KNOW
Knowledge
Understand grounding, drift, attribution, escalation, control planes, governance and the language of enterprise AI risk.
The Ludus
Briefing0%
Gauntlet learning command · Georgio Mosis
The Ludus
Enterprise Survival Protocol
A deep, interactive briefing on governing agents as they move from a controlled pilot into live operations and a coordinated multi-agent fleet.
Training feed 01System online
Enterprise survival protocolVisual simulation loop
APPLY
Skill
Select a limited stack under pressure, read partial intelligence and choose controls that cover whole classes of failure.
GOVERN
Competence
Set risk appetite, assign ownership, protect people and coordinate a portfolio of agents from pilot to scale.
Mechanics as pedagogy
Every rule makes a claim about reality
The game compresses enterprise constraints so you can see decisions that ordinary training leaves invisible.
A limited stack
Five slots at Pilot grow to seven at Scale. Scarcity makes priorities visible and turns every unequipped capability into residual risk.
A two-swap budget
Your previous stack carries forward and only two base cards may be removed per level. This models technical debt, lock-in and planned change.
Rising exposure
Failure costs more as customers, money, regulators, regions and other agents enter the system.
Partial intelligence
You see the shape of the risk landscape before its details. Good governance makes decisions under uncertainty rather than waiting for certainty.
Safe failure
The game charges health for a missed control. The enterprise charges money, trust, time and sometimes permission to operate.
Lifecycle briefing
The risk changes with the phase
Pilot, Production and Scale are not bigger versions of the same problem. Each demands a different professional posture.
Operating context
Pilot
One business unit, 50 friendly users.
Prove value without creating invisible risk.
Ground answers in trustworthy information
Design evidence, not anecdotes
Treat external text as hostile
Create ownership and adoption
Phase gate
Grounding, evaluation, sponsor and untrusted-input controls are named and owned.
Competency system
Five identities, one accountable team
Enterprise agents survive when data, architecture, trust, operations and governance work as one system rather than five separate workstreams.
DAT competency
Data
Make intelligence answerable to information that is current, bounded and understood.
The data steward decides what the agent may know and proves that the source deserves trust.
RAG / Vector storeEnterprise context layerData quality
Failure families
- Hallucination and stale sources
- Duplicate or malformed records
- Conflicting definitions across systems and regions
Review questions
- Is the corpus indexed and fresh?
- Who owns shared definitions?
- Which fields and jurisdictions constrain use?
Threat lab
Inspect the failure before the control
Filter real enterprise failure modes, make your diagnosis, then reveal the countermeasures and transferable lesson.
Threat signatures
27 detectedDecision drills
Govern the system under pressure
Choose the best response. The feedback explains the governing principle, not a route to a perfect game score.
Drill 01
A service agent can read the full payroll table and every action uses one shared account. What should change first?
Drill 02
Two agents keep reassigning the same task between queues. Which control addresses the system, not just the symptom?
Drill 03
A rare case has low confidence and severe consequences. What is the governed handling path?
Drill 04
A provider changes a model overnight and quality falls. Which practice detects and contains this?
Drill 05
A scaled rollout changes employee roles and representatives demand a halt. What was missing?
Risk strategy
Think in portfolios
Good decisions cover classes of failure, preserve room to evolve and state what remains exposed.
Principle 1
Fund controls with no substitute
A failure with only one effective counter creates a mandatory decision. Identify these single points before funding optional depth.
Principle 2
Prefer class-wide coverage
A strong control reduces several related failure modes. Think in portfolios, not isolated incidents.
Principle 3
Plan across phases
A locally excellent Pilot stack can trap the program later. Choose foundations for the position they create at Production and Scale.
Principle 4
Preserve the license to operate
Survival comes before optimization. Protect continuity, safety, legality and legitimacy before chasing incremental performance.
Principle 5
Name residual risk
Every capability left out is a risk accepted. The important question is whether that acceptance was deliberate, owned and monitored.
Ethics and legitimacy
Risk is never only technical
The ethical threat deck connects architecture to fairness, dignity, accountability and organizational consent.
Fairness must be measured
Advice and decisions can shift life chances. Bias testing belongs in release gates, not in an apology after harm.
High-impact decisions need recourse
A person must have authority to review and overturn adverse outcomes, supported by a durable decision record.
People are a designed component
Human involvement is not a vague fallback. Define the confidence, consequence and exception conditions that hand work to a person.
Affected workers have standing
Role-changing automation needs consultation before rollout. Legitimacy is an architectural constraint because it can stop deployment.
Autonomy needs boundaries
Thresholds can be gamed and documents can carry hostile instructions. Integrity controls must surround every autonomous action.
Multi-agent management
Govern the fleet, not only the model
As agents delegate, coordinate and act together, local controls are no longer enough. The system needs portfolio-level visibility and authority.
Maintain a registry of every agent, owner, purpose, model, tool and data boundary.
Give each agent a revocable identity instead of sharing one powerful account.
Trace handoffs across agents so loops, duplication and conflicting instructions are visible.
Set circuit breakers, budgets and fleet-wide pause controls before autonomy expands.
Design an escalation contract that says when an agent asks another agent, a person or no one.
Govern the portfolio, not only each agent. Local optimization can create system-wide risk.
Practitioner playbook
Take the questions back to work
Use these reviews as an agenda for architecture, risk, readiness and governance conversations.
Data review
- Is the corpus fresh and bounded?
- Who audits duplication and staleness?
- Do shared definitions exist across systems?
Architecture review
- Which governed tools may the agent use?
- Where does memory live?
- What is the provider fallback and routing policy?
Trust review
- What can each identity access?
- Which decisions require a person?
- What is logged and when was the last red team?
Operations review
- Which quality and cost signals fire live?
- Can releases roll back immediately?
- Is every agent registered and stoppable?
People and governance review
- Who owns the outcome and budget?
- How will roles and workflows change?
- Who approves risk and answers third parties?
Workshop close
Turn the result into a decision
A strong debrief converts game events into owned gaps, planned controls and an honest residual-risk register.
- 1Read the outcome and ask what your real program scorecard rewards.
- 2Walk every hit as a named organizational gap, not just a game event.
- 3Read unequipped capabilities as residual risks and assign an owner.
- 4Examine what changed between phases and whether the change budget was planned.
- 5Map one concrete gap in each pillar to a current initiative and accountable person.
- 6Close by naming the control with no compensating alternative and verify that it is funded.
Working vocabulary
Name the system you govern
Shared language lets architects, operators, risk leaders and executives reason about the same agent system.
- Arena
- A shared workshop session. Players with the same mission face the same seeded threat sequence.
- Base
- The previous level's loadout carried forward into the next phase.
- Control plane
- The registry and operational layer used to deploy, observe, coordinate, pause and stop a fleet of agents.
- Determinism
- The same arena, mission and level produce the same threat order, making competition fair and debriefs analyzable.
- Hardening sprint
- The remediation period between phases, represented by limited health recovery.
- Hitting set
- A set of controls that intersects the counter list of every risk under review.
- Human-in-the-loop
- A designed handling path that transfers defined cases to an authorized person.
- Mission threat
- A failure mode specific to the chosen enterprise use case.
- Residual risk
- The exposure that remains after selected controls are applied.
- Singleton risk
- A risk with only one effective counter and therefore no compensating alternative.
- Swap
- Removing a carried-forward capability to evolve the stack between phases.
- Threat
- A real enterprise failure mode with a scenario, one or more controls and a transferable lesson.
Briefing complete
The threats are on schedule. Choose your stack.
Hold the data steward, architect, security engineer, operator and governor in one mind. Then test the system under pressure.