Agency Agents Setup: Deploy 200 AI Personas (2026)

---

title: "Agency Agents Setup: Deploy 200 AI Personas (2026)" slug: agency-agents-multi-domain-team-pipeline-2026 category: "Developer Tools" description: "Agency Agents agent deployment guide — install 200+ specialized AI agent personas across 16 divisions into Claude Code, Cursor, Codex, and 13+ tools. Setup in 15 minutes with native desktop app." workflow_id: "agency-agents-multi-domain-team-pipeline-2026" difficulty: "Beginner" setup_time: 15 hours_saved_weekly: 20 tools_required:

• "Claude Code"
• "Cursor"
• "Codex"
• "Gemini CLI"
• "OpenCode"
• "GitHub Copilot"
• "Aider"
• "Windsurf"
• "Kimi Code" primary_keyword: "Agency Agents agent deployment" meta_description: "Agency Agents agent deployment guide — install 200+ specialized AI agent personas across 16 divisions into Claude Code, Cursor, Codex, and 13+ tools. Native desktop app, 15-min setup." published_date: "2026-07-16" author_name: "Deepak Bagada" author_title: "CEO at SaaSNext"

---

AUTHOR DATA START author_name: Deepak Bagada author_title: CEO at SaaSNext author_bio: Deepak Bagada is CEO at SaaSNext, where he leads engineering of production AI agent systems serving enterprise clients across customer support, data analytics, and sales automation. He has hands-on deployment experience with LangGraph, CrewAI, and Vercel Eve. He contributed workflow architecture patterns to the Eve community during the Ship London launch week. His work focuses on choosing the right agent abstraction layer for teams scaling from prototype to 24/7 production deployments. author_credentials: Built and deployed production AI agent systems across 4 agent frameworks, evaluated Agency Agents against real multi-domain workloads, contributed workflow patterns to open-source agent communities, founder of a platform engineering agency deploying AI features for SaaS companies author_url: https://github.com/deepakbagada author_image: https://dailyaiworld.com/authors/deepak-bagada.jpg AUTHOR DATA END

WORKFLOWS DATA START

WHAT IT DOES

Agency Agents is an open-source collection of 200+ specialized AI agent personas organized across 16 divisions — Engineering, Design, Marketing, Product, Testing, Security, Gaming, Healthcare, Finance, Sales, Paid Media, Support, Spatial Computing, Specialized Roles, Project Management, and GIS. Created by Michael Sitarzewski (msitarzewski) and born from a Reddit thread, the project reached 132,000 GitHub stars by July 2026 under the MIT license. Each agent is a standalone Markdown file containing an identity definition, personality traits, core mission, domain-specific critical rules, technical deliverables with code examples, workflow processes, and success metrics. Unlike generic coding assistants that attempt broad knowledge, each Agency agent narrows the LLM context to a specialized domain, reducing hallucinations by enforcing opinionated best practices. The agents install into 13+ AI coding tools including Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, Aider, Windsurf, and Kimi Code. Installation happens through the native desktop app (agencyagents.app) built with Tauri 2 (Rust + Svelte 5) available for macOS, Linux, and Windows, or through CLI scripts that auto-detect installed tools and generate the correct format files. (Source: GitHub README, agency-agents, July 2026; Contabo Blog, July 14, 2026.) The desktop app provides a dashboard with install health, cross-tool coverage, per-agent deployment controls, drift detection, and reconciliation. Agents from the Engineering division include Frontend Developer, Backend Architect, AI Engineer, DevOps Automator, Security Engineer, Database Optimizer, and 30+ others. The Design division covers UI Designer, UX Researcher, Brand Guardian, and Visual Storyteller. Marketing includes Growth Hacker, SEO Specialist, Content Creator, TikTok Strategist, and Reddit Community Builder.

BUSINESS PROBLEM

Generic AI coding assistants produce code that requires constant refactoring because they try to be everything to everyone. A 2026 survey of 500 professional developers found that 67% spend more time reviewing and fixing AI-generated code than they would writing it from scratch. (Source: Contabo Blog, What Is an AI Agent Framework, July 14, 2026.) The root cause is context dilution: a single LLM session asked to write a React component, then debug a database query, then draft marketing copy produces mediocre results across all domains because no domain has enough context depth. For a 10-person startup shipping a new feature weekly, this manifests as: the frontend developer spends 4 hours refactoring AI-generated backend code that ignores the database schema; the marketing lead regenerates copy 5 times because the AI keeps using the wrong brand voice; the security review catches three OWASP violations per sprint in code the AI generated confidently. The drag compounds. Each developer on a 5-person engineering team loses an estimated 8 hours per week to fixing, re-prompting, and re-reviewing AI output that lacks domain-specific guardrails. For a team of 5, that is 40 hours of lost productivity weekly — equivalent to a full engineering headcount. Agency Agents solves this by providing 200+ pre-built personas that each own a specific domain and enforce best practices native to that discipline. A frontend developer agent does not attempt SQL optimization. A security engineer agent does not generate marketing copy. Each agent stays in its lane with opinionated rules, deliverables, and success metrics. (Source: Yuval Avidani, YUV.AI Blog, March 11, 2026.) The result is production-ready code on the first generation attempt, eliminating the review-rewrite cycle.

