Introducing Agentcy: Multi-Agent Creative Collaboration

We're releasing Agentcy, an experimental platform for creative work with multiple AI agents. Built on the AutoGen framework, it coordinates specialized agents—research, design, writing, project management—under human direction.

This isn't a finished product. It's an experiment in how creative teams might work with AI.

Why Build This

Creative work involves repetitive tasks that eat time: background research, draft generation, asset variations, project coordination. These tasks are necessary but not where the interesting thinking happens.

Agentcy distributes these tasks to specialized AI agents. The goal: free up human attention for judgment, direction, and the work that actually requires a person.

We're also testing whether structured multi-agent coordination works better than single-model prompting for complex creative projects. Early results suggest it does, but we're still learning.

The Agents

Each agent handles a specific type of work:

Research Agent — Gathers and synthesizes information. Market research, competitor analysis, reference collection, summarizing long documents. Feeds the knowledge base that other agents draw from.

Creative Director Agent — Reviews work against project goals and brand guidelines. Provides structured feedback on whether generated content aligns with strategy. Works alongside human creative directors, not instead of them.

Designer Agent — Generates visual concepts from text descriptions. Useful for rapid exploration—dozens of logo variations, layout options, or style directions in the time it takes to make one manually. Human designers select and refine.

Writer Agent — Drafts written content: headlines, website copy, social posts. Adapts tone and format to different platforms and audiences. First drafts, not final copy.

Project Manager Agent — Coordinates the other agents. Tracks progress, assigns tasks, flags blockers, keeps humans informed. The connective tissue.

These agents work together under human guidance. The structure matters: defined roles, clear handoffs, human checkpoints.

Open Source

Agentcy is open source. We want other people building on this, finding the problems we haven't found, and pushing the approach further.

The code is at github.com/SCTY-Inc/agentcy.

Open source also means transparency. You can see exactly how the agents work, what prompts they use, and where the limitations are. No black box.

What We're Working On

This is early. Here's where we're focused:

Better reasoning — Moving agents beyond generation toward actual creative judgment. Can an agent evaluate whether a concept is strategically sound, not just grammatically correct?

Context depth — Agents that understand brand history, audience nuance, and project context well enough to produce relevant work without constant hand-holding.

Coordination patterns — How should agents hand off work? When should they collaborate vs. work in parallel? We're testing different approaches.

Output quality — Generated work needs to meet professional standards. We're iterating on prompts, agent configurations, and quality checks.

What This Isn't

Agentcy doesn't replace creative professionals. It's a tool for creative professionals—one that might change how work gets divided between humans and machines, but not who makes the decisions that matter.

We're skeptical of AI tools that promise to automate creativity. Creativity isn't the bottleneck. Time is. Agentcy is about time.

---

If you're interested in this space—creative tooling, multi-agent systems, human-AI collaboration—we'd like to hear from you. The repo is open, and so are we.