Pattern: Sequential Pipeline | Team size: 5 agents
A chapter-by-chapter production line that turns the book concept into a finished manuscript with consistent narrative, examples, and technical accuracy. The sequential flow works because each stage depends on the previous one: vision and outline, then workflows, then code, then editing and cohesion.
Imagine a visionary marketer harnessing the power of Codex by OpenAI to reshape the entire marketing landscape. In this book, show—step by step—how Codex can be used not just to create ads, but to revolutionize strategy. Detail how Codex can automate customer research, build full-scale campaign blueprints, predict market trends, and even craft interactive customer experiences. Offer practical code snippets and workflows for each application—making Codex the marketer’s partner in creating a future where AI drives not just tactics, but entire marketing visions.
Create an agent team to produce a complete, chapter-by-chapter manuscript for a practical business/tech book titled “Codex Marketing: How OpenAI Codex Reinakes Strategy (Not Just Ads)” (working title; finalize in Stage 1). Use a Sequential Pipeline with five agents. Each agent must write specific deliverables to the exact file paths below. Every stage must explicitly read the prior stage outputs before producing its own. Do not skip dependencies. PROJECT NAME (for all paths): codex_marketing_book_factory GLOBAL CONSTRAINTS (apply to all agents) - Target reader: modern marketers, growth leads, and marketing ops who can run light code (Python/JS) and want AI-first strategy, not generic AI tips. - Voice: visionary but practical; step-by-step; no hype; show workflows, constraints, and QA steps. - Codex framing: treat “Codex by OpenAI” as the coding agent that helps build marketing systems (data pipelines, analysis, automation, interactive experiences). Acknowledge that product naming/capabilities may evolve; write examples using the general “OpenAI API” and “Codex-style coding assistant” pattern without making unverifiable claims about product packaging. - Must include: privacy, compliance, bias, safety, and measurement guardrails. - Manuscript structure: Front matter + 10–12 chapters + back matter. Each chapter must include (a) Why it matters, (b) Step-by-step system, (c) A “Codex workflow” playbook, (d) Code snippets section, (e) Pitfalls & QA checklist, (f) Mini case study. - All code must be runnable (best-effort) and include: dependencies, environment variables, brief run instructions, and safe placeholders (no real keys). - Collaboration mechanics: each agent must leave “Notes for next agent” and “Open questions / risks” sections at end of their main deliverable. Later agents must answer or resolve earlier open questions, or explicitly document why not. - Challenge mechanism: each agent must include a “Challenge & alternatives” subsection where they critique at least 3 assumptions from previous stage and propose improvements. - Length targets: total manuscript 45,000–65,000 words. Keep chapters balanced. PIPELINE OVERVIEW & DEPENDENCIES (do not violate) Stage 1 (Book Architect) must complete before Stage 2 begins. Stage 2 (Marketing Strategy Author) must complete before Stage 3 begins. Stage 3 (Workflow Engineer) must complete before Stage 4 begins. Stage 4 (Code Snippet Builder) must complete before Stage 5 begins. Stage 5 (Technical Editor & Consistency QA) must complete before final Synthesis/Review. DELIVERABLE FILES (write to these exact paths) - outputs/agent_teams_demo/codex_marketing_book_factory/00_brief.md - outputs/agent_teams_demo/codex_marketing_book_factory/01_outline.md - outputs/agent_teams_demo/codex_marketing_book_factory/02_strategy_chapters.md - outputs/agent_teams_demo/codex_marketing_book_factory/03_workflow_playbooks.md - outputs/agent_teams_demo/codex_marketing_book_factory/04_code_snippets.md - outputs/agent_teams_demo/codex_marketing_book_factory/05_tech_edit_and_consistency_report.md - outputs/agent_teams_demo/codex_marketing_book_factory/06_final_manuscript.md AGENT 1: Book Architect (Outline & Positioning) Objective: Define the book promise, audience, positioning, learning path, and a detailed chapter outline that moves from tactical uses to strategic transformation. Input: Business problem statement only. Output files to produce: 1) outputs/agent_teams_demo/codex_marketing_book_factory/00_brief.md (1,200–1,800 words) Required sections (use these exact headings): - Working Title + Subtitle (3 options) - One-Sentence Promise - Who This Book Is For / Not For - Reader Starting Assumptions (skills, tools, constraints) - The “Codex-as-Partner” Thesis (5 bullet claims; each must be testable) - Key Differentiators vs Other AI Marketing Books (8 bullets) - Ethical/Legal/Privacy Position (specific) - Learning Outcomes (10 outcomes, measurable) - Notes for next agent - Open questions / risks 2) outputs/agent_teams_demo/codex_marketing_book_factory/01_outline.md (1,800–2,400 words) Required structure: - Final Title + Subtitle (choose 1) - Table of Contents (10–12 chapters + intro + conclusion + appendices) - Chapter-by-Chapter Blueprint (for each chapter include): - Strategic goal (1–2 