duration
HARD RULE: Leaders review every shift. No exceptions.
The daily rhythm IS the quality system.
Framework
What Is Leaders Standard Work?
Leaders Standard Work (LSW) is a structured daily routine that production leaders follow to maintain control, ensure quality, and drive improvement. It's borrowed from Lean manufacturing, where supervisors have defined checkpoints throughout the day rather than reactive firefighting.
In human+agent systems, LSW serves the same purpose: it creates predictable control points where humans review agent output, make decisions, and course-correct before problems compound.
Why This Matters
Without LSW, production becomes reactive. Problems are discovered too late. Quality degrades gradually. Improvement opportunities are missed. LSW creates the discipline that makes sustained excellence possible.
"Standard work is the baseline for continuous improvement. You cannot improve a process that is not standardized."
— Taiichi Ohno, Toyota Production SystemFramework
Signal / Learn / Operate
Every shift review follows the same three-phase structure. This creates consistency and ensures nothing is missed.
| Phase | Question | Activities | Output |
|---|---|---|---|
| Signal | What does the data say? | Review metrics, check alerts, scan logs | Observations noted in Shift Log |
| Learn | What does it mean? | Analyze patterns, identify root causes, assess risks | Insights captured, improvement ideas logged |
| Operate | What do we do about it? | Make decisions, assign actions, adjust plans | Run Plan updated, actions assigned |
Order Matters
Always Signal before Learn, always Learn before Operate. Jumping straight to action without understanding the data leads to solving the wrong problems.
Daily Rhythm
AM Shift Review
Morning Control Window
~8:00 AM · 15-30 minSignal
- Review any overnight automated runs
- Check quality scores from previous day's late runs
- Scan for system alerts or errors
- Review improvement tickets from previous shift
Learn
- Identify any failures or quality issues
- Note patterns across multiple runs
- Assess capacity and resource availability
- Flag any blockers or dependencies
Operate
- Set today's Run Plan with priorities
- Assign work to agents/team members
- Escalate any issues from overnight
- Open today's Shift Log entry
Daily Rhythm
Midday Shift Review
Midday Control Window
~12:00 PM · 10-20 minSignal
- Check progress against Run Plan
- Review QC results from morning runs
- Note any new alerts or anomalies
- Check resource utilization
Learn
- Are we on track to meet daily targets?
- Any emerging blockers or risks?
- What's causing variance from plan?
- Are agents performing as expected?
Operate
- Adjust afternoon priorities if needed
- Escalate blockers that can't be self-resolved
- Re-allocate resources if behind
- Update Shift Log with midday status
Daily Rhythm
PM Shift Review
Evening Control Window
~5:00 PM · 15-30 minSignal
- Review all completed runs for the day
- Finalize quality metrics
- Collect all improvement tickets generated
- Note any outstanding work in progress
Learn
- What worked well today?
- What didn't go as planned?
- Are there systemic issues to address?
- What patterns are emerging over time?
Operate
- Close out today's Shift Log
- Queue improvement tickets for batch review
- Set up any overnight automated runs
- Brief incoming shift (if applicable)
Controls
Daily Scoreboard
Three metrics tell you if the production system is healthy. Track them every day.
Speed
Cycle Time
Quality
First-Pass Yield
Cost
Resources Used
| Metric | What It Measures | Target | Red Flag |
|---|---|---|---|
| Cycle Time | Time from request to delivery | ≤ target per deliverable type | > 2x expected duration |
| First-Pass Yield | % of runs that pass QC without rework | ≥ 85% | < 70% |
| Resources Used | Tokens, compute, human hours | Within budget | > 20% over budget |
Leading vs Lagging
First-pass yield is a leading indicator—declining yield predicts future problems. Cycle time and cost are lagging indicators—they measure what already happened. Watch yield most closely.
Controls
Andon Triggers
These conditions require immediate escalation. Stop work and involve a human decision-maker.
- QC Failure Loop: QC score <75% after 2 regeneration attempts on the same deliverable
- Off-Topic Output: Agent produces content unrelated to the assigned task
- Missing Context: Context file is missing, corrupted, or contains invalid data
- Duration Exceeded: Run takes >2x expected time without clear reason
- Resource Overrun: Token/compute usage >3x expected for a single run
- System Error: Agent crashes, hangs, or returns error repeatedly
Non-Negotiable
Andon triggers are not suggestions. When triggered, the leader must stop and investigate before resuming production. Do not "push through" hoping it will resolve itself.
Controls
Required Artifacts
Three artifacts must be maintained to ensure traceability and enable improvement.
Shift Log
A timestamped record of observations, decisions, and actions taken during each shift. The source of truth for what happened.
Run Plan
The queue of work to be completed, with assignments, priorities, and target completion times. Updated at each shift.
Improvement Ticket
A captured learning or suggestion for system improvement. Generated during runs and batched for periodic review.
Roles
Roles & Accountability
Clear ownership prevents gaps and confusion. Every role has defined responsibilities.
| Role | Responsibilities | Accountable For |
|---|---|---|
| Production Leader | Conducts shift reviews, maintains artifacts, escalates issues | Daily metrics, shift log completeness, escalation timeliness |
| Agent Operator | Configures and runs agents, performs QC, logs observations | Run execution, first-pass yield, improvement ticket submission |
| Process Owner | Maintains Process/Example/Context/Quality files, reviews improvements | Process quality, file currency, improvement implementation |
| Escalation Point | Receives andon triggers, makes judgment calls, resolves blockers | Response time, resolution quality, systemic issue identification |
One Person, Multiple Roles
In small teams, one person may hold multiple roles. That's fine—what matters is that every responsibility has a clear owner. Ambiguity is the enemy.
"Management is doing things right; leadership is doing the right things."
— Peter DruckerLeaders Standard Work bridges both: the structured rhythm ensures things are done right, while the Signal/Learn/Operate framework ensures the right things are prioritized.