Paul Barrick

Profile

I have been building agent-orchestrated software delivery systems since the ground floor of the practice — starting in early 2024, before mainstream agentic frameworks existed, with a custom Claude + MCP harness in our enterprise codebase. Three years on, I lead the agentic enablement function at MyComplianceOffice: a production agent platform that coordinates 8+ specialized agents, integrates with Jira / Confluence / GitLab via MCP, runs four-phase human-in-the-loop pipelines, and ships delivery packages that engineers actually merge.

My approach: agents are operational systems, not demos. Every agent in our stack has explicit stage contracts, observability hooks, cost tracking, resumability across sessions, and policy-enforced safety guardrails. The harness layer is provider-abstracted so the org isn't coupled to a single LLM vendor — Claude today, anything tomorrow.

Core Strengths

What I've Built (Selected — MyComplianceOffice / Fairwords)

Custom agent platform

• Planner v3.0 — a four-phase HITL delivery orchestrator with mandatory phase gates, Phase.Stage numbering, Confluence/Jira integration, and stage-contract enforcement. Generates structured delivery packages (PRDs, task breakdowns, complexity analysis, test plans). v1 → v3 evolution from April 2024 → April 2026.
• PR Orchestrator — multi-agent coordination for cross-repository code review. Delegates to six specialized sub-agents (security, backend, UI, performance, cost, code quality) with weighted PR scoring (security 30%, code quality 20%, testing 20%, cost 10%, …) and critical-issue caps.
• Implementer Agent — hierarchical execution agent with discovery cache, phase resumability, per-task artifact structure, structured blocker capture, and automated standup reporting. Treats execution like a real operational system, not a one-shot prompt.
• Specialized agents — analyzer (codebase intelligence), task-orchestrator (multi-task sequencing), ui-orchestrator, detection-expert, cost-tracker. Each registered through a unified agent framework with dual Claude Code + Kilo Code compatibility.

Custom harness layer (agentic-wrapper)

• Built a provider-abstracted harness via MCP — Taskmaster-AI for planning, Context7 for codebase analysis, Atlassian HTTP MCP for Jira/Confluence (read-only, audit-friendly), Figma + GitLab as needed.
• Unified .mcp.json as single source of truth for all MCP servers; AGENTS.md as authoritative source for agent registration across runtimes.
• Environment & secrets isolation — wrapper scripts (scripts/with-env.sh, scripts/with-secret.sh) load secrets without exposing them to LLM context.
• Repo started 2024. Year-over-year evolution from agent stubs to production-grade execution framework — the receipts are in the git history.

Custom context engineering

• Context library in docs/system/context/ — repository architecture, business-domain knowledge base (compliance communications regulatory requirements), cross-repo orchestration patterns. Every agent draws context from a curated source, not raw codebase searches.
• Pipeline configurations in docs/system/pipeline-configs/ — planner-pipeline-config (4-phase HITL with stage contracts, resumability), pr-scoring-algo, test-pipeline-config. Pipelines are declarative and versioned.
• Protocol library — Playwright MCP, GraphQL debugging, structured logging. Agents follow these protocols; humans audit them.

Production characteristics

• G9 two-tier observability hooks — pre-task and post-phase, full execution audit trails. Every run is reviewable.
• Cost tracking — automated daily monitoring (cost-monitor:start, cost:summary) so token spend is visible per-team and per-pipeline.
• Safety guardrails — ASCII-only source-file rule, critical-issue caps on PR scoring, mandatory HITL gates between phases. Compliance-friendly defaults.
• Session resumability protocol — halt/resume gates with task-progress preservation across interruptions. Long-running pipelines survive restarts.
• Test harness — unified test pipeline config with per-task artifact structure.

Strategic transformation work

• Led the org-wide shift from traditional PI planning to agent-orchestrated delivery with humans in the loop. Sprint preparation cycle time dropped, consistency across teams went up.
• Introduced model-agnostic agent design — outcomes decoupled from specific LLM providers. We can swap Claude → GPT → Gemini at the harness layer without touching agent definitions or pipeline configs.
• Partnered with engineering, product, and leadership to scale AI-assisted workflows safely across the eComms organization, an environment with strict regulatory and audit constraints.

Professional Experience

MyComplianceOffice (formerly Fairwords, acquired 2023)

VP of Engineering — eComms Platform & Head of Agentic Enablement

2021 – Present

Hired as Director of Engineering at Fairwords; promoted to VP prior to the acquisition. Retain platform and organizational leadership for the eComms division and own the company-wide agentic enablement function.

Agentic enablement (lead):

• Designed and shipped the agent platform described above — planner / PR-orchestrator / implementer / 5+ specialized agents, harness layer, context library, pipeline configs.
• Lead architect on the model-agnostic harness, MCP integration strategy, and HITL gate design.
• Partner with security, compliance, and legal to ensure agentic workflows meet the same audit and governance bar as the rest of the eComms platform.

eComms platform leadership:

• Own all technology decisions for the eComms division — architecture, platform evolution, compliance alignment.
• Lead three multinational, fully remote engineering teams (up to 21 engineers, QA, technical leads).
• Architectural lead for ingestion, archival, analytics, and compliance workflows.
• Senior engineering sponsor for major platform initiatives and cross-team architectural decisions.

Agentic stack: Claude Agent SDK · MCP (Taskmaster-AI, Context7, Atlassian HTTP, Figma, GitLab) · custom harness · multi-agent orchestration
Platform stack: Node.js · TypeScript · Angular · GraphQL · MongoDB Atlas · Apache Kafka · AWS · Docker

Paya

Tech Lead

2019 – 2021

• Built and scaled a payment-gateway platform with high reliability and security requirements.
• Owned local dev environments, CI/CD pipelines, cloud deployment workflows.
• Designed integrations and asynchronous pipelines across multiple systems.

Technologies: Node.js · AWS (Lambda, SQS) · Serverless Framework · React · Ember · Docker

Intengo

Senior Full-Stack Engineer

2012 – 2019

• Designed and shipped enterprise analytics platforms used for large-scale market research.
• Led architectural migration from monolith/LAMP to MEAN-based microservices.
• Implemented GraphQL APIs, WebSockets, and real-time data streaming.

Technologies: Angular · Node.js · Express · MongoDB · GraphQL · Docker · AWS · Laravel

Imperative Design, Inc.

Founder / Lead Engineer

2007 – Present

• Built and shipped multiple products and client systems from scratch.
• Translated ambiguous requirements into working software for direct clients.
• Currently the home of Optimus — an agent-native engineering platform with its own production-grade nexus (data layer), test-driver (verification layer), and harness (inference layer). Public companion site at paul.barrick.dev/optimus.

Education

North Georgia College & State University
Bachelor of Marketing · Associate Degree in Spanish · Studied abroad in Costa Rica.

Languages

Fluent in Spanish and Romanian.