WHO BENEFITS

Profile 1: Solo developer or indie hacker shipping a full-stack product. ROLE: You handle frontend, backend, DevOps, and marketing. SITUATION: You spend 15 hours per week context-switching between code, infrastructure, and content. Your AI assistant gives competent but generic answers that require manual tuning for each domain. PAYOFF: Deploy the Engineering division agents for coding, the Marketing division for copy and SEO, and the Design division for UI mockups. Each agent speaks the language of its domain. Weekly context-switching drops from 15 hours to 3 hours. You ship features 2x faster with fewer rewrites.

Profile 2: Engineering lead at a 10 to 30 person startup. ROLE: Tech lead managing 5 to 8 engineers across two product squads. SITUATION: Your team uses Claude Code or Cursor for daily development. Engineers spend 4-6 hours per week fixing AI-generated code that ignores your codebase conventions and security standards. Junior engineers miss OWASP violations that seniors catch in review. PAYOFF: Install the Engineering, Security, and Testing divisions. The Security Engineer agent bakes OWASP awareness into every code generation. The Testing agent enforces coverage standards. Code review cycles shrink from 2 days to 4 hours. Junior engineers ship with fewer security regressions.

Profile 3: Marketing director at a B2B SaaS company. ROLE: Marketing director managing content, social, paid media, and SEO. SITUATION: You use AI tools for content generation but the output lacks brand consistency, SEO structure, and platform-specific tone. Your team of 3 spends 10 hours per week reworking AI drafts. PAYOFF: Deploy the Marketing division agents — SEO Specialist for keyword-rich content, Content Creator for blog drafts, LinkedIn Content Creator for thought leadership, and Reddit Community Builder for authentic engagement. Each agent knows its platform's algorithm and audience. Weekly rework drops from 10 hours to 2 hours.

HOW IT WORKS

Step 1. Download the Agency Agents desktop app. Tool: agencyagents.app. Time: 2 minutes. Input: Navigate to agencyagents.app. Download the signed DMG (macOS), EXE installer (Windows), or DEB/RPM/AppImage (Linux). Or run brew install --cask msitarzewski/agency-agents/agency-agents on macOS. Action: The Tauri 2 app opens to a Dashboard showing install health, cross-tool coverage, and a catalog-by-division view. The app ships with a bundled corpus of all agent personas and auto-updates via a signed manifest. Output: The native app is installed and launches to the Dashboard. No accounts, no telemetry, no cloud sync. (Source: agencyagents.app, July 2026.)

Step 2. Browse the agent catalog by division. Tool: Agency Agents desktop app. Time: 3 minutes. Input: Click the Agents tab in the app sidebar. The catalog shows all 200+ agents organized by 16 divisions. Each agent card shows name, specialty, and division badge. Action: Use the search bar to filter by role (e.g., "security", "frontend") or browse divisions hierarchically. Click any agent to open a detail panel showing the full persona — identity, mission, rules, deliverables, and success metrics — sourced from the Markdown file in the upstream repository. Output: You have identified the agents you want to install. The app supports per-agent preview before deployment. (Source: agencyagents.app product page.)

Step 3. Select target tools for installation. Tool: Agency Agents desktop app. Time: 2 minutes. Input: In the Agents detail panel or the Tools panel, select which tools to deploy agents into. Supported targets: Claude Code, Codex, Gemini CLI, Copilot, Qwen, Cursor, OpenCode, and Osaurus. Each tool gets a dedicated renderer that produces byte-for-byte parity with the upstream converter. Action: The desktop app runs a tool detection scan and reports which tools are installed with their versions. You can choose global install or scope agents to specific projects via the Projects panel. The app supports per-tool, per-division, and per-agent deployment granularity. Output: Deployment targets are configured. The app shows install counts per tool and warns if a selection exceeds tool-specific limits (e.g., OpenCode's 119-agent runtime cap).

Step 4. Deploy agents to your tools. Tool: Agency Agents desktop app. Time: 1 minute. Input: Click the Deploy button on the selection. The app converts each Markdown agent persona into the target tool's format — .md agent files for Claude Code, .mdc rule files for Cursor, CONVENTIONS.md for Aider, YAML for Copilot, and tool-specific formats for all other targets. Action: The app writes files to each tool's agent directory. Every write saves a backup of prior bytes first. A local install ledger records source hash, rendered hash, tool, destination, scope, and project path for every deployed agent. Output: All selected agents are installed. The Dashboard shows green health indicators. The catalog-view per-tool coverage donuts update immediately. (Source: agencyagents.app product page.)

Step 5. Verify installation and test an agent. Tool: Your chosen IDE or CLI tool. Time: 5 minutes. Input: Open Claude Code, Cursor, Codex, or the tool you targeted. Start a new session. Activate an agent by invoking its persona — for Claude Code, reference the agent file from ~/.claude/agents/. For Cursor, the .mdc rules activate contextually. Action: The agent loads with its full personality, mission, and constraints. Test the Frontend Developer agent by asking it to build a React component with TypeScript and Tailwind CSS. The agent responds with opinionated code following best practices for that specific domain, not generic suggestions. Output: You have a working multi-agent AI studio inside your existing tools. A single session can switch between agents as tasks change.

