What if you could clone your best analyst four times, point each clone at a different problem, and have all four deliver their findings simultaneously?
That is essentially what Claude Code Agent Teams does. And after running it on a real client project, we can confirm: it changes the way technical work gets done.
Agent Teams is the flagship feature in Anthropic's latest Claude Code update, and it is the one we have been most eager to test. We ran it against a genuine multi-site SEO project for a plumbing franchise across Sydney and Melbourne - and the results were remarkable enough that we rewrote our internal workflows the same week.
Here is everything we learned.
Agent Teams Guide
Part 1: Multi-Agent Editing
Run parallel AI agents on real projects
What Are Claude Code Agent Teams?
Agent Teams lets you spin up multiple AI agents that work on different parts of a project simultaneously, coordinated by a single team lead agent.
Think of it like a project manager running a kick-off meeting. The team lead reads the brief, breaks it into workstreams, assigns each workstream to a specialist teammate, then monitors progress and synthesises the results.
How the Architecture Works
The system has three layers:
- Team Lead - The orchestrator. It reads your instructions, decides what sub-tasks are needed, delegates them to teammates, and compiles the final output.
- Teammates - Independent agents that each receive a focused brief. They can read files, run commands, search the web, and produce deliverables - all within their own context window.
- Communication Layer - Teammates report back to the team lead with their findings. The lead can ask follow-up questions or request revisions before accepting their work.
Each teammate operates in its own thread with its own context. This means they do not pollute each other's thinking or run into context window limits from unrelated information.
Agent Architecture: Parallel Dispatch
Real Case Study: Auditing a Plumbing Franchise Network
We did not want to test this on a toy example. We wanted to know if Agent Teams could handle the kind of messy, multi-dimensional work that actually fills our client engagements.
The Brief
Our client, Mr Splash Plumbing, operates a franchise network with over 20 microsites across Sydney and Melbourne suburbs. Each site targets local keywords like "plumber Parramatta" or "emergency plumber St Kilda."
The problem: the network had grown organically, and nobody had done a comprehensive audit across all sites simultaneously. Some sites were performing well. Others were practically invisible. We needed to understand the full picture - fast.
The Team We Built
We set up a team of four specialised agents:
| Agent | Role | Workstream |
|---|---|---|
| GSC Analyst | Google Search Console data analysis | Indexation health, crawl issues, performance trends |
| Location Optimiser | Suburb page content analysis | Content duplication, local relevance, optimisation gaps |
| SEO Researcher | Backlink and keyword analysis | Link profile health, ranking opportunities, spam detection |
| Dev Planner | Technical audit and task mapping | WordPress configuration, site speed, development roadmap |
Each agent received a focused brief with access to the project's data files, API credentials, and relevant frameworks.
What Happened Next
All four agents started working at the same time. While the GSC Analyst was pulling indexation data, the Location Optimiser was comparing content across suburb pages, the SEO Researcher was analysing backlink profiles, and the Dev Planner was crawling site configurations.
Fifteen minutes later, we had seven deliverables on the table.
The Findings
GSC Analyst discovered a critical indexation crisis:
- 300 pages across 7 sites were blocked from Google by a WordPress "Discourage search engines" checkbox that someone had ticked and forgotten about
- The overall index rate across the network was just 56.3%
- Dozens of pages that should have been ranking were completely invisible to search engines
Location Optimiser uncovered a massive duplication problem:
- 97% duplicate content across 43 suburb pages
- The same generic plumbing copy had been pasted across nearly every location page, with only the suburb name swapped out
- An 8-phase remediation plan was produced, prioritised by traffic potential and competitive opportunity
SEO Researcher mapped the full link profile:
- 63,100 total backlinks analysed across the network
- 4 spam domains identified that needed to be disavowed
- 11 keywords sitting in position 2 or 3 - just one spot away from page one, representing quick-win opportunities
Dev Planner created a comprehensive roadmap:
- 12 pending development tasks mapped, totalling 48 to 70 hours of work
- Independently flagged the same WordPress noindex issue the GSC Analyst found
- Prioritised tasks by impact and dependency order
The Validation Effect
Here is what made us sit up: two agents working on completely different workstreams, with no communication between them, independently identified the same critical issue - the WordPress "Discourage search engines" checkbox.
The GSC Analyst found it through indexation data patterns. The Dev Planner found it through WordPress configuration audits. Neither knew what the other was working on.
This kind of independent corroboration is something you normally only get from expensive, time-consuming parallel review processes. Agent Teams delivered it as a natural byproduct of parallel execution.
Each Agent Gets a Clear Scope
The key to multi-agent success is separation. Each agent owns its own files, its own context, and its own deliverable. No overlap means no conflicts.
How to Set Up Your First Agent Team
Setting up an Agent Team in Claude Code is straightforward. Here is the process we use.
Step 1: Define Your Workstreams
Before involving any AI, write down the distinct workstreams your project needs. Good workstreams are:
- Independent - Each can proceed without waiting for another
- Focused - Narrow enough for one agent to handle thoroughly
- Concrete - Clear deliverable at the end
Bad workstreams are vague ("look into SEO stuff") or deeply interdependent ("implement the frontend, but wait for the backend agent to finish the API first").
Step 2: Write the Team Brief
Your prompt to Claude Code should describe:
- The overall project goal
- Each teammate's role and specific task
- What files or data each agent needs access to
- What format you want the deliverables in
Here is a simplified example:
Create a team to audit our client website:
Team of 3:
1. Technical SEO agent - crawl the site, check robots.txt,
sitemap, page speed. Output: technical-audit.md
2. Content analyst - review the top 20 pages for keyword
targeting, thin content, duplication. Output: content-audit.md
3. Competitor researcher - analyse the top 3 competitors
for link gaps and content gaps. Output: competitor-report.md
Project files are in shared/projects/client-name/
Step 3: Let the Team Lead Orchestrate
Once you submit the brief, the team lead agent takes over. It will:
- Parse your requirements
- Spawn the teammate agents
- Assign each one their workstream
- Monitor progress
- Compile results
You can watch the progress in real time. Each teammate's work appears in the Claude Code interface as it happens.
