Definition
A human-agent squad is a small working team in which one human lead directs several specialized AI agents toward a shared business outcome. The human sets goals, guardrails, and escalation conditions. The agents handle sensing, pattern recognition, and repetitive execution.
The 1-to-5 model is the most common configuration: one human lead overseeing roughly five specialized agents. Greg Kihlström introduces both ideas in Marketing Operations 3.0 (De Gruyter Brill), where the squad is described as “the core working unit of the Commercial System.” It’s organized around outcomes, such as onboarding continuity or churn prevention, rather than channels like email or paid social.
See also: AI Agent · Marketing Operations · Next Best Action · Human-in-the-Loop (HITL)
Why it matters for marketing
Most marketing teams are still organized by channel or function. One person owns email, another owns paid search, a third runs the CRM. AI tools get added to each of those roles one at a time, and each person prompts their own tools.
That arrangement doesn’t scale well once agents start acting on their own. Five agents working independently, each optimizing its own channel, will eventually collide. A pricing agent pushes a discount while a retention agent is trying to protect margin. A campaign agent sends an upbeat offer to a customer with an open complaint.
A squad gives those agents one shared outcome and one accountable human. The book’s argument is that the skill that matters now is orchestration — designing how several agents, policies, and workflows work together — more than prompting any single tool. As the book puts it, before a company can orchestrate commercially, “the team itself has to become able to be orchestrated.”
How it works
The human lead
The human lead (the book calls this role the Bot Manager) is accountable for the squad’s results. The job includes:
- Defining the outcome and success criteria
- Setting authority limits and escalation thresholds
- Resolving conflicts between agents with competing objectives
- Auditing agent reasoning and correcting drift
- Stepping in on high-stakes moments, such as pricing exceptions, renewal disputes, or reputationally sensitive replies
The agents
The book’s example growth squad includes five agent roles, each with a distinct mandate:
| Agent | Standard mandate | In a Next Best Action environment |
|---|---|---|
| Content Agent | Real-time creative optimization; testing many ad and message variants | Generates or selects the message that best fits the customer’s current context |
| Data Agent | Cleans and integrates data; finds patterns humans might miss | Identifies the live signals that decide the next action |
| Campaign Agent | Runs always-on programs; adjusts send times and audience paths | Orchestrates timing, channel, and order of actions in real time |
| Value Agent | Customer value realization and post-purchase nurture | Protects retention and reduces effort; not every best action is a sale |
| Sales or Support Agent | Commercial progression or service resolution | Handles the transaction or service step the moment calls for |
The book is careful to say the specific roles aren’t the point. What matters is that each agent has a distinct mandate, the handoffs between agents are orchestrated, and the human stays accountable for overall performance.
Why “about five”
The ratio is a design guideline, not a rule. The squad needs to be small enough that the human can still exercise real judgment, with enough context to understand why each agent did what it did. The book describes the 1-to-5 model as “a practical unit of managed leverage.” Adding more agents helps only if oversight quality holds.
Supervised execution
Squads are especially useful for work that falls in the middle of the risk spectrum: too risky for full autonomy, too dynamic for rigid manual control, and too valuable to sit in slow approval chains. Agents move quickly within set thresholds, and the human handles escalations and edge cases.
How to utilize human-agent squads
- Pick an outcome, not a channel. Good first squads cover onboarding continuity, early churn prevention, plan or product recommendation, or post-purchase value realization.
- Name the human lead before deploying agents. Accountability should be clear from day one.
- Give each agent one mandate. Overlapping mandates are where agent conflicts start.
- Share context across the squad. Every agent should read from the same current view of the customer instead of keeping its own copy.
- Define escalation triggers in advance, based on deal size, sentiment, compliance exposure, or contract deviation.
- Scale by adding squads, not by stretching one. The book treats the squad as the basic unit for growing cross-functional orchestration under Commercial Operations Orchestration (CO2).
Comparison: human-agent squads vs. other team models
| Human-Agent Squad (1-to-5) | Spotify-model squad | Amazon two-pizza team | Channel-based marketing team | |
|---|---|---|---|---|
| Members | 1 human lead + ~5 AI agents | 6–12 people, cross-functional | Small enough to feed with two pizzas | Specialists by channel |
| Organized around | A business outcome | A product area or mission | A service or product | A channel (email, social, search) |
| Who executes | Mostly agents | Humans | Humans | Humans, with tools |
| Human role | Set goals, guardrails, escalation; audit agent reasoning | Build and ship | Build and own a service | Plan and run campaigns |
| Main risk | Ungoverned agent behavior or agent conflict | Coordination across squads | Dependencies between teams | Silos and handoff delays |
| Scales by | Adding squads | Grouping squads into tribes | Adding teams | Adding headcount |
The name echoes the squad structure that Spotify popularized in agile software teams, described by Henrik Kniberg and Anders Ivarsson in 2012. Spotify’s squads are made up of people. In a human-agent squad, most of the execution is done by agents.