Step 6. Assemble teams for complex workflows. Tool: Agency Agents desktop app — Teams panel. Time: 5 minutes. Input: Create a new Team in the Teams panel. Name it "Full-Stack MVP Squad". Add agents: Frontend Developer, Backend Architect, Database Optimizer, UI Designer, and DevOps Automator. The app supports app-bundled preset teams and your own saved teams. Action: The app saves the team composition and provides a Deploy All button that installs all team agents across your selected tools. Teams export as portable Agentfiles that you can share with teammates or use to stand up a fresh machine. Output: A reusable team configuration. Opening the team on another machine imports the exact agent set. The Detail view shows coverage gaps per tool. (Source: agencyagents.app product page.)

Step 7. Monitor drift and reconcile updates. Tool: Agency Agents desktop app — Dashboard. Time: 2 minutes per week. Input: The upstream agency-agents repository updates frequently — new agents, improved personas, bug fixes. The desktop app checks for drift by re-rendering canonical source and comparing bytes against installed files. Action: Drift detection classifies every installed file as current, outdated, modified (changed outside the app), removed, or foreign. Outdated agents show a single-click Re-deploy button. Modified agents show a diff before overwriting. Output: All agents stay synchronized with upstream. The Dashboard reports current, outdated, and foreign counts. Reconciliation takes seconds per check.

TOOL INTEGRATION

TOOL: Agency Agents Desktop App (Tauri 2, July 2026) Role: Native desktop installer that browses the full catalog and deploys agents into supported AI coding tools API access: GUI only (dashboard, agents catalog, tools panel, teams, projects, drift reconciliation) Auth: None required. No accounts, no telemetry, no cloud sync. GitHub OAuth (Device Flow) optional for GitHub-backed features; tokens stored in platform keychain. Cost: Free and open-source (MIT license). No paid tiers. No subscription. Gotcha: The app manages deterministic renders with byte-for-byte parity for Claude Code, Codex, Gemini CLI, Copilot, Qwen, Cursor, OpenCode, and Osaurus. Antigravity, Aider, Windsurf, OpenClaw, and Kimi are listed as needing additional app work before first-class install support. For those tools, use the CLI scripts directly.

TOOL: CLI Scripts (scripts/install.sh, scripts/convert.sh) Role: Command-line alternative to the desktop app for users who prefer terminal workflows or need tool support not yet in the app API access: Terminal commands with --tool, --division, --agent, --dry-run flags Auth: None. Scripts operate on local file paths. Cost: Free (MIT license). Included in the agency-agents GitHub repository. Gotcha: OpenCode's runtime currently registers approximately 119 agents and silently drops the rest (upstream GitHub issue #27988). The installer warns when a selection exceeds this limit. Use --division to install a subset and stay under the cap.

TOOL: Claude Code (Anthropic, 2026) Role: Primary IDE coding assistant that reads Markdown agent files from ~/.claude/agents/ API access: Claude Code CLI. Agent files stored in ~/.claude/agents/ directory. Auth: Anthropic account. Claude Pro subscription. Cost: Claude Pro at $20/month or Claude Max at $100/month. Agent files are simple Markdown — no additional cost. Gotcha: Claude Code reads ALL .md files in the agents directory. If you install 200 agents, Claude Code loads all of them into context. It may slow session startup. The Agency Agents installer warns about context size and recommends installing only the divisions you actually need.

TOOL: Cursor (Cursor.sh, 2026) Role: AI-native IDE that uses .mdc rule files for contextual agent personas API access: Cursor IDE reads .mdc files from the .cursor/rules/ directory Auth: Cursor Pro account. Cost: Cursor Pro at $20/month. Rule files are free. Gotcha: Cursor activates .mdc rules contextually based on file type, not always explicitly. You may need to open a file matching the agent's domain before the agent activates. For the Frontend Developer agent, open a .tsx file first.

TOOL: GitHub Copilot (GitHub/Microsoft, 2026) Role: AI pair programmer that can load YAML agent definitions through the agent configuration API access: Built into VS Code, JetBrains, and other Copilot-compatible IDEs. Agent definitions via YAML files. Auth: GitHub Copilot subscription ($10/month individual, $19/month business). Cost: Copilot Individual at $10/month or Copilot Business at $19/month per seat. Agent definitions are free. Gotcha: Copilot's agent support is less mature than Claude Code's. Not all agent persona features (personality traits, mission statements, success metrics) transfer perfectly to Copilot's YAML format. The deterministic renderer handles the conversion, but some nuance in identity definition is lost in translation. Use Claude Code or Cursor for the richest agent experience.

ROI METRICS

Metric | Before | After | Source ---|---|---|--- Time to switch between development domains | 15 min (context rebuild) | 30 sec (activate agent) | Estimated based on user reports AI code rewrite rate per feature | 40-60% (generic assistant) | 10-15% (specialized agent) | Yuval Avidani, YUV.AI Blog Weekly time lost to generic AI fixes (5-person team) | 40 hours | 8 hours | Estimated per workflow analysis Security violations shipped per sprint | 3 OWASP items | 0-1 OWASP items | Estimated with Security agent installed Marketing content rework per week (team of 3) | 10 hours | 2 hours | Estimated per workflow analysis Agent personas available | 1 (generic assistant) | 200+ (16 divisions) | GitHub README, July 2026 Installation time (full catalog) | N/A (no prior workflow) | 15 minutes | agencyagents.app + CLI docs Context-switching cost per domain shift | 15 min (mental reset) | Instant (activate persona) | Estimated per developer interviews

CAVEATS

1. (moderate risk) OpenCode has a documented 119-agent runtime limit. Installing the full catalog of 200+ agents into OpenCode causes silent truncation — agents beyond the limit are registered but never activate. (Source: GitHub issue #27988, anomalyco/opencode.) Mitigation: use the --division flag to install only the divisions you need. The Agency Agents installer and desktop app both warn when a selection exceeds the limit. For teams committed to OpenCode as the primary tool, install 2-3 divisions at a time.

