Generative Practice: Synthetic Teaming
A Practical Guide to Leading AI Collaborators for Complex Work
Synthetic teaming is the Generative Practice of assembling and managing a team of AI collaborators—each with specialized roles and clear focus—to accomplish complex, multi-faceted work.
Rather than using AI as a single tool you switch between tasks, synthetic teaming is a practice that creates an ongoing team structure where different AI “team members” maintain context, contribute their expertise, and work together through shared documents and coordinated workflows.
This practice transforms AI from a productivity hack into a genuine collaboration model, mirroring how human teams coordinate around shared goals while leveraging AI’s unique capabilities for parallel processing and instant expertise access.
I’ve been developing my synthetic team for about a year now, and it’s fundamentally changed how I work. I started simply—just me and one coding thread—but quickly realized I needed separation between asking questions and doing focused work. That led to two threads, then three, and eventually to my current team of about eight specialized collaborators across multiple projects.
My core team includes Albert (my assistant who helps me track tasks), a tech geek for questions, and a cofounder AI for long-term vision work. My development team has Ava (architect), Dara (design lead), Gavin (tech setup), David (implementation), and Grace (producer who coordinates everyone). Each morning, I record an audio standup during my drive, and Grace transforms it into a structured document the whole team accesses. I’ve found that giving them names makes it easier to remember who is who!
What surprised me most is how similar this feels to leading human teams. I spend time ensuring they have good context through architecture documents and design specs. I check in multiple times during the day. When Ava designs a backend architecture, I copy it into project files so David can implement it and Gavin can configure the environment correctly. When I get stuck on UI design, I sketch something for Dara, who transforms it into proper UX documentation.
The efficiency is startling—I can accomplish in an afternoon what used to take my five-person team weeks. Because I can leverage their experience, I’m able to work at the edge of my capabilities across multiple domains simultaneously. I’m not a great backend architect anymore, but Ava is. I’m rusty on modern tooling, but Gavin keeps me current. Together, we’re far more capable than I am alone, and far more coordinated than I’d be jumping between disconnected AI conversations.
When should you practice Synthetic Teaming?
Synthetic teaming is a collaboration practice that helps coordinate complex, multi-faceted work across different domains of expertise. Engage this practice when:
Your project requires multiple distinct types of expertise. When you’re building something that needs multiple types of expertise, synthetic teaming prevents you from forcing one AI conversation to be all things at once.
You’re working at the edge of your capabilities. When your project demands expertise beyond your current skills, synthetic teaming lets you access specialized knowledge without becoming an expert yourself. You lead the vision while team members contribute domain expertise.
Context keeps getting lost between sessions. When you find yourself re-explaining your project goals, architecture decisions, or working style repeatedly, you need shared context that persists across conversations. Synthetic teaming uses project documents as collective memory, so each team member starts with understanding rather than requiring orientation.
You need parallel progress, not sequential handoffs. When your work doesn’t flow linearly, synthetic teaming enables simultaneous conversations with different specialists.
You’re building something over weeks or months, not hours. When your project has enough scope that you’ll return to it across multiple sessions and need to maintain momentum, synthetic teaming creates continuity.
You want genuine collaboration, not just task execution. When you’re tired of using AI as a productivity tool that you manage transaction-by-transaction, synthetic teaming creates something closer to what happens on a team of humans. You’re leading a team with distributed intelligence.
The Core Method
Step 1: Select an AI platform that supports team-based workflows.
Choose a tool that allows you to bundle multiple chat threads with shared context—Claude Projects, ChatGPT with custom GPTs and projects, or Gemini Gems with project folders. The key capability you need is the ability to give different AI instances access to common documents (architecture specs, design docs, daily standups) so they stay synchronized like a real team. While custom GPTs create reusable starting points for specific roles, projects allow multiple specialized threads to work from the same knowledge base simultaneously. This shared context is what transforms disconnected AI conversations into genuine team coordination.
Step 2: Start with core context documents.
Create a series of foundational documents. Here are three:
A document about yourself (your background, skills, and working style)
A document describing your synthetic team structure (who’s on the team and their roles)
A document outlining your team’s rituals and routines (how you manage files, daily rhythms, and key discoveries about working together).
These documents ensure every team member understands the shared context.
Step 3: Create different team members.
Determine who is on your team. Create each new team member with a focused role definition—architect, designer, producer, tech guide. Keep initial prompts light, describing the essence of the role rather than exhaustive job descriptions. You’ll refine roles retroactively based on how you actually work together over time.
Step 4: Establish daily rhythms like a real team.
Conduct daily standups using audio logs: record what you accomplished yesterday, what you’re focusing on today, and where you’re blocked. Have your producer transform these into standup documents saved in your shared project files. When you begin working with a specific team member, you direct their attention to the standup: “Good morning! Check today’s standup document. I need help architecting the backend.” This ensures they understand the current context and priorities before diving into focused work. Add midday reflections and end-of-day summaries to maintain momentum and create a living record of progress.
Step 5: Work across team members fluidly.
Move between team members based on what the work needs—architecture discussions with your architect, design questions with your designer, coordination with your producer. Unlike chaining tasks sequentially, you’re having parallel conversations with specialists who all share access to the same project documents.
Step 6: Reconstitute team members when they get tired.
When conversations get long and AI team members lose focus or hit token limits, use reconstitution prompts. Ask them to retroactively describe their role, everything you’ve accomplished together, and how you work best. Use this richer description to seed a fresh conversation thread—like giving someone a coffee break and a clear head.
Why It Works
Synthetic teaming draws upon research of how optimal human teams are structured.
Research on “distributed cognition” demonstrates that complex problem-solving improves when expertise is distributed across specialized agents rather than concentrated in generalists.
Studies on human team coordination show that shared mental models—the equivalent of your project documents—significantly improve team performance by reducing coordination costs and misalignment.
Research consistently shows that team members with well-defined, focused roles outperform those with ambiguous or overly broad responsibilities. By giving each AI a tight scope, you prevent the “overgeneralization problem“ where AI makes too many assumptions and paints you into corners.
Management practices on “leading by context“ rather than “leading by control” show that teams perform best when given clear vision, constraints, and coordination mechanisms—then trusted to execute.
Principles Behind the Practice
Give each team member an extremely clear, focused role. Avoid ambiguous or overlapping responsibilities. AI performs dramatically better with tight constraints than broad mandates. A specialized architect produces better work than a generalist “helper.”
Maintain shared documents religiously. Architecture specs, design documents, daily standups, and reflections keep everyone aligned. When one team member creates something valuable, immediately add it to project files so others can build on it.
Work with your team, not through your team. Don’t chain tasks sequentially from one AI to another. Instead, have parallel conversations with whoever you need, when you need them—just like you’d walk between desks in an office.
Refresh team members proactively before they fail. Don’t wait until AI loses focus completely. When you notice degraded performance, reconstitute that thread. It’s faster to refresh early than to clean up confused work later.
How to Start
Begin with just two chat threads: one for questions and exploration, another for focused work. Don’t build the full team structure yet.
Notice when you’re diluting your focused thread with tangential questions—that tension signals you need separation. Then add a third specialist role for whatever keeps pulling you off-track.
Build your team gradually over weeks, adding roles only when the need becomes obvious.
Start each new team member with a simple incantation prompt (200 words maximum), then let them prove their value before expanding further.
After a month, you’ll naturally see what team configuration serves your actual work rather than an imagined ideal.
This Wall Street Journal article is an example of someone using a Synthetic Team.