AI Project Showcase: IT Influentials Agent

Section 1 — Product Overview

1.1 Product name and tagline

Name: IT Influentials Agent
Tagline: 12 specialized agents for B2B media strategy workflows
Current status: Active development — v1.0.4
First commit / project start: Between January 8 and March 2026 — derived from POC (dated Jan 8, 2026) via Plugin Clone Safety Checker; POC archived March 2026 [CLAUDE NOTE: inferred from POC release notes dates and ARCHIVED.md]

1.2 What it is

IT Influentials Agent is a WordPress plugin that implements a 12-agent AI system for B2B media strategy workflows. Each agent specializes in a different aspect of the consulting workflow — from intake conversation and research through strategy, audience analysis, content planning, revenue modeling, technology assessment, financial projections, project planning, quality assurance, and final report generation. The system uses an orchestrator to sequence agents through a defined workflow with human checkpoints at Strategy and Report phases. Agents draw from a file-backed knowledge base (consulting frameworks, industry benchmarks, experience proof points) and 14 prompt templates, all importable into the WordPress database for admin editing. The plugin provides a full admin dashboard for project management, agent execution, document generation, and prompt library management.

1.3 What makes it meaningfully different

• 12 specialized agents, not a single chatbot: Each agent has a distinct role, system prompt, and output schema — Research, Strategy, Prompt Engineer, Audience, Content, Revenue, Technology, Financial, Project, Report, QA, and Intake
• Orchestrated consulting workflow: Defined execution sequence with human review checkpoints at Strategy and Report phases — not just parallel API calls
• File-backed knowledge base with DB import: Consulting frameworks, benchmarks, and proof points are authored as Markdown files, then imported into WordPress database for runtime use and admin editing via Prompt Library UI
• Dual prompt source architecture: Agents use DB-stored prompts (admin-editable) with fallback to PHP-defined prompts — enabling prompt iteration without code changes
• Optional n8n integration: ITI_Workflow_Adapter routes through n8n when available, with fallback to direct Claude API

1.4 Platform and deployment context

Platform: WordPress plugin (GPL v2+)
Deployment: Self-hosted WordPress (WP 6.0+, PHP 8.0+)
Primary interface: WordPress admin dashboard with 7 admin pages (Dashboard, Projects, New Project, Knowledge Base, Documents, Prompt Library, Settings)

---

Section 2 — User Needs and Problem Statement

2.1 Target user

Primary user: IT Influentials consultants running B2B media strategy engagements
Secondary users: Business analysts, content strategists, and revenue planners within client organizations
User environment: WordPress admin panel on ITI’s internal infrastructure

2.2 The problem being solved

B2B media consulting engagements require deep analysis across multiple domains — audience, content, revenue, technology, financial projections, and project planning. Manually producing these analyses is time-intensive (weeks) and requires expertise across many specializations. A single AI chatbot cannot reliably produce the depth and structured output needed across all these domains simultaneously.

2.3 Unmet needs this addresses

Need	How the product addresses it	Source of evidence
Multi-domain consulting analysis	12 specialized agents each focusing on one domain with tailored prompts and output schemas	includes/agents/ — 13 agent classes
Structured consulting workflow	Orchestrator sequences agents: Intake → Research → Strategy → [checkpoint] → Domain agents → Report → QA → [checkpoint]	class-orchestrator.php workflow logic
Reusable consulting frameworks	Knowledge base files (consulting-framework.md, industry-benchmarks.md, experience-proof-points.md) injected into agent context	knowledge-base/ directory
Iterative prompt improvement	Prompt Library admin UI with DB-stored prompts, versioning, and usage tracking	templates/admin/prompts.php, class-knowledge-importer.php
Document generation	Report agent produces structured JSON; Document Generator creates HTML (with optional TCPDF/PHPWord) deliverables	class-document-generator.php, class-report-agent.php

2.4 What users were doing before this existed

Manually conducting each analysis domain separately, synthesizing findings in documents, and iterating through multiple rounds of revision. The ITI Agent Plugin POC (v1.0.7, now archived) was the predecessor, proving the multi-agent concept before this production version was built.