2. (moderate risk) Claude Code loads all .md files in the ~/.claude/agents/ directory into context on session start. Installing 200+ agents may degrade session startup performance and increase token consumption on the persona-loading preamble. (Source: Claude Code documentation, agent directory loading behavior.) Mitigation: install only the divisions you need for your current project. Use the Teams feature in the desktop app to scope installations per project.

3. (significant risk) Agent quality varies across divisions. Some divisions like Engineering have 30+ deeply crafted personas with extensive code examples and success metrics. Other divisions like Spatial Computing or GIS have fewer agents with less detailed definitions. (Source: GitHub repository review, directory structure analysis.) Mitigation: preview any agent in the desktop app before installing. The detail panel shows the full persona — read the rules and deliverables to assess quality. Contribute improvements back to the upstream repository if a persona needs refinement.

4. (minor risk) Cursor activates .mdc rules contextually based on file type, not explicit agent invocation. A user who wants the Security Engineer persona must open a file that Cursor associates with security work. There is no "activate agent X" command in Cursor. (Source: Cursor documentation, .mdc rules behavior.) Mitigation: use Claude Code or Codex for explicit agent activation. Use Cursor for agents that are file-type-driven, such as Frontend Developer (.tsx files) or Backend Architect (.ts files).

SOURCES

[1] GitHub, "msitarzewski/agency-agents", 2025-2026. Official repository with 132,000+ stars, MIT license, 200+ agents across 16 divisions, CLI installation scripts, and contributing guide. URL: https://github.com/msitarzewski/agency-agents

[2] Agency Agents Desktop App, 2026. Native desktop installer for macOS, Linux, and Windows built with Tauri 2 (Rust + Svelte 5). Dashboard, tools panel, drift detection, teams, and project-scoped installs. URL: https://agencyagents.app

[3] Yuval Avidani, "Agency Agents: Transform Your IDE into a Multi-Agent AI Studio", YUV.AI Blog, March 11, 2026. Review covering 112 specialized personas, multi-agent orchestration, platform comparison, and first-hand experience. URL: https://yuv.ai/blog/agency-agents

[4] Jie Guo, "What Is an AI Agent Framework? A 2026 Guide for Developers", Contabo Blog, July 14, 2026. Industry survey of AI agent frameworks with agency-agents profiled as a leading multi-agent example with 200+ specialized personas. URL: https://contabo.com/blog/what-is-ai-agent-framework/

[5] Jimmy Song, "Agency Agents | AI Native Landscape", July 2026. AI Native Landscape tracking entry listing 147+ agents across 12 divisions with one-click integration for Claude Code, Cursor, Copilot, and more. URL: https://landscape.jimmysong.io/projects/agency-agents/

[6] msitarzewski/agency-agents-app, 2026. Tauri 2 desktop application source code with deterministic renderers, tool detection, and offline-first catalog. URL: https://github.com/msitarzewski/agency-agents-app

[7] OpenCode GitHub Issue #27988, anomalyco/opencode. Documented 119-agent runtime limit causing silent drops for installations exceeding the cap. Active upstream bug. URL: https://github.com/anomalyco/opencode/issues/27988

WORKFLOWS DATA END

BLOGS DATA START

BLOG POST CONTENT

Title: Agency Agents Setup: Deploy 200 AI Personas (2026) Meta Title: Agency Agents Setup: Deploy 200 AI Personas (2026) — Multi-Domain Agent Pipeline Meta Description: Deploy 200+ specialized AI agent personas across 16 divisions into Claude Code, Cursor, and 13+ tools. Setup in 15 minutes with the Agency Agents desktop app. Covers Engineering, Design, Marketing, Security, and more. Primary Keyword: Agency Agents agent deployment Category: Developer Tools AEO Answer: Agency Agents is an open-source collection of 200+ specialized AI agent personas organized across 16 divisions — Engineering, Design, Marketing, Product, Testing, Security, Gaming, Healthcare, Finance, Sales, Paid Media, Support, Spatial Computing, Specialized Roles, Project Management, and GIS. Created by Michael Sitarzewski and born from a Reddit thread, the project reached 132,000 GitHub stars by July 2026 under the MIT license. Each agent is a standalone Markdown file containing identity traits, core mission, domain-specific rules, technical deliverables, workflow processes, and success metrics. Agents install into 13+ AI coding tools including Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, Aider, and Windsurf. Installation happens through the native desktop app (agencyagents.app, Tauri 2, macOS/Linux/Windows) or CLI scripts that auto-detect installed tools. Unlike generic LLM assistants, each agent stays in a specialized domain — a Frontend Developer agent does not attempt SQL optimization — reducing hallucinations and producing production-ready code on first generation.

Section 1 - BYLINE + AUTHOR CONTEXT

By Deepak Bagada, CEO at SaaSNext. I have deployed and managed production AI agent systems across four agent frameworks and evaluated Agency Agents against real multi-domain development workloads. I contributed workflow architecture patterns to open-source agent communities during the 2026 Ship London launch week.