How to measure a squad
- Outcome metrics tied to the squad’s purpose (activation rate, retention, resolution time, conversion)
- Return on Agent (ROA): whether the squad’s value exceeds the full cost of running and governing it
- Hybrid Squad Efficiency Ratio (HSER): how effectively the human lead expands productive output through the agents. The book treats this as a driver metric, not a replacement for ROA.
- Guardrail metrics: escalation appropriateness, drift rate, compliance exceptions, error recovery time
- Toil: time the human lead still spends on manual cleanup or ritual approvals, which should go down
Best practices
- Write down what each agent may decide on its own and what requires human review.
- Plan for agent conflict. Decide in advance how competing objectives (margin vs. retention, speed vs. caution) get resolved.
- Test agents in a controlled environment before giving them live authority.
- Keep the human lead’s span manageable, and expand only when escalation quality and guardrails stay stable.
- Give the squad cross-functional data access. A squad that can’t see service or product data will repeat the old silo problems.
- Review squad performance on a regular cadence, not only after incidents.
Future trends
- Squads as the default unit for AI-driven commercial work, replacing channel-based team structures in more marketing organizations.
- Reusable agents. The book notes that a single data agent may support campaign planning, sales prospecting, and post-purchase support. Agents may increasingly be shared across squads.
- Standard agent roles. Expect common agent “job descriptions” (content, data, campaign, value, service) to show up in vendor platforms.
- Larger ratios as tooling matures. Better monitoring and automated auditing may let one human oversee more agents, provided governance keeps pace.
- New performance reviews that evaluate the human lead on squad outcomes rather than personal output.
FAQs
What is a human-agent squad? A small team where one human lead directs several specialized AI agents toward a shared business outcome.
What is the 1-to-5 model? A squad design in which one human oversees roughly five specialized agents. Greg Kihlström describes it in Marketing Operations 3.0 as a practical unit for scaling AI-driven work without losing accountability.
Does it have to be exactly five agents? No. The number varies by business model and risk level. The principle is that the squad stays small enough for the human to supervise with real judgment.
What agents are typically in a squad? The book’s example uses a Content Agent, Data Agent, Campaign Agent, Value Agent, and a Sales or Support Agent.
Who leads the squad? A human lead, often called a Bot Manager or Agent Orchestrator, who is accountable for the squad’s results.
How is this different from a Spotify squad? Spotify’s squads are small cross-functional teams of people. A human-agent squad has one person, and most of the execution is done by AI agents.
What happens when two agents disagree? The human lead resolves it, ideally using rules set in advance. The book gives the example of a pricing agent and a retention agent with competing goals.
What kind of work suits a squad? Outcome-focused work that needs speed and continuity but still carries some risk, such as onboarding, retention, recommendations, and post-purchase support.
How do squads scale across a company? By adding more squads around more outcomes, coordinated through shared data, policy, and governance. In the book’s model, that coordination happens under Commercial Operations Orchestration (CO2) and its governing council.
Related Terms
- AI Agent
- Agentic AI
- Marketing Operations
- Next Best Action
- Human-in-the-Loop (HITL)
- Tuckman’s Stages of Group Development
- RACI Matrix
- Bot Manager
- Commercial Operations Orchestration (CO2)
- Hybrid Squad Efficiency Ratio (HSER)
Listen: related episodes of The Agile Brand
- Omnissa CMO Renu Upadhyay on balancing AI innovation with org chart disruption
- Readying an AI-augmented workforce with John Munsell, Bizzuka
- Expert Mode: Beyond the prompt — agentic AI and the new era of marketing orchestration
Sources
- Kihlström, G. Marketing Operations 3.0. De Gruyter Brill. https://www.degruyterbrill.com/document/doi/10.1515/9783112218839/html
- Kniberg, H., & Ivarsson, A. “Scaling Agile @ Spotify with Tribes, Squads, Chapters & Guilds.” October 2012. https://holub.com/goodies/SpotifyScaling.pdf
- Amazon Web Services. “Amazon’s Two Pizza Teams.” https://aws.amazon.com/executive-insights/content/amazon-two-pizza-team/