---

Section 3 — Market Context and Competitive Landscape

3.1 Market category

Primary category: AI-powered consulting automation / multi-agent business intelligence
Market maturity: Early — multi-agent systems for consulting are emerging but most are SaaS platforms, not self-hosted WordPress plugins
Key dynamics: ⚡ Growing interest in “AI agents” for business workflows; most solutions are cloud-only SaaS

3.2 Competitive landscape

Product / Company	Approach	Strengths	Key gap this project addresses	Source
⚡ McKinsey Lilli / BCG proprietary	Enterprise AI assistants	Deep consulting methodology, massive training data	Proprietary, enterprise-only, not customizable or self-hosted	[CLAUDE NOTE: inferred from market knowledge]
⚡ Custom GPTs / Claude Projects	Single-persona AI assistants	Easy setup, conversational	Single agent, no workflow orchestration, no structured multi-domain output	[CLAUDE NOTE: inferred]
⚡ LangChain/CrewAI agent frameworks	Developer tools for multi-agent	Flexible orchestration, Python ecosystem	Require custom development, no WordPress integration, no consulting-specific knowledge	[CLAUDE NOTE: inferred]

3.3 Market positioning

Internal tool that demonstrates ITI’s advanced AI orchestration capabilities. Positioned as a showcase of how multi-agent systems can automate complex consulting workflows while maintaining human oversight at critical decision points. The WordPress deployment model is unusual — most competitors are SaaS or Python-based.

3.4 Defensibility assessment

Knowledge base curation: Consulting frameworks, industry benchmarks, and experience proof points represent proprietary consulting methodology. Prompt engineering: 14 agent prompt templates (8,000+ lines combined) encode deep domain expertise in structured format. Architecture: 12-agent orchestration with checkpoints and dual prompt sources (DB + PHP) creates a system that’s non-trivial to replicate. [CLAUDE NOTE: defensibility inferred from codebase complexity and knowledge base depth]

---

Section 4 — Requirements Framing

4.1 How requirements were approached

Evolved from the ITI Agent Plugin POC (archived March 2026) which proved the multi-agent concept. Production version refined the architecture with proper prefixing (ITI_Agent_ vs ITI_Agent_POC_), added Prompt Library UI, and cleaned up integration points. Testing guide (391 lines) documents validation approach.

4.2 Core requirements

1. 12 specialized AI agents with distinct roles and structured JSON output schemas
2. Orchestrator-managed workflow with defined execution sequence
3. Human review checkpoints at Strategy and Report phases
4. File-backed knowledge base with WordPress database import
5. Admin-editable Prompt Library with versioning and usage tracking
6. Project management (CRUD) with engagement type classification
7. Document generation (HTML with optional TCPDF/PHPWord)
8. Intake conversational agent for client discovery
9. Full admin dashboard with 7 pages
10. Optional n8n workflow integration via ITI_Workflow_Adapter

4.3 Constraints and non-goals

Hard constraints: WordPress 6.0+, PHP 8.0+; admin-only access (no public-facing interface); requires mysqli, json, curl extensions
Explicit non-goals: Public-facing shortcodes are NOT implemented (admin-only) — contrast with the POC which had user-facing shortcodes