Step 4: Review and Iterate
When the team finishes, review the deliverables. If something needs more depth, you can ask the team lead to send a follow-up task to a specific teammate, or spawn a new agent for additional investigation.
Parallel vs Sequential: The Time Advantage
When Agent Teams Make Sense (and When They Don't)
Ideal Use Cases
- Multi-site audits - Like our plumbing franchise case study
- Codebase refactoring - One agent handles types, another handles tests, another handles documentation
- Research projects - Multiple agents investigating different aspects simultaneously
- Content production - We used Agent Teams to write this very blog series you are reading (more on that in a moment)
- Data analysis - Different agents analysing different data sources or metrics
When to Stick with a Single Agent
- Simple, linear tasks - If steps must happen in sequence, parallelism adds overhead without benefit
- Tiny tasks - Fixing a typo does not need a team
- Highly interdependent work - If every step depends on the previous step's output, teammates will spend most of their time waiting
The Meta Example
We used Agent Teams to produce this blog series. One agent wrote the hub post. Other agents wrote spoke posts in parallel. Each agent had the series brief, the keyword research, and the brand guidelines - and they all worked simultaneously.
The result: a complete content series drafted in a fraction of the time it would take a single writer working through each post sequentially. Every post maintains consistent messaging because every agent started from the same brief.
Solo vs Team
1 agent, sequential tasks, hours
4 agents, parallel execution, 15 min
Token Cost and Practical Considerations
Agent Teams is powerful, but it is not free. Here is what you should know about the economics.
Token Usage Scales Linearly
Each teammate consumes tokens independently. A team of four agents uses roughly four times the tokens of a single agent. There is no magic compression - each agent needs its own context to do its work.
When the Cost Is Worth It
For our plumbing franchise audit, the token cost was a fraction of what the equivalent human labour would have cost. Four analysts working for a full day versus 15 minutes of parallel AI execution - the maths speaks for itself.
The cost calculus favours Agent Teams when:
- Time is valuable - Parallel execution saves hours or days
- The work is genuinely parallelisable - Agents are not waiting on each other
- Accuracy matters - Independent corroboration catches errors that sequential work misses
- The project is complex enough - Simple tasks do not benefit from team overhead
Tips for Managing Costs
- Be specific in your briefs - Vague instructions cause agents to explore unnecessarily
- Scope teammates tightly - Each agent should know exactly what to deliver
- Start small - Test with 2 agents before scaling to 5
- Use Plan Mode first - Map out your workstreams before spawning the team (more on Plan Mode in our companion post)
Solo Agent Workflow
- Sequential task execution
- One perspective at a time
- Context switching overhead
- Hours to cover all angles
Agent Team Workflow
- Parallel execution
- Multiple perspectives simultaneously
- Dedicated context per agent
- Minutes to cover all angles
What This Means for Agencies and Teams
Agent Teams is not just a developer tool. It is a force multiplier for any team doing complex knowledge work.
At Jordan James Media, we have integrated Agent Teams into our audit and research workflows. A project that previously required three team members working across two days can now be scoped, executed, and compiled in under an hour.
This does not replace our team. It amplifies them. Our strategists spend less time on data gathering and more time on the analysis and creative thinking that actually moves the needle for clients.
For agencies evaluating AI tools, Agent Teams represents a step change. It is the difference between having one AI assistant and having a coordinated AI department.
Mr Splash Audit: Results Dashboard
Key Takeaways
- Agent Teams runs multiple AI agents in parallel, coordinated by a team lead that delegates, monitors, and compiles results
- Real-world tested - We ran it on a 20+ site plumbing franchise and got 7 deliverables in 15 minutes
- Independent corroboration - Two agents flagged the same critical issue without communicating, providing built-in quality assurance
- Token costs scale linearly - A team of 4 costs roughly 4x a single agent, but the time savings often justify the spend
- Best for parallelisable work - Audits, research, content production, and multi-file refactoring are ideal use cases
- Plan your workstreams first - Use Plan Mode to scope the team before you launch it
- We used it to write this series - The content you are reading was produced by an Agent Team working in parallel
Setting Up Your First Agent Team
Ready to Put AI Teams to Work on Your Projects?
Jordan James Media is a Sydney-based digital marketing agency that uses AI-first workflows to deliver faster, more thorough results for Australian businesses. If you want to see what Agent Teams and advanced AI tooling can do for your marketing, we would love to chat.
Talk to Us About AI-Powered Marketing
Related Reading:
- Claude Code's Biggest Update: 7 Features That Changed How We Build
- Claude Code Plan Mode: Think Before You Build
- Claude Code Checkpoints: Never Lose Your Work Again
Launching Parallel Agents
Each agent runs in its own terminal with a focused task. They work simultaneously without interfering with each other's file operations.
Social Media Carousel
7 cards — Scroll to browse, click any card to download
Agent Teams Guide
Run parallel AI agents on real projects
GSC Analyst, Location Optimiser, SEO Researcher, and Dev Planner — all working at once
Solo vs Team
1 agent, sequential tasks, hours
4 agents, parallel execution, 15 min
Best Use Cases
Coordination Strategy
Give each agent a distinct scope. Overlap causes conflicts — separation enables speed.
Key Insight
Agent Teams turn hours into minutes by parallelising what used to be sequential.
Scale Your AI Ops
Let us build your agent workflow
Share This Article
Spread the knowledge
Download as PDF
Save this article for offline reading