sentences) - Core concepts (5 bullets) - Step-by-step path (6–10 steps) - Required workflow artifact(s) (e.g., “Persona research pipeline”, “Campaign blueprint generator”) - Required code artifact(s) (e.g., “Python: survey clustering”, “JS: interactive experience widget”) - Measurement plan (3–5 metrics + instrumentation idea) - Mini case study premise (2–3 sentences) - Appendix Plan (at least 3 appendices; include “Prompt patterns”, “Compliance checklist”, “Starter repos structure”) - Challenge & alternatives (critique likely weak spots in the plan; propose fixes) - Notes for next agent - Open questions / risks Collaboration mechanics: - Write explicit “handoff instructions” for Agent 2: which chapters are most important, tone constraints, and how to maintain the thesis. AGENT 2: Marketing Strategy Author Dependency: Must read outputs/agent_teams_demo/codex_marketing_book_factory/00_brief.md and 01_outline.md before writing. Do not begin until Stage 1 is complete. Objective: Draft the strategy narrative chapters that explain how Codex reshapes marketing operating models: research, planning, experimentation, measurement, and governance. Output file to produce: - outputs/agent_teams_demo/codex_marketing_book_factory/02_strategy_chapters.md (18,000–26,000 words) Structure requirements: - Use the exact Table of Contents from 01_outline.md. - For each chapter, write 1,500–2,500 words and include these exact headings: - Chapter Summary (5 bullets) - Why This Changes Strategy (not tactics) - Step-by-Step System (numbered steps; align to outline) - Operating Model: People, Process, Tech (PPP) - Measurement & Experimentation Loop - Risks, Ethics, and Compliance Guardrails - Mini Case Study (narrative + outcomes) - What You Will Build Next (bridge to workflow engineer) - Include an Introduction (1,200–1,800 words) and Conclusion (800–1,200 words) using the same voice and thesis. - Add a recurring motif: “Codex as the marketing systems engineer” with concrete examples. Collaboration mechanics: - At the end of EACH chapter include: - “Notes for Workflow Engineer” (workflow artifacts needed, inputs/outputs, QA checks) - “Open questions / risks” - Include a “Challenge & alternatives” section near the end of the file where you critique at least 3 assumptions from the outline and propose changes; if proposing changes, specify how Agent 3 should adapt playbooks. AGENT 3: Workflow Engineer (Codex Playbooks) Dependency: Must read 00_brief.md, 01_outline.md, and 02_strategy_chapters.md before writing. Do not begin until Stage 2 is complete. Objective: Convert each chapter into executable, end-to-end marketer workflows (playbooks) that a Codex-like coding assistant can help implement. These should be usable even if the reader only has spreadsheets + basic scripting. Output file to produce: - outputs/agent_teams_demo/codex_marketing_book_factory/03_workflow_playbooks.md (10,000–16,000 words) Global playbook template (must be used for EVERY chapter as a separate playbook section): - Playbook Name - When to Use / When Not to Use - Inputs (data, tools, access, assumptions) - Outputs (artifacts; include file names the reader would produce) - Codex Prompt Pack (5–10 prompts; include role, constraints, and acceptance criteria per prompt) - Step-by-Step Workflow (10–20 steps) - QA Checks (accuracy, bias, privacy, compliance, reproducibility) - Failure Modes & Debug Plan - Time/Cost Estimate (rough) - Team Handoff (what marketing/ops/legal needs to review) - “Challenge & alternatives” (at least 2 alternative approaches) Additional requirements: - Create at least 8 reusable “workflow primitives” (e.g., “Data intake & cleaning”, “Insight synthesis”, “Hypothesis generator”, “Experiment design”, “Creative variant factory”, “Attribution sanity-checker”, “Trend signal detector”, “Interactive experience builder”). Define each primitive once, then reference it across playbooks. - Ensure each playbook clearly calls out which code snippets are required from Agent 4 (by a stable snippet ID you define, e.g., SNIPPET-PY-01). - End with: - Notes for Code Snippet Builder (prioritized list of snippet IDs with requirements) - Open questions / risks AGENT 4: Code Snippet Builder (Python/JS + APIs) Dependency: Must read 00_brief.md, 01_outline.md, 02_strategy_chapters.md, and 03_workflow_playbooks.md before writing. Do not begin until Stage 3 is complete. Objective: Produce runnable, copy-pasteable code snippets (Python and JavaScript/TypeScript) that implement the playbooks: customer research automation, campaign blueprint generation, trend modeling, and interactive experiences. Output file to produce: - outputs/agent_teams_demo/codex_marketing_book_factory/04_code_snippets.md (8,000–14,000 words) Mandatory constraints: - Do not hardcode secrets. Use environment variables (e.g., OPENAI_API_KEY). - Prefer standard libraries plus minimal dependencies; list dependencies per snippet. - Provide “How to run” steps for each snippet. - Include basic tests or sanity checks for at least 6 snippets. - Provide synthetic/sample data in-code where needed (tiny inline datasets) so snippets can run without external files. - Use generic “OpenAI API” patterns; avoid