4.4 Key design decisions and their rationale

Decision	Alternatives considered	Rationale	Evidence source
12 specialized agents vs single all-purpose agent	Single Claude call with massive prompt, fewer agents with broader scope	Each agent can have focused prompts, structured JSON output schemas, and independent iteration	13 agent class files in includes/agents/
File → DB import for knowledge/prompts	DB-only, file-only	Markdown files are version-controllable and AI-tool-friendly; DB storage enables admin editing without code deploys	class-knowledge-importer.php
Admin-only (no shortcodes)	User-facing shortcodes (as in POC)	Production version focused on consultant workflow, not end-user self-service	No add_shortcode in iti-agent.php
Dual prompt source (DB preferred, PHP fallback)	DB-only, PHP-only	Enables prompt iteration via admin UI while maintaining working defaults	class-context-manager.php build_agent_context()
Optional TCPDF/PHPWord for documents	Always-HTML, mandatory PDF	Graceful degradation — HTML output always works, richer formats when libraries available	class-document-generator.php

---

Section 5 — Knowledge System Architecture

5.1 Knowledge system overview

KB type: File-backed Markdown → WordPress database import, with optional Pinecone vector RAG and Tavily web search
Location in repo: knowledge-base/ (3 knowledge files + 14 prompt templates)
Estimated size: 3 knowledge files (~1,376 lines) + 14 prompt templates (~9,267 lines) = ~10,643 lines of curated content

5.2 Knowledge system structure

knowledge-base/
├── consulting-framework.md          # 391 lines — 4-pillar consulting model
├── experience-proof-points.md       # 442 lines — case-style proof points
├── industry-benchmarks.md           # 543 lines — revenue/audience/content/tech benchmarks
└── prompt-templates/
    ├── intake-agent-prompt.md        # 202 lines
    ├── research-agent-prompt.md      # 315 lines
    ├── strategy-agent-prompt.md      # 410 lines
    ├── prompt-engineer-agent-prompt.md # 369 lines
    ├── audience-agent-prompt.md      # 831 lines
    ├── content-agent-prompt.md       # 1,037 lines
    ├── revenue-agent-prompt.md       # 1,025 lines
    ├── technology-agent-prompt.md    # 1,258 lines
    ├── financial-agent-prompt.md     # 600 lines
    ├── project-agent-prompt.md       # 720 lines
    ├── report-agent-prompt.md        # 708 lines
    ├── qa-agent-prompt.md            # 602 lines
    ├── synthesis-agents-prompt.md    # 1,191 lines (combined: financial, project, report, QA)
    └── domain-agents-prompt.md       # 799 lines (legacy/reference bundle)

5.3 Knowledge categories

Category	Files / format	Purpose	Update frequency
Consulting framework	consulting-framework.md (Markdown)	4-pillar model, engagement models, deliverable templates	As methodology evolves
Experience proof points	experience-proof-points.md (Markdown)	Case-style evidence for revenue, audience, tech, roles	As new engagements complete
Industry benchmarks	industry-benchmarks.md (Markdown)	Revenue, audience, content, and technology benchmarks for B2B media	Quarterly [CLAUDE NOTE: inferred]
Agent prompt templates	14 Markdown files in prompt-templates/	System prompts with role definitions, output schemas, behavioral rules	As agents are refined

5.4 How the knowledge system was built

1. Consulting framework authored from ITI’s consulting methodology (4-pillar model)
2. Experience proof points compiled from real engagement outcomes
3. Industry benchmarks aggregated from B2B media industry data
4. 14 agent prompt templates crafted with detailed role definitions, output JSON schemas, and behavioral guardrails
5. class-knowledge-importer.php handles file → database import: maps knowledge files to categories (framework, experience, benchmarks); extracts individual agent prompts from synthesis-agents-prompt.md via regex section headers
6. Activation hook calls reimport_all() which deletes existing knowledge/prompt rows and re-imports from files
7. Prompt Library admin UI (templates/admin/prompts.php) enables runtime editing, versioning, and usage tracking

5.5 System prompt and agent configuration

System prompt approach: Dual source — DB-stored prompts preferred (admin-editable via Prompt Library), PHP $this->system_prompt as fallback
Key behavioural guardrails: Each agent has role-specific guardrails; QA agent runs six-category quality review; human checkpoints at Strategy and Report
Persona / tone configuration: Each agent has 12 distinct agent personas for domain-specific reasoning (e.g., Research = “B2B media research analyst”, Strategy = “B2B media strategy consultant”)
Tool use / function calling: No formal LLM tool use; Research agent has Tavily web search pre-injected; Pinecone query_knowledge_base() exists in base agent class but is not called by any agent in current code