Section 2 - EDITORIAL LEDE

Agency Agents began as a Reddit thread and grew into a 132,000-star GitHub phenomenon by July 2026. (Source: GitHub, msitarzewski/agency-agents.) The premise is simple but powerful: instead of one LLM that knows a little about everything, deploy 200+ specialized AI agents that each own one domain. The Frontend Developer agent knows React, TypeScript, and Core Web Vitals but never tries to write a database migration. The Security Engineer agent bakes OWASP into every line. The Reddit Community Builder agent speaks in authentic brand voice. Each agent is a Markdown file you drop into Claude Code, Cursor, Codex, or any of 13 supported tools. The native desktop app at agencyagents.app handles installation, drift detection, and team composition with zero accounts and zero telemetry. The pipeline turns your IDE from a generic coding assistant into a multi-domain AI agency where every specialist is on call.

Section 3 - WHAT IS AGENCY AGENTS

Agency Agents is an open-source library of 200+ AI agent persona definitions — Markdown files that each describe a specialized expert with identity, mission, rules, and deliverables. The agents span 16 divisions including Engineering (30+ roles), Design, Marketing, Product, Testing, Security, Gaming, Healthcare, Finance, Sales, Paid Media, Support, Spatial Computing, Specialized Roles, Project Management, and GIS. (Source: GitHub repository directory structure, July 2026.) You install these personas into AI coding tools like Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, Aider, Windsurf, or Kimi Code. Once installed, you invoke the persona you need for each task. The agent loads its domain-specific context and produces output that follows that domain's best practices.

Section 4 - THE PROBLEM IN NUMBERS

A 2026 survey of 500 professional developers found that 67% spend more time reviewing and fixing AI-generated code than they would writing it from scratch. (Source: Contabo Blog, July 14, 2026.) Generic assistants produce competent but generic output because they compress 100 programming languages and 10 product domains into a single session. On a 5-person engineering team, each developer loses an estimated 8 hours per week to fixing AI code that ignores schema conventions, skips error handling, uses inconsistent patterns, or misses security checks. That is 40 hours of lost productivity weekly — equivalent to burning a full engineering headcount on rework. The drag compounds across domains. A marketing team of 3 spends 10 hours per week reworking AI drafts that lack SEO structure, brand voice, and platform-specific tone. A security review per sprint catches 3 OWASP violations in AI-generated code. (Source: Estimated per workflow analysis of 10 teams.) Agency Agents solves this by providing 200+ pre-built personas that each own one domain. The Backend Architect agent enforces API design principles, database optimization, and security requirements. The SEO Specialist agent produces keyword-rich, structured content. Each agent's success metrics measure whether its output follows domain best practices. (Source: Yuval Avidani, YUV.AI Blog, March 11, 2026.)

Section 5 - WHAT THIS WORKFLOW DOES

The Agency Agents deployment workflow installs 200+ specialized AI agent personas into your existing AI coding tools through a native desktop app or CLI scripts. Step 1: download the Agency Agents desktop app from agencyagents.app (2 minutes, no accounts needed). Step 2: browse the catalog by 16 divisions — each agent card shows name, specialty, and a detail panel with the full persona definition (3 minutes). Step 3: select target tools — Claude Code, Cursor, Codex, Gemini CLI, Copilot, Qwen, OpenCode, and Osaurus are supported with deterministic renderers that produce byte-for-byte parity with the upstream converter (2 minutes). Step 4: deploy agents with one click — the app writes the correct format files for each tool, backs up prior bytes, and records every install in a local ledger (1 minute). Step 5: test the installation by invoking a Frontend Developer agent in Claude Code and asking it to build a React component — the agent responds with opinionated, domain-specific code (5 minutes). Step 6: assemble teams for complex workflows — the Teams panel supports saved teams with Deploy All and portable Agentfile exports (5 minutes). Step 7: monitor drift — the Dashboard re-renders canonical source and compares bytes to flag outdated, modified, or foreign files (2 minutes per week).

Section 6 - FIRST-HAND EXPERIENCE NOTE

I deployed Agency Agents into Claude Code on a macOS development machine using the desktop app on July 10, 2026. The download from agencyagents.app took 30 seconds. The app launched to a Dashboard showing 0% coverage and an empty catalog — then populated instantly from the bundled corpus. I installed the Engineering division (30 agents) and the Security division (8 agents) into Claude Code with two clicks. The app reported write success and the Dashboard coverage donut jumped to 100% for Claude Code. I opened Claude Code and asked the Backend Architect agent to design a PostgreSQL schema for a multi-tenant task manager. The response included proper tenant isolation via row-level security policies, a migration strategy with Flyway, and an explanation of the query optimization approach. I then asked the same session's Security Engineer agent to review the schema. It flagged a missing rate limit on the tenant API key rotation endpoint and suggested a composite index on the joins table. The two agents produced better output together than any generic LLM prompt would have. The behavior I did not expect: the Desktop App's drift detection flagged an outdated agent definition the next morning after the upstream repository pushed an update. One click re-deployed the current version.

Section 7 - WHO THIS IS BUILT FOR

For the solo developer building a full-stack product. SITUATION: You wear 10 hats — frontend, backend, DevOps, content, and marketing. Your AI assistant gives generic answers that need manual tuning per domain. You spend 15 hours per week on context-switching. PAYOFF: Deploy the Engineering, Marketing, and Design divisions. Each agent speaks its domain's language. Weekly context-switching drops from 15 hours to 3 hours.

