Sneeze It Digital Agency

One seat, one owner. No agent shares responsibility with another agent.

Why: Shared responsibilities create blame diffusion and tuning conflicts.

Failure mode: Two agents both handle client performance. Tuning one breaks the other.

Scope: Organization-wide

Every agent writes to its own shared state file. No agent reads data sources directly.

Why: Pre-computed shared state decouples scan timing from compile timing.

Failure mode: Orchestrator silently re-queries sources. Cannot tell what version of reality it saw.

Scope: Organization-wide

All external communications require founder approval before sending.

Why: Reputational damage from bad messages is hard to reverse.

Failure mode: Agent sends email with incorrect performance numbers. Client loses trust.

Scope: Organization-wide

File-based state is authoritative over AI memory.

Why: AI memory drifts. Files do not.

Failure mode: Agent acts on stale memory instead of canonical file. Wrong decisions.

Scope: Organization-wide

Performance Analyst reports patterns but never recommends actions.

Why: Recommendations without client context were ignored.

Failure mode: Analyst recommends budget increase for cash-strapped client.

Scope: Organization-wide

Retention agent overrides Sales agent when client is at risk.

Why: Sales expansion to at-risk client accelerates churn.

Failure mode: Sales proposes upsell to declining client. Client cancels.

Scope: Organization-wide

Only one agent (EA) has authority to send external communications.

Why: Multiple senders create duplicate and inconsistent messages.

Failure mode: Sales and Retention both email same client with contradictory messages.

Scope: Organization-wide

Each agent has documented role, ownership, and explicit non-ownership boundaries.

Why: Without boundaries, agents drift into overlapping responsibilities.

Failure mode: Two agents both track project status with conflicting updates.

Scope: Organization-wide

AI Call Center Manager manages 3 humans via daily data-driven Slack coaching.

Why: Data-driven daily coaching was not possible with a human manager at this scale.

Failure mode: If messages sound generic, human employees disengage within 3 days.

Scope: Organization-wide

Evaluator agent scores maturity. Learning agent implements. Evaluator re-scores.

Why: Self-improvement requires separated diagnosis and action.

Failure mode: Evaluator diagnoses correctly but implementer fails. Score stagnates.

Scope: Organization-wide

17 overnight agents pre-compute data. Morning briefing reads cached results.

Why: Eliminates 30+ min serial latency during founders most valuable hours.

Failure mode: Founder starts the day waiting for scans instead of acting.

Scope: Organization-wide

Agents coordinate via structured message bus with 5 message types and 3-exchange limit.

Why: Structured messaging makes coordination visible and bounded.

Failure mode: Without structure, agents coordinate through undocumented side channels.

Scope: Organization-wide

Critical alerts auto-escalate: founder, then COO after 48h, then Strategic agent.

Why: Unanswered critical alerts create financial risk.

Failure mode: Agent detects +139% overspend. Alert unanswered 16 days. Client overspends $1,348.

Scope: Organization-wide

Morning briefing compiles 10 parallel scanners into one unified document.

Why: 10 data sources cannot be queried sequentially in under 5 minutes.

Failure mode: One scanner failure blocks entire briefing without fault-tolerant architecture.

Scope: Organization-wide

If data is stale, flag it visibly. Never silently present old info as current.

Why: Stale data presented as current causes wrong decisions.

Failure mode: Briefing shows yesterdays spend as todays. Wrong budget decisions.

Scope: Organization-wide

If 3+ tasks from one person are overdue, flag as capacity pattern.

Why: Individual overdue = forgotten. Pattern = overwhelmed.

Failure mode: Manager assumes laziness. Actual problem is capacity. Problem worsens.

Scope: Organization-wide

Gave analyst write access to campaigns. Optimized for wrong metrics.

Why: Lacked client context. Technically correct, strategically wrong.

Failure mode: Decreased spend on clients strategic brand campaign. Client frustrated.

Scope: Organization-wide

Single shared state file caused 12 corruption events in month one.

Why: Concurrent writes from multiple agents created data races.

Failure mode: Two agents wrote simultaneously. One update lost. Manual cleanup required.

Scope: Organization-wide

Built message bus without embedding triggers in workflows. Zero transactions.

Why: Protocol described communication. No workflow included steps to use it.

Failure mode: 13 inbox files deployed. All empty. Dead infrastructure.

Scope: Organization-wide

Agent reported escalation overdue for 17 days without executing it.

Why: Spec treated as documentation, not executable logic.

Failure mode: Critical overspend detected. Escalation specified. Never executed. 17 days.

Scope: Organization-wide

Negative constraints improve AI message quality. Structural requirements degrade it.

Why: Telling AI what NOT to do produces natural variation.

Failure mode: Added example messages: quality dropped 8.4 to 8.2. Added ban rules: quality rose to 8.8.

Scope: Organization-wide

All external comms, pricing, contracts, hiring, firing are founder-only.

Why: Legal, financial, and relationship consequences AI cannot fully assess.

Failure mode: Agent agrees to terms violating margin floors. Agent sends wrong positioning.

Scope: Organization-wide

Emotional and relational domains remain human.

Why: AI cannot substitute for human connection, empathy, or presence.

Failure mode: AI-managed employee feels managed by a system. Engagement drops.

Scope: Organization-wide

AI writing must not sound like AI. No em dashes, no filler, no hedging.

Why: AI-sounding writing destroys trust with clients, team, and prospects.

Failure mode: Coaching message with ChatGPT patterns. Employee realizes manager is AI.

Scope: Organization-wide