---

Section 6 — Build Methodology

6.1 Development approach

Evolved from the ITI Agent Plugin POC (archived March 2026) to production version. Prefix migration from ITI_Agent_POC_ to ITI_Agent_. Plugin Clone Safety Checker was built specifically to validate the clone process. Development used Cursor with AI assistance.

6.2 Build phases

Phase	Approximate timeframe	What was built	Key commits or milestones
POC (archived)	NOT FOUND — add manually	12-agent system with user-facing shortcodes, v1.0.7	Proved multi-agent concept
Production clone	NOT FOUND — add manually	Clean prefix migration, admin-only focus	iti-agent-plugin/ created from POC
v1.0.0	NOT FOUND — add manually	Initial production release	README changelog entry
v1.0.4	NOT FOUND — add manually	Current version — Prompt Library, refinements	Plugin header + constant

6.3 Claude Code / AI-assisted development patterns

Cursor-assisted development. The Plugin Clone Safety Checker (plugin-clone-checker/) was created as a development tool specifically for validating plugin cloning operations — scanning for leftover original prefixes across 19 check categories.

6.4 Key technical challenges and how they were resolved

Challenge	How resolved	Evidence
POC to production migration	Built Plugin Clone Safety Checker; systematic prefix rename	plugin-clone-checker/ tool in same iti-agents/ directory
Prompt template management at scale (14 files)	Knowledge Importer with file → DB import; synthesis-agents-prompt.md extraction via regex	class-knowledge-importer.php
Maintaining prompt editability without code deploys	Dual prompt source: DB preferred with PHP fallback; Prompt Library admin UI	class-context-manager.php, templates/admin/prompts.php
Pinecone integration wired but unused	query_knowledge_base() in base agent class — vector RAG ready but no agent currently calls it	class-base-agent.php

---

Section 7 — AI Tools and Techniques

7.1 AI models and APIs used

Model / API	Provider	Role in product	Integration method
Claude (configurable model)	Anthropic	All 12 agent AI calls	PHP wp_remote_post to https://api.anthropic.com/v1/messages, anthropic-version: 2023-06-01
Tavily Search API	Tavily	Research agent web search	PHP wp_remote_post to https://api.tavily.com/search
Pinecone Vector Database	Pinecone	Semantic knowledge retrieval (wired but not actively called)	PHP wp_remote_post to Pinecone host endpoints
Marked.js	jsDelivr CDN	Markdown rendering in admin UI	CDN script enqueue

7.2 AI orchestration and tooling

Tool	Category	Purpose
ITI_Agent_Orchestrator	Custom orchestrator	Sequences 12 agents through defined workflow with checkpoints
ITI_Agent_Context_Manager	Context assembly	Builds agent context from DB prompts, knowledge, and prior agent outputs
ITI_Workflow_Adapter (optional)	n8n routing	Routes through n8n consulting-pipeline when available; falls back to direct Claude
ITI_Agent_Document_Generator	Output formatting	Generates HTML deliverables (optional TCPDF/PHPWord for PDF/Word)

7.3 Prompting techniques used

• [x] System prompt engineering with role-specific personas (12 distinct agents)
• [x] Structured JSON output schemas (each agent defines expected output structure)
• [x] Context injection from knowledge base (consulting framework, benchmarks, proof points)
• [x] Prior agent output chaining (later agents receive earlier agents’ outputs as context)
• [x] Human-in-the-loop checkpoints (Strategy and Report phases)
• [x] Prompt versioning and admin editing via Prompt Library
• [ ] Few-shot examples
• [ ] Chain-of-thought prompting (not explicitly)
• [ ] Tool use / function calling