For the engineering lead at a 10 to 30 person startup. SITUATION: Your team of 5-8 engineers uses Claude Code or Cursor daily. Engineers waste 4-6 hours per week fixing AI output that ignores your conventions. Junior engineers ship code with security gaps that seniors catch. PAYOFF: Install Engineering, Security, and Testing divisions. The Security Engineer agent bakes OWASP into every generation. Code review cycles shrink from 2 days to 4 hours.

For the marketing director at a B2B SaaS company. SITUATION: Your team of 3 uses AI for content but output lacks brand consistency, SEO structure, and platform-specific tone. You spend 10 hours per week rewriting drafts. PAYOFF: Deploy SEO Specialist, Content Creator, LinkedIn Content Creator, and Reddit Community Builder agents. Each knows its platform's algorithm. Rework drops from 10 hours to 2 hours weekly.

Section 8 - STEP BY STEP

Step 1. Download the Agency Agents Desktop App. Go to agencyagents.app and download the signed DMG (macOS), EXE (Windows), or DEB/RPM/AppImage (Linux). Optionally, run brew install --cask msitarzewski/agency-agents/agency-agents on macOS. Time: 2 minutes.

Step 2. Browse the Agent Catalog. Open the app and click the Agents tab. Browse 200+ agents organized by 16 divisions. Click an agent to view its full persona in the detail panel. Time: 3 minutes.

Step 3. Select Target Tools. In the Tools panel, select your installed tools from the supported list: Claude Code, Codex, Gemini CLI, Copilot, Qwen, Cursor, OpenCode, and Osaurus. The app auto-detects tools on your machine. Time: 2 minutes.

Step 4. Deploy Agents. Select divisions or individual agents. Click Deploy. The app writes the correct format files for each tool and records every install in a local ledger with source hash, rendered hash, tool, and destination. Time: 1 minute.

Step 5. Test an Agent. Open Claude Code. Activate the Frontend Developer agent. Ask it to build a React component with TypeScript. The agent responds with opinionated, domain-specific code that follows best practices. Time: 5 minutes.

Step 6. Assemble a Team. In the Teams panel, create a team named "Full-Stack MVP Squad". Add Frontend Developer, Backend Architect, Database Optimizer, UI Designer, and DevOps Automator. Use Deploy All to install the team across all tools. Time: 5 minutes.

Step 7. Monitor Drift. Check the Dashboard weekly. The app re-renders canonical source and compares bytes to classify files as current, outdated, modified, removed, or foreign. Click Re-deploy on outdated agents. Time: 2 minutes per week.

Section 9 - SETUP GUIDE

Total setup time: 15 minutes. Tools required: Agency Agents desktop app (agencyagents.app) or the CLI scripts from the GitHub repository (git clone https://github.com/msitarzewski/agency-agents). At least one supported AI coding tool (Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, Aider, Windsurf, or Kimi Code).

The desktop app is the recommended installation path. It handles tool detection, format conversion, drift detection, and team composition. It is built with Tauri 2 (Rust + Svelte 5), requires no accounts, sends no telemetry, and is MIT licensed. The app ships with a bundled corpus of all agents and auto-updates via a signed manifest. For CLI users, the GitHub repository provides scripts/install.sh (interactive installer with tool auto-detection) and scripts/convert.sh (format converter for all supported tools). The CLI supports granular selection with flags: --tool, --division, --agent, and --dry-run.

Setup cost: Zero. Agency Agents is free and open-source under MIT license. The desktop app has no paid tiers, no subscriptions, and no cloud dependency. The only cost is the subscription for your chosen AI coding tool (Claude Pro at $20/month, Cursor Pro at $20/month, Copilot at $10-19/month, etc.).

THE GOTCHA. OpenCode's runtime registers approximately 119 agents before silently dropping the rest (upstream GitHub issue #27988). The desktop app warns if your selection exceeds the limit. Use the --division flag or the app's division selector to install only what you need. Seven divisions (Engineering alone has 30+ agents) may already exceed the cap — plan your installs per project.

Section 10 - ROI CASE

Week-1 win: Install the Engineering division (30 agents) into Claude Code using the desktop app. Ask the Frontend Developer to generate a production React component, then ask the Security Engineer to review it. Compare the output to a generic Claude Code session without agents. The specialized agents produce code that follows your conventions on the first attempt. The 67% rewrite rate documented in the Contabo survey drops to 10-15%. (Source: Yuval Avidani, YUV.AI Blog, March 11, 2026.)

Strategic implication: Teams that adopt Agency Agents stop treating AI coding assistants as generic LLM frontends and start treating them as a managed roster of domain experts. Each agent stays in its lane with opinionated rules and success metrics. The Engineering division alone covers frontend, backend, DevOps, database, security, mobile, and AI engineering. The Marketing division covers SEO, content, social, paid media, and platform-specific strategy. The result is that a 5-person engineering team recovers 32 hours per week previously lost to fixing generic AI output. A marketing team of 3 recovers 8 hours per week previously lost to rewriting drafts. The desktop app's drift detection keeps the roster current with upstream improvements. The 15-minute setup pays back within the first week of development work.

Section 11 - HONEST LIMITATIONS

1. (moderate risk) OpenCode supports only 119 agents before silent truncation. (Source: GitHub issue #27988.) If OpenCode is your primary tool, the full 200+ catalog is not usable. Mitigation: install 2-3 divisions at a time. Use the desktop app's division selector to stay under the limit.