unverifiable product-specific endpoints. If you include LLM calls, wrap them in functions with clear docstrings and “replace with your provider’s SDK call” notes. Snippet catalog requirements: - Provide 14–20 snippets total, each with: - Snippet ID (matching Agent 3 references) - Purpose - Language (Python or JS/TS) - Dependencies - Environment variables - Code (fully included) - How to run - Expected output - Pitfalls & safeguards Coverage must include at least: - Data ingestion/cleaning pipeline for marketing datasets (Python) - Customer research automation: - Scraping-safe alternative (use provided text corpora / survey exports) - Topic modeling or clustering (Python) - Insight summarization pipeline with citations to input rows (Python) - Campaign blueprint generator: - From product brief to segmentation, positioning, channel plan, creative angles (Python) - Output to structured JSON + Markdown (Python) - Experiment design and measurement: - Power/sample size rough calculator OR experiment tracker template generator (Python) - UTM/link builder + event schema generator (Python or JS) - Trend prediction / market signal detection: - Time-series anomaly detection on synthetic data (Python) - Lightweight forecasting baseline (Python) - Interactive customer experience: - Simple web widget or Node/Express mini-app that serves an interactive quiz/product finder backed by rules + optional LLM call (JS/TS) - Conversation state handling and safety filters (JS/TS) - Governance: - PII redaction utility (Python) - Prompt logging + audit trail structure (Python) - Bias/quality checklist automation (Python) Collaboration mechanics: - Include a “Notes for Technical Editor” section: known limitations, where assumptions were made, and which snippets are most fragile. - Include “Challenge & alternatives”: for 3 key snippets, propose an alternate implementation (e.g., no-LLM baseline, different library) and tradeoffs. AGENT 5: Technical Editor & Consistency QA Dependency: Must read all previous outputs (00–04) before writing. Do not begin until Stage 4 is complete. Objective: Validate technical accuracy, ensure consistent terminology and narrative cohesion, tighten prose, and add guardrails (privacy, bias, compliance). Produce an actionable edit report AND assemble the final manuscript. Output files to produce: 1) outputs/agent_teams_demo/codex_marketing_book_factory/05_tech_edit_and_consistency_report.md (3,000–5,000 words) Required sections: - Executive Summary (10 bullets: biggest fixes) - Terminology & Naming Consistency (table: term, preferred usage, occurrences to fix) - Technical Accuracy Review (by snippet ID: pass/fail + fix notes) - Security/Privacy/Compliance Review (PII handling, consent, retention, SOC2-style practices, ad platform policies) - Bias & Fairness Risks (where they appear; mitigations) - Measurement Validity & Causal Claims Review (flag overclaims; propose rewrites) - Structure & Flow Issues (chapter transitions, repetition, missing prerequisites) - Required Revisions (prioritized checklist) - Optional Enhancements (nice-to-have) - Notes for final synthesis - Open questions / risks 2) outputs/agent_teams_demo/codex_marketing_book_factory/06_final_manuscript.md (45,000–65,000 words) Assembly requirements: - Produce a single cohesive manuscript with: - Title page - Copyright/Disclaimer page (include “not legal advice”) - Dedication (1–2 lines) - Table of Contents - Introduction - Chapters 1–N (from outline; consistent headings) - Conclusion - Appendices (from outline; include Prompt Patterns appendix with at least 12 patterns) - Glossary (20–30 terms) - Integrate: strategy content (Agent 2), playbooks (Agent 3), and code snippets (Agent 4) into each chapter in the specified chapter format: - Why it matters - Step-by-step system - Codex workflow playbook (condensed in-chapter; link to full playbook section if needed) - Code snippets (embed the most relevant; reference others by ID) - Pitfalls & QA checklist - Mini case study - Ensure every chapter includes at least: - 1 workflow prompt pack excerpt (3 prompts minimum) - 1 code snippet excerpt (or a clear reference + short excerpt) - 1 compliance/bias safeguard note - Maintain consistent formatting: - H2 for chapters - H3 for the required chapter headings - Code fences with language tags FINAL SYNTHESIS / REVIEW STEP (must be performed after Agent 5 completes) After writing 06_final_manuscript.md, perform a final integrated review inside the same file at the very end under heading “Synthesis Review (Post-Production)”. Include: - Manuscript readiness score (1–10) with justification - Top 12 remaining risks (technical, legal, narrative, usability) - 10-point launch checklist (what to verify before publishing) - Change log (high-level) describing what was altered during editing - If any open questions remain, list them with proposed resolutions Execution instruction: - Run the pipeline sequentially. Each agent must explicitly cite which prior files they read and which open questions they resolved.
Create your own AI agent team at Build Agents Store. Describe your business problem and get specialized agent teams with ready-to-use prompts for Claude Code.