7.4 AI development tools used to build this

Tool	How used in build
Cursor	Primary development environment; AI-assisted code generation
Claude (via Cursor)	Code generation, prompt template authoring, architecture decisions
Antigravity	Autonomous test execution, browser QA, visual regression testing — used per global CLAUDE.md tool lane
Plugin Clone Safety Checker	Validated prefix migration from POC to production

---

Section 8 — Version History and Evolution

8.1 Version timeline

Version / Phase	Date	Summary of changes	Significance
POC (v1.0.7)	Pre-March 2026	12-agent system with user-facing shortcodes, ITI_Agent_POC_* prefix	Proof of concept — archived March 2026
v1.0.0	NOT FOUND — add manually	Initial production release with ITI_Agent_* prefix, admin-only	README changelog
v1.0.4	NOT FOUND — add manually	Current version — Prompt Library, refinements	Plugin header + constant

8.2 Notable pivots or scope changes

Major pivot from user-facing shortcode interface (POC) to admin-only consultant tool (production). The POC had 6 shortcodes ([iti_agent_poc], [iti_agent_poc_projects], etc.) for logged-in users; the production version removed all public-facing shortcodes and focused entirely on the admin workflow.

8.3 What has been cut or deferred

• User-facing shortcodes (present in POC, removed in production)
• Pinecone vector RAG (wired in base agent class but no agent calls it)
• brief column referenced in orchestrator n8n path but not in DB schema
• Context format fields (urgency_level, key_challenges, current_state) referenced in format_project_context_injection but not in projects table schema
• README/TESTING-GUIDE version drift (still say 1.0.0 while code is 1.0.4)

---

Section 9 — Product Artifacts

9.1 Design and UX artifacts

Artifact	Path	Type	What it shows
Admin CSS	assets/css/admin.css	CSS (708 lines)	Full admin dashboard styling
Admin JS	assets/js/admin.js	JS (431 lines)	AJAX interactions, project management, agent execution UI

9.2 Documentation artifacts

Document	Path	Type	Status
README	README.md	User/developer docs (151 lines)	Partially outdated (says v1.0.0)
Testing Guide	TESTING-GUIDE.md	Test documentation (391 lines)	Partially outdated (says v1.0.0)
ITI Agents CLAUDE.md	../documentation/CLAUDE.md	AI context document (66 lines)	Current

9.3 Data and output artifacts

Artifact	Path	Description
Consulting framework	knowledge-base/consulting-framework.md	391-line 4-pillar consulting methodology
Industry benchmarks	knowledge-base/industry-benchmarks.md	543-line B2B media benchmarks
Experience proof points	knowledge-base/experience-proof-points.md	442-line case-style proof points
14 prompt templates	knowledge-base/prompt-templates/*.md	~9,267 total lines of agent prompt templates

---

Section 10 — Product Ideation Story

10.1 Origin of the idea

The ITI Agent system evolved from the Executive Advisor plugin concept into a multi-agent architecture that could replicate the depth and breadth of a full consulting team. The idea was to automate the structured consulting workflow that ITI performs for B2B media clients — breaking it into specialized AI agents that each handle one domain of expertise.

10.2 How the market was assessed

NOT FOUND — add manually

10.3 The core product bet

That 12 specialized AI agents, each with focused prompts and structured output schemas, would produce higher-quality consulting deliverables than a single general-purpose AI — and that an orchestrated workflow with human checkpoints would maintain the quality control needed for professional consulting output.