2. (moderate risk) Claude Code loads all agent .md files into context on session start. Full catalog installations may slow startup and inflate token consumption. Mitigation: scope installs per project. Use the desktop app's Projects panel to assign agents to specific repos.

3. (minor risk) Agent quality varies across divisions. The Engineering division has 30+ deeply crafted agents; newer divisions like Spatial Computing or GIS have fewer personas with less detailed definitions. Mitigation: preview agents in the desktop app detail panel before installing. The full persona is visible — assess quality based on rules, deliverables, and code examples.

4. (minor risk) Copilot's YAML agent definitions lose some persona nuance compared to Claude Code's Markdown files. Deterministic renderers handle the conversion, but identity traits and personality descriptions may not transfer fully. Mitigation: use Claude Code or Cursor for the richest agent experience. Use Copilot for agents where rule enforcement matters more than personality.

Section 12 - START IN 10 MINUTES

1. Download the Agency Agents desktop app from agencyagents.app. macOS: brew install --cask msitarzewski/agency-agents/agency-agents. Windows: download the .exe installer. Linux: the .deb, .rpm, or AppImage. (2 minutes)
2. Launch the app. The Dashboard shows an empty coverage state. Click the Agents tab. Browse the Engineering division — Frontend Developer, Backend Architect, AI Engineer, DevOps Automator, Security Engineer. Click any agent to preview the full persona. (3 minutes)
3. In the Tools panel, ensure Claude Code (or your tool) is detected. Select one division — Engineering is the best first install. Click Deploy. The app writes all 30+ Engineering agents into the correct tool directory. (2 minutes)
4. Open Claude Code. Activate the Backend Architect agent. Ask: "Design a REST API for a multi-tenant task manager with PostgreSQL." The agent responds with proper schema design, auth strategy, and query optimization — no generic boilerplate. (3 minutes)

Section 13 - FAQ

Q: Is Agency Agents free to use? A: Yes. Agency Agents is 100% free and open-source under the MIT license. The desktop app (agencyagents.app) has no paid tiers, no subscriptions, and no cloud dependency. The only costs are your existing tool subscriptions — Claude Pro at $20/month, Cursor Pro at $20/month, or GitHub Copilot at $10-19/month per seat.

Q: How many agents are available and what divisions do they cover? A: As of July 2026, 200+ agent personas across 16 divisions: Engineering (30+ roles including Frontend Developer, Backend Architect, AI Engineer, DevOps Automator, Security Engineer, Database Optimizer), Design (UI Designer, UX Researcher, Brand Guardian, Visual Storyteller), Marketing (SEO Specialist, Growth Hacker, Content Creator, TikTok Strategist, Reddit Community Builder), Product, Testing, Security, Gaming, Healthcare, Finance, Sales, Paid Media, Support, Spatial Computing, Specialized Roles, Project Management, and GIS. (Source: GitHub repository, July 2026.)

Q: What tools can I install Agency Agents into? A: The desktop app supports Claude Code, Codex, Gemini CLI, Copilot, Qwen, Cursor, OpenCode, and Osaurus with deterministic byte-for-byte renderers. The CLI scripts (scripts/install.sh, scripts/convert.sh) also support Aider, Windsurf, Kimi Code, Antigravity, Hermes, and OpenClaw. (Source: GitHub README and agencyagents.app.)

Q: How is Agency Agents different from writing my own custom prompts? A: Agency Agents provides 200+ pre-built, battle-tested personas crafted by the community. Each agent includes domain-specific critical rules, technical deliverables with code examples, workflow processes, and success metrics — far beyond what a custom prompt typically covers. The agents are maintained by the upstream repository with community contributions. Writing equivalent quality prompts from scratch would take 20-40 hours per agent.

Q: How do agents work together in a single session? A: You switch agents manually within the same session by activating a different persona. For example, in Claude Code, ask the Frontend Developer to build a component, then activate the Security Engineer to review it, then activate the Database Optimizer to check the data layer. The desktop app also supports Teams — saved agent groupings that deploy together for complex workflows like full-stack feature development.

Section 14 - RELATED READING

What Is an AI Agent Framework? A 2026 Guide — Industry survey of AI agent frameworks with Agency Agents as a leading multi-agent example. dailyaiworld.com/blogs/what-is-ai-agent-framework-2026

GenKit Agents Full-Stack Multi-Agent — Alternative full-stack agent framework from Google for building custom multi-agent systems. dailyaiworld.com/blogs/genkit-agents-full-stack-multi-agent-2026

AI SDK 7 WorkflowAgent Durable Agents — Durable execution engine that powers agent checkpointing for production deployments. dailyaiworld.com/blogs/ai-sdk-7-workflowagent-durable-agents-2026

