6.2 The "Mitosis" Strategy: Growing Through Division

When AI portfolios need to grow, the instinct is often to make existing teams bigger. The AI Innovation model takes a different approach inspired by cell biology: successful pods divide into two smaller pods, each carrying the DNA of the parent but focusing on distinct products. This "mitosis" strategy preserves the two-pizza team benefits while enabling exponential growth.

Why Division, Not Growth?

Adding people to a team has rapidly diminishing returns. A 15-person "pod" has lost most benefits of the small team model. Instead of growing teams, we multiply teams. Each division creates two pods that can eventually divide again, enabling exponential scaling while maintaining the intimate team dynamics that make pods effective.

The Biology Analogy

How Cells Divide

In biological mitosis, a single cell divides into two daughter cells, each receiving a complete copy of the genetic material and the ability to function independently. The AI Innovation mitosis mirrors this:

Biological Mitosis Pod Mitosis
Cell grows beyond optimal size Pod exceeds two-pizza threshold or scope
DNA replicates Knowledge, culture, and practices transfer to both
Cell divides into two Pod splits into two distinct pods
Each daughter cell is complete Each new pod has full capabilities
Daughter cells can grow and divide again New pods can mature and eventually split

The Exponential Advantage

Mitosis enables exponential scaling:

1
Year 0: Pilot Pod
2-3
Year 1: First Division
5-8
Year 2: Second Wave
12-20
Year 3: Scaled

Division Triggers

When to Consider Mitosis

Pod division should be triggered by specific conditions, not arbitrary growth targets:

Size Trigger

Pod has grown beyond 10 people and coordination overhead is becoming noticeable.

Signal: Standups take too long, people don't know what others are doing

Scope Trigger

Pod is working on multiple distinct AI products that don't share significant dependencies.

Signal: Subgroups have formed with different focus areas

Domain Trigger

Pod expertise has grown to cover multiple distinct domains that could benefit from specialization.

Signal: Context switching between unrelated problem spaces

Success Trigger

Pod has proven the model and organization is ready to expand AI capabilities.

Signal: Demand for AI products exceeds single-pod capacity

Readiness Assessment

Before dividing, verify that conditions support successful mitosis:

The Division Process

Phase 1: Planning (4-6 weeks before)

1

Define the Split

Determine which products/capabilities will go to each pod. Draw clear boundaries to minimize ongoing dependencies.

2

Identify Leadership

Select the STO for the new pod. Ideally an experienced team member who has been groomed for the role.

3

Plan Team Composition

Determine which team members go to which pod, ensuring both have critical capabilities covered.

4

Prepare Charters

Create Model Cards and charters for any new products the new pod will own.

Phase 2: Execution (Division Week)

Day 1

Formal Announcement

Communicate the division to both pods and broader organization. Celebrate as a milestone of success, not a splitting up.

Day 2-3

Physical/Virtual Separation

Establish distinct team spaces, communication channels, and tooling configurations. Make the separation tangible.

Day 4-5

Knowledge Transfer

Intensive sessions to transfer any critical knowledge that isn't going with the person. Document thoroughly.

Phase 3: Stabilization (4-8 weeks after)

Critical Period

The weeks after division require careful attention:

  • Weekly check-ins between the two STOs to resolve any boundary issues
  • Escalation path to AI Council if pods struggle to collaborate
  • Permission to ask: new pod members should feel comfortable reaching back
  • Success metrics: Track both pods' health during transition

Common Division Patterns

Pattern 1: Product Split

Most common pattern when a pod has grown to support multiple AI products:

Product Split Example

Before: Single pod owns fraud detection AND customer churn prediction

After:

  • Pod A: Fraud Detection (original STO stays)
  • Pod B: Churn Prediction (new STO from senior team member)

Key consideration: Ensure both products have sufficient complexity to justify dedicated pods

Pattern 2: Domain Split

When a successful AI capability needs to expand to new domains:

Domain Split Example

Before: Pod owns recommendation engine for retail products

After:

  • Pod A: Retail Recommendations (established domain)
  • Pod B: Financial Products Recommendations (new domain, different regulations)

Key consideration: New domain often requires different expertise and governance approach

Pattern 3: Platform Extraction

When shared capabilities should become a platform serving multiple pods:

Platform Extraction Example

Before: Pod builds and operates product AND maintains MLOps infrastructure

After:

  • Pod A: Product pod (uses platform services)
  • Pod B: MLOps Platform team (serves multiple pods)

Key consideration: Platform team needs different operating model (see Section 6.3)

Pattern 4: Geographic/Market Split

When expansion requires region-specific versions:

Geographic Split Example

Before: Pod owns AI product for US market

After:

  • Pod A: US/Americas (original regulatory context)
  • Pod B: EU/APAC (GDPR, AI Act compliance, localization)

Key consideration: Significant regulatory differences often justify dedicated pods

Anti-Pattern: Escape Division

Warning

Division should never be used to escape problems:

  • Don't divide because of personality conflicts (fix the conflict first)
  • Don't divide a struggling pod (fix the problems first)
  • Don't divide to give someone a promotion (find real product boundaries)
  • Don't divide prematurely (ensure real need exists)

The Mitosis Mindset

Successful organizations build mitosis into their AI culture from the beginning. STOs are always developing future STOs. Team members expect that success leads to division. The goal isn't to build one amazing team—it's to build an organism that grows by multiplying its successes while preserving what makes it effective.