10.4 How the idea evolved

Began as an Executive Advisor plugin, evolved into a POC with 12 agents and user-facing shortcodes (v1.0.7), then pivoted to an admin-only production tool when the user-facing interface proved premature. The POC was archived in March 2026, and the production version focused on the consultant workflow rather than client self-service. The Express Agents plugin was spun off as a lightweight 5-agent variant for quicker engagements. [CLAUDE NOTE: evolution inferred from ARCHIVED.md, CLAUDE.md, and codebase comparison]

---

Section 11 — Lessons and Next Steps

11.1 Current state assessment

What works well: 12-agent architecture with distinct roles and output schemas; file-backed knowledge with DB import; Prompt Library admin UI; orchestrator with human checkpoints; comprehensive admin dashboard
Current limitations: Pinecone RAG wired but not used; README/Testing Guide outdated (1.0.0 vs 1.0.4); brief column missing from DB schema (n8n path likely inactive); some context format fields not in DB schema; domain-agents-prompt.md not imported by Knowledge Importer
Estimated completeness: 75% [CLAUDE NOTE: inferred — core multi-agent workflow functional, but RAG integration incomplete and documentation outdated]

11.2 Visible next steps

1. Update README and TESTING-GUIDE to reflect v1.0.4
2. Activate Pinecone RAG in agent workflows (currently wired but unused)
3. Add brief column to projects table or fix orchestrator n8n path
4. Align format_project_context_injection fields with actual DB schema
5. Import domain-agents-prompt.md and standalone financial/project/report/qa prompt files
6. Migrate to ITI Shared Library components where applicable
7. Add document export to admin UI (currently AJAX-only)

11.3 Lessons learned

_Manual input required — this section cannot be populated automatically._

---

Section 12 — Claude Code Validation Checklist

• [x] Every placeholder has been replaced or marked NOT FOUND
• [x] All externally-sourced competitive data is marked with ⚡
• [x] All inferences are marked with [CLAUDE NOTE]
• [x] Version history is derived from actual git log where available
• [x] Knowledge system paths reflect real directory structure
• [x] AI tools are confirmed from code/config, not guessed
• [x] Section 11.3 is left blank for manual input
• [x] Document header shows today’s date and files examined

---

Sources Examined

File / Path	What it contributed
iti-agents/documentation/CLAUDE.md (66 lines)	Project overview, features, related projects
iti-agent-plugin/iti-agent.php (886 lines)	Plugin identity, version, constants, AJAX actions, admin menu, activation
iti-agent-plugin/README.md (151 lines)	Features, workflow, API list, requirements
iti-agent-plugin/TESTING-GUIDE.md (391 lines)	Testing approach, validation checklist
iti-agent-plugin/includes/class-database.php (1,146 lines)	Database schema, 7 tables
iti-agent-plugin/includes/class-orchestrator.php (493 lines)	Workflow sequencing, agent execution
iti-agent-plugin/includes/class-context-manager.php (485 lines)	Context assembly, dual prompt source
iti-agent-plugin/includes/class-knowledge-importer.php (379 lines)	File → DB import, prompt extraction
iti-agent-plugin/includes/class-claude-api.php (300 lines)	Claude API integration
iti-agent-plugin/includes/class-tavily-api.php (226 lines)	Tavily web search
iti-agent-plugin/includes/class-pinecone-api.php (346 lines)	Pinecone vector RAG
iti-agent-plugin/includes/class-api-handler.php (111 lines)	API routing, n8n adapter
iti-agent-plugin/includes/class-document-generator.php (433 lines)	Document generation
iti-agent-plugin/includes/agents/class-base-agent.php (261 lines)	Base agent class
iti-agent-plugin/includes/agents/*.php (12 agent classes)	Individual agent roles and prompts
iti-agent-plugin/knowledge-base/*.md (3 files, ~1,376 lines)	Consulting knowledge
iti-agent-plugin/knowledge-base/prompt-templates/*.md (14 files, ~9,267 lines)	Agent prompt templates
iti-agent-plugin/templates/admin/*.php (7 templates)	Admin UI pages
iti-agent-plugin/assets/css/admin.css (708 lines)	Admin styling
iti-agent-plugin/assets/js/admin.js (431 lines)	Admin interactions