Claude-Based Autonomous PR & Journalist Relationship Agent Conversational, Self-Learning, Human-Review Gated Objective Build a Claude-based AI agent that operates as a fully autonomous PR associate in the advertising / media intelligence industry. This agent must: Learn our company, positioning, research, and content Monitor reporters and competitors continuously Generate unique pitch ideas Hold real conversations with journalists Nurture relationships over time Track all outreach and engagement Only escalate to humans when: Reporter wants to interview us Reporter requests comment Coverage opportunity is confirmed We want the agent to operate autonomously — but with a configurable human review gate before messages go out (for now). Core Functional Requirements 1️⃣ Industry Knowledge Memory Layer The agent must maintain structured memory of: Our positioning Spokespeople + expertise areas Podcast episodes Blog posts Original research Quotes we have previously given Topics we can credibly comment on We need: A secure upload mechanism for new research/content Agent must: Ingest it Index it Generate derivative story angles Propose pitch concepts from it 2️⃣ Reporter Monitoring Engine The agent must: Continuously monitor target reporters Track: Recent articles Topic themes (last 30–90 days) Tone shifts Who they quote Publishing cadence Summarize each reporter’s recent output Detect emerging patterns Dashboard must show: Reporter name Outlet Recent topics covered Last 5 article summaries Suggested pitch angles tied to those themes 3️⃣ Competitor Narrative Monitoring The agent must track: Competitor announcements Competitor blog posts Competitor podcast appearances Messaging shifts Claims being made Output: Narrative gaps Overused talking points Opportunities to counter-position Trend alerts 4️⃣ Pitch Synthesis Engine Must combine: Reporter’s recent writing Competitor positioning Our content/research Industry momentum Output weekly: Ranked pitch ideas Rationale for why that reporter Supporting data angle Suggested headline Suggested quote draft 5️⃣ Autonomous Conversational Outreach Critical requirement: The agent must: Initiate outreach Reply to journalist responses Hold ongoing conversation threads Ask clarifying questions Provide supporting materials Suggest follow-ups Only escalate to human when: Interview requested Quote approval needed High-value opportunity confirmed For now: All outbound messages must require human approval We want the option to later allow automated nurture messages without approval 6️⃣ Weekly Nurture System Agent must: Maintain warm cadence (without spam) Send relevant insights periodically Share useful content tied to reporter’s beat Avoid repetition Track engagement history 7️⃣ Reporting & Monitoring Dashboard Must include: Outreach Tracking Who was contacted When Message status Open rate (if available) Replies Follow-up schedule Reporter Monitoring What they’ve written recently Article summaries Beat trend visualization Performance Response rates Pitch type performance Reporter engagement score 8️⃣ Guardrails No fabricated facts Citations required internally Confidence scoring Stop conditions after non-response Frequency caps Technical Expectations Claude (Anthropic SDK) required Persistent memory storage Structured JSON outputs Conversation thread memory CRM-style contact schema Dashboard UI (lightweight acceptable)