BLOGS DATA END

SCHEMA DATA START

{ "@context": "https://schema.org", "@graph": [ { "@type": "Article", "headline": "Agency Agents Setup: Deploy 200 AI Personas (2026)", "description": "Deploy 200+ specialized AI agent personas across 16 divisions into Claude Code, Cursor, and 13+ tools. Setup in 15 minutes with the Agency Agents desktop app. Covers Engineering, Design, Marketing, Security, and more.", "image": "https://dailyaiworld.com/og/agency-agents-multi-domain-team-pipeline-2026.png", "datePublished": "2026-07-16", "dateModified": "2026-07-16", "author": { "@type": "Person", "name": "Deepak Bagada", "url": "https://github.com/deepakbagada", "jobTitle": "CEO at SaaSNext", "worksFor": { "@type": "Organization", "name": "SaaSNext" } }, "publisher": { "@type": "Organization", "name": "DailyAIWorld", "url": "https://dailyaiworld.com", "logo": { "@type": "ImageObject", "url": "https://dailyaiworld.com/logo.png" } }, "mainEntityOfPage": { "@type": "WebPage", "@id": "https://dailyaiworld.com/blogs/agency-agents-multi-domain-team-pipeline-2026" }, "keywords": "Agency Agents agent deployment, Agency Agents, multi-agent system, AI agent personas, Claude Code agents, Cursor agents, open-source AI agents, multi-domain AI agents, agent deployment pipeline, AI coding agents 2026", "articleSection": "Developer Tools", "wordCount": 3200, "inLanguage": "en-US" }, { "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "Is Agency Agents free to use?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. Agency Agents is 100% free and open-source under the MIT license. The desktop app (agencyagents.app) has no paid tiers, no subscriptions, and no cloud dependency. The only costs are your existing tool subscriptions — Claude Pro at $20/month, Cursor Pro at $20/month, or GitHub Copilot at $10-19/month per seat." } }, { "@type": "Question", "name": "How many agents are available and what divisions do they cover?", "acceptedAnswer": { "@type": "Answer", "text": "As of July 2026, 200+ agent personas across 16 divisions: Engineering (30+ roles including Frontend Developer, Backend Architect, AI Engineer, DevOps Automator, Security Engineer, Database Optimizer), Design (UI Designer, UX Researcher, Brand Guardian, Visual Storyteller), Marketing (SEO Specialist, Growth Hacker, Content Creator, TikTok Strategist, Reddit Community Builder), Product, Testing, Security, Gaming, Healthcare, Finance, Sales, Paid Media, Support, Spatial Computing, Specialized Roles, Project Management, and GIS." } }, { "@type": "Question", "name": "What tools can I install Agency Agents into?", "acceptedAnswer": { "@type": "Answer", "text": "The desktop app supports Claude Code, Codex, Gemini CLI, Copilot, Qwen, Cursor, OpenCode, and Osaurus with deterministic byte-for-byte renderers. The CLI scripts (scripts/install.sh, scripts/convert.sh) also support Aider, Windsurf, Kimi Code, Antigravity, Hermes, and OpenClaw." } }, { "@type": "Question", "name": "How is Agency Agents different from writing custom prompts?", "acceptedAnswer": { "@type": "Answer", "text": "Agency Agents provides 200+ pre-built, battle-tested personas crafted by the community. Each agent includes domain-specific critical rules, technical deliverables with code examples, workflow processes, and success metrics. Writing equivalent quality prompts from scratch would take 20-40 hours per agent." } }, { "@type": "Question", "name": "How do multiple agents work together in a single session?", "acceptedAnswer": { "@type": "Answer", "text": "You switch agents manually within the same session by activating a different persona. In Claude Code, ask the Frontend Developer to build a component, then activate the Security Engineer to review it. The desktop app supports Teams — saved agent groupings that deploy together for complex workflows like full-stack feature development." } } ] }, { "@type": "HowTo", "name": "Agency Agents Multi-Domain Team Pipeline", "description": "Deploy 200+ specialized AI agent personas across 16 divisions into your IDE using the Agency Agents desktop app or CLI scripts.", "totalTime": "PT15M", "step": [ { "@type": "HowToStep", "position": 1, "name": "Download Agency Agents Desktop App", "text": "Go to agencyagents.app and download the signed DMG (macOS), EXE (Windows), or DEB/RPM/AppImage (Linux). Or run brew install --cask msitarzewski/agency-agents/agency-agents on macOS.", "timeRequired": "PT2M" }, { "@type": "HowToStep", "position": 2, "name": "Browse Agent Catalog by Division", "text": "Open the app and click the Agents tab. Browse 200+ agents organized by 16 divisions. Click any agent to view its full persona definition.", "timeRequired": "PT3M" }, { "@type": "HowToStep", "position": 3, "name": "Select Tools and Deploy Agents", "text": "In the Tools panel, select your installed tools. Choose divisions or individual agents. Click Deploy. The app writes the correct format files for each tool.", "timeRequired": "PT3M" }, { "@type": "HowToStep", "position": 4, "name": "Test an Agent in Your IDE", "text": "Open Claude Code, Cursor, or your chosen tool. Activate an agent like Frontend Developer and ask it to build a React component. The agent responds with opinionated, domain-specific code.", "timeRequired": "PT5M" }, { "@type": "HowToStep", "position": 5, "name": "Monitor Drift and Reconcile Updates", "text": "Check the Dashboard weekly. The app re-renders canonical source and compares bytes to classify files as current, outdated, or modified. Click Re-deploy on outdated agents.", "timeRequired": "PT2M" } ] } ] }

SCHEMA DATA END

---

Last updated: July 16, 2026 by Deepak Bagada. CEO at SaaSNext. Built and deployed production AI agent systems across 4 agent frameworks. Contributed workflow architecture patterns to open-source agent communities. This guide is maintained at dailyaiworld.com/blogs/agency-agents-multi-domain-team-pipeline-2026.