FrankX.AI

AI agent playbooks

From a chat window to agents that ship real work.

Five steps, in the order I would take them again: pick the machine and the stack, build one agent that works, package it as skills, run a crew with quality gates, and architect it for production. Every link is a working guide, tool, or open repository.

Step 01

Set up the machine and the stack

What should I run my agents on, and with which tools?

Pick the hardware, the model, and the tools before you write a single prompt. Most agent failures start as stack decisions nobody made on purpose.

  1. AI Computer Field GuideField guideRTX 5090 workstations, cloud GPU rental, AI mini PCs, DGX Spark, and private AI factory setups for founders and creators.
  2. LLM HubComparisonCompare model providers, pricing, and sources, then evaluate them against your own workload.
  3. Agent HubComparisonClaude Code, Codex, Cursor, Antigravity, LangGraph, CrewAI, and the Agent SDKs: what each is for and what to watch.
  4. The FrankX StackLibraryThe templates, agents, infrastructure, databases, and dev tools I actually use and recommend.

Step 02

Build your first agents and skills

How do I go from a chat window to an agent that does real work?

One working agent beats ten demos. Start with a guided build, then package what works as reusable SKILL.md modules.

  1. Build Your First AI AgentWorkshopA 90-minute central path with the Vercel AI SDK, plus six branches for where you go next.
  2. Claude Code & Agent Skills HubHubInstall, build, and evaluate skills in the open SKILL.md format, and compare skills, plugins, subagents, and MCP servers.
  3. Skill BuilderToolFill the fields, copy the generated SKILL.md, and drop it into a Claude, Codex, Gemini, or custom agent repo.
  4. Prompt LibraryLibraryAn open prompt corpus, ranked, attributed, and tagged by lab, category, and model lane.
  5. Implementation guidesField guideStep-by-step guides distilled from real agent-assisted builds, with the decisions kept visible.

Step 03

Run agents as a crew

How do I operate many agents without losing control of quality?

Operating agents is a management problem: roles, bounded scopes, quality gates, and memory that compounds instead of drifting.

  1. ACOSOpen sourceThe open-source creator operating system for Claude Code: auto-activating skills, specialised agents, and commands.
  2. Agent catalogLibraryExplore the ACOS catalog and set up the public source installer.
  3. Agentic CreatorField guideNine roles for people who run AI as a crew, from creator and producer to orchestrator and curator.
  4. Agentic Life OSField guideDecide the highest-leverage move, execute with bounded agents, keep the evidence, and turn results into assets.
  5. Agentic Builder LabHubBuild-in-public field notes on the agentic dev stack: Claude Code, Codex, Gemini, Antigravity, and Firebase.

Step 04

Architect for production

What does this look like when a team, a budget, and real users depend on it?

Retrieval, multi-agent orchestration, LLM gateways, MCP tool servers, and the operating model that keeps them governed.

  1. AI ArchitectureHubRetrieval apps, bounded agent teams, persistent cloud workers, and automation, with source-backed diagrams.
  2. Architecture blueprintsBlueprintInteractive reference designs for multi-agent orchestration, enterprise RAG, LLM gateways, and MCP servers.
  3. Multi-cloud AI comparisonComparisonAWS Bedrock, Azure AI Foundry, Google Vertex AI, and OCI GenAI side by side.
  4. AI Operations researchHubLLM infrastructure, memory systems, and multi-agent orchestration for production operations.
  5. AI Center of ExcellenceField guideStrategy, governance, talent, technology, data, and ethics in one operating model.

Step 05

Go deeper with Frank

When is it worth bringing in a second architect?

The playbooks are free. When the decision is consequential or the system has to be installed into a real business, work with me directly.

  1. WorkshopsWorkshopDelivered workshops and studio curricula on AI agents, MCP, accountable memory, and coordinated work.
  2. AI Architecture StudioEngagementArchitecture, coaching, and builds for teams that need the system to ship.
  3. FoundryEngagementComplete AI operating systems installed into businesses: website, agent harness, quality gates, and memory.
  4. Founder's CircleEngagementA quarterly strategic AI retainer for founders whose problem is judgment under uncertainty.

Open source

Clone the parts that already work

Claude Skills LibraryOpen repository of modular SKILL.md definitions and evaluation harnesses.github.com/frankxai/claude-skills-libraryMCP DoctorDiagnostic CLI for Model Context Protocol server configurations.github.com/frankxai/mcp-doctor

Questions people ask first

What is an AI agent playbook?
An AI agent playbook is a tested sequence for choosing a stack, building an agent, packaging its behaviour as reusable skills, and operating it with quality gates. The FrankX playbooks link each step to a working guide, tool, or open-source repository.
Where should I start if I have never built an agent?
Start with the Build Your First AI Agent workshop, then turn whatever worked into a SKILL.md module with the Skill Builder. Hardware and model choices only matter once you know what the agent has to do.
Do I need a local AI computer to run agents?
No. Most agent work runs well against hosted models. A local workstation or mini PC becomes worth it for private data, heavy local inference, or predictable monthly cost; the AI Computer Field Guide compares the options.
Are the playbooks free?
Yes. The guides, hubs, skill tools, and open-source repositories are free. Workshops, the AI Architecture Studio, the Foundry, and the Founder’s Circle are paid engagements for teams that want direct help.

Start with one agent this week.

The 90-minute workshop path gets a working agent on your screen. Everything else on this page builds on it.

Build your first agent

Want the long reads instead? Open the magazine.