AI Agent Anti-Patterns & Best Practices Workbook

📋 What It Is

An 8-tab enterprise risk assessment workbook that catalogs 45 known failure modes across 14 categories of AI agent projects — and gives your team a structured, repeatable process to score them, mitigate them, and prove readiness at every phase gate from initial feasibility through post-production monitoring. This isn't a checklist you glance at once. It's an operational instrument with 257 live formulas that connect your risk scores to mitigation ownership, phase gate readiness, and an auto-generated executive Dashboard — so the assessment stays current as your project evolves.

Most agent initiatives don't fail because the team lacked talent. They fail because predictable anti-patterns — hallucination normalization, gold-plated reasoning, evaluation theater, context bloat from eager skill loading — show up late, after the prototype is praised and the budget is committed. This workbook forces early confrontation with those risks, when the cost of correction is still low.

Includes full lifecycle coverage with 6 phase gates (from Discovery/Feasibility through Post-Production Review) and deep treatment of the emerging Skills & Capability Architecture domain — including progressive skill loading, skill version control, and Agent Development Lifecycle (ADLC) integration patterns that reflect how enterprise agent platforms are actually being built in 2026.

👥 Who It's For

• AI/ML engineering leads evaluating architectural risk before decisions become structural and irreversible
• AI program managers overseeing agent initiative portfolios across multiple teams and use cases
• Security, GRC, and compliance reviewers conducting go/no-go assessments before production deployments
• AgentOps / SRE teams establishing operational baselines and monitoring thresholds for agent systems
• Platform and infrastructure architects validating that agent designs don't create hidden infrastructure debt
• Business stakeholders and sponsors who need a shared vocabulary for what can go wrong — and evidence that the team is managing it

Best used cross-functionally. Score as a team — disagreements about a score are where hidden risks surface.

⏱ When to Use It

The workbook maps to a 6-gate lifecycle. Use it at the gate that matches where your project stands:

• Discovery/Feasibility — before committing to an agent approach. Validate the use case, assess technical feasibility, and identify skill gaps before architecture work begins
• Architecture Review — when design patterns become structural and hard to reverse. Score early, while you can still change course
• Scope Expansion — before adding new tools, workflows, models, or agent capabilities. Each expansion reintroduces risk
• Pre-Production — the final quality gate before deployment. Surface unknown unknowns before they become production incidents
• Go-Live — deployment authorization. Verify all Critical mitigations are complete, rollback plans are tested, and monitoring is operational
• Post-Production Review — 30/60/90-day validation. Audit drift detection, cost actuals vs. projections, and make the scale/continue/retire decision with data
• Post-incident retrospectives — map incidents back to the anti-pattern registry. Was this a known failure mode you scored low? Recalibrate
• Quarterly health checks — agent systems drift. Scores change. New anti-patterns emerge. Re-assess periodically

📦 What It Produces

A completed assessment generates:

• Executive Dashboard — auto-calculated risk summary with overall risk profile, top 5 risks by score (deduplicated), phase gate readiness across all 6 gates, and mitigation progress tracking across P1/P2/P3 priorities
• Scored Risk Register — 45 anti-patterns scored by likelihood × impact, auto-classified into Critical (≥20), High (12–19), Medium (6–11), and Low (1–5) risk levels with conditional formatting
• Prioritized Mitigation Plan — 122 pre-built mitigations with owner assignments (8 RACI roles via dropdown), due dates, verification methods, and status tracking
• Phase Gate Readiness Report — 57 checkpoints across 6 gates with pass rates that auto-feed the Dashboard
• Best Practice Roadmap — 38 implementation patterns across Basic, Intermediate, and Advanced maturity tiers with distribution chart
• Shared Risk Vocabulary — a 27-term glossary covering agent-specific concepts (MCP, A2A, ADLC, Skill Composition, Progressive Loading) plus calibrated scoring scales

Functions as a repeatable risk discipline for agent engineering — not a one-time audit, but a living document that tracks your project's risk posture from feasibility through post-production.

🚀 How to Use It — Quickstart

• Step 1. Open the Scoring Guide tab. Walk through the Likelihood Scale (1–5), Impact Scale (1–5), and Risk Matrix with your team before anyone starts scoring. This step prevents the single biggest source of scoring inconsistency.
• Step 2. Switch to Risk Assessment. Fill in project metadata (Project, Agent, Assessed By, Date, Review Stage), then walk through all 45 anti-patterns as a team. For each: set Applicable (Yes/No/Partial), score Likelihood (1–5) and Impact (1–5). Risk Score and Risk Level auto-calculate.
• Step 3. Return to the Dashboard tab. Your overall risk profile, top 5 risks, phase gate readiness across all 6 gates, and mitigation progress are auto-generated. This is your executive summary.
• Step 4. Open Mitigation Actions. For your Critical and High risks, 122 pre-built mitigations are already mapped with verification methods. Assign owners using the 8-role dropdown, set due dates, and confirm P1/P2/P3 priority.
• Step 5. Complete the Phase Gate Checklist for your current gate. Mark each checkpoint as Pass, Fail, Conditional, or N/A. The Dashboard's Phase Gate Readiness auto-updates: ≥80% = Ready, ≥50% = Conditional, <50% = Not Ready.
• Step 6. Set a review cadence. Re-score after scope changes, incidents, or quarterly milestones. Use the Post-Production Review gate for 30/60/90-day validation cycles.

👁 Preview — What's Inside

📝 Version History