ctaio.dev Ask AI Subscribe free

AI Team Org Chart: Embedded, Centralized, or Hybrid

Three Real Structures at Three Company Stages

Stop debating org charts in the abstract. Here are three real structures that work at different company stages: what they look like, when they break, and when to transition to the next model. I've implemented all three at different points in my career, and the answer is always "it depends on where you are right now."

AI Team Org Chart: Embedded, Centralized, or Hybrid

30-second executive takeaway

  • 30-person startup: Embed 3 ML engineers directly in product. No separate AI team. Report to CTO.
  • 200-person scaleup: Centralize into a 12-person AI team with embedded liaisons. Report to Head of AI/CAIO.
  • 1000+ enterprise: Three-layer model. AI Platform (infra), AI CoE (governance), and embedded engineers in business units.

Not theory. Structures I've built and seen work.

Each structure is appropriate for a specific company stage. The transition points are clear. The failure modes are predictable. Most AI team design failures come from applying the wrong model for the current stage.

The 30-person startup

Headcount: 3 ML engineers embedded in product Reports to: CTO

Three ML engineers sit directly in the product team. No separate AI team. They attend the same standups, work from the same backlog, and deploy through the same CI/CD pipeline as everyone else. The CTO provides technical oversight and makes architecture decisions for AI workloads.

Roles

  • Senior ML Engineer (architect + IC)
  • ML Engineer (data + training)
  • MLOps Engineer (deployment + monitoring)

What works

Tight product feedback loop. ML engineers understand user context deeply. No coordination overhead between teams. Fast iteration — a model improvement goes from idea to production in days, not weeks.

What breaks

When you hire ML engineer number 4 or 5, there is no career path, no peer review, and no technical leadership specific to AI. The engineers start building ad-hoc infrastructure that is not production-grade because nobody owns AI platform concerns. Scaling beyond 5 requires a structural change.

The 200-person scaleup

Headcount: 12-person centralized AI team + embedded liaisons Reports to: Head of AI / CAIO

A dedicated AI team of 12 reports to the Head of AI or CAIO. The team owns the ML platform (model serving, feature store, experiment tracking), builds shared models, and sets standards. Two to three embedded liaisons sit in the highest-priority product squads, translating product needs into AI specifications and AI capabilities into product features.

Roles

  • Head of AI / CAIO
  • ML Platform Lead
  • Senior Applied ML Engineers (3-4)
  • Data Engineers (2)
  • MLOps Engineers (2)
  • AI Product Manager
  • Embedded Liaisons (2-3)

What works

Shared infrastructure eliminates duplication. Consistent model quality and evaluation standards. Career path for AI specialists. Peer review and knowledge sharing. The platform enables product teams to self-serve on simpler AI tasks (classification, recommendations) without waiting for the centralized team.

What breaks

If the centralized team becomes a bottleneck — product teams waiting weeks for AI features while the central team prioritizes platform work. If liaisons are perceived as "spies" rather than collaborators. If the team optimizes for technical elegance over product impact. Requires strong AI product management to prevent the ivory-tower failure mode.

The 1000+ enterprise

Headcount: AI Platform + AI CoE + embedded engineers in BUs Reports to: CAIO with dotted lines to CTO and CPO

Three distinct layers. (1) AI Platform team: builds and maintains the ML infrastructure — model serving, feature store, data pipelines, experiment tracking, LLM gateway. Reports to CTO. (2) AI Center of Excellence: sets standards, governance, runs audits, defines best practices, provides consulting to business units. Reports to CAIO. (3) Embedded AI engineers in each business unit: build AI features specific to their domain using the platform. Report to BU engineering leads with dotted line to CoE for standards compliance.

Roles

  • CAIO (executive)
  • VP AI Platform
  • VP AI CoE
  • Platform Engineers (8-12)
  • CoE Consultants (4-6)
  • BU-Embedded AI Engineers (varies)

What works

Scales to hundreds of AI practitioners across dozens of business units. Platform eliminates infrastructure duplication. CoE ensures consistency without being a bottleneck. Business units have autonomy to move fast on domain-specific AI while benefiting from shared infrastructure and standards.

What breaks

If the CoE becomes a paper-pushing governance body with no delivery mandate. If the platform team is under-resourced and business units build shadow infrastructure. If there is no clear escalation path when a business unit violates CoE standards. Requires strong executive sponsorship and clear decision rights documentation.

Who should the AI team report to?

The reporting line is not an administrative detail — it signals organizational priority. Where the AI team sits in the hierarchy determines what gets optimized and what gets sacrificed.

Reporting Line When It Fits Advantage Risk
Reports to CTO AI is primarily infrastructure. Model serving, data pipelines, ML platform. The AI team builds capabilities that product teams consume. Technical excellence. Strong alignment with engineering standards. Clean infrastructure. AI becomes disconnected from product value. Team optimizes for platform elegance, not business impact.
Reports to CPO AI IS the product. The company's core value proposition is AI-powered. Think: an AI-native startup where the model IS the product. Tight product-market fit. AI decisions driven by user value. Fast iteration on model improvements that users feel directly. Technical debt accumulates. Infrastructure is treated as second-class. Long-term platform health sacrificed for short-term feature velocity.
Reports to CAIO AI is a strategic cross-cutting capability. Multiple business units benefit from AI. Investment exceeds 15% of engineering headcount. Cross-BU coordination. Strategic investment allocation. Governance and standards from day one. AI treated as an enterprise capability, not a team hobby. Additional executive layer adds coordination cost. CAIO must have real authority (budget, headcount, veto) or the role becomes advisory theater.
Reports to CDO AI is primarily data-driven (analytics, recommendations, predictions) and the data organization is already strong. Tight coupling between data quality and model quality. Data infrastructure shared between analytics and ML. CDO organizations often lack engineering culture. Models may never reach production quality because the team is analytically-minded, not engineering-minded.

Frequently Asked Questions

Should AI engineers be embedded in product teams or centralized?
It depends on your stage. Below 50 engineers total, embed them in product teams — the coordination overhead of a separate AI team is not worth it, and AI work needs tight product context to deliver value. Between 50 and 200 engineers, centralize into a dedicated AI team with embedded liaisons in each product squad. Above 200, you need a hybrid: a platform team building shared infrastructure, a governance layer setting standards, and embedded engineers in business units consuming the platform. The mistake most organizations make is centralizing too early (creating an ivory tower) or staying embedded too long (creating fragmented, inconsistent AI implementations).
Who should the AI team report to: CTO, CPO, or CAIO?
If AI is infrastructure (model serving, ML pipelines, data platforms), it reports to the CTO. If AI is the product (the core value proposition is AI-powered), it reports to the CPO. If AI is a strategic capability being deployed across multiple business units, it reports to the CAIO or CDO. The reporting line signals organizational priority: CTO ownership says "AI is a technical capability." CPO ownership says "AI is a product differentiator." CAIO ownership says "AI is a cross-cutting strategic investment." Most growth-stage companies should start with CTO and transition to CAIO when AI investment exceeds 15% of total engineering headcount.
How many ML engineers does a startup need to start doing AI seriously?
Three. One senior ML engineer who can architect the system and make technology choices. One ML engineer focused on data pipelines and model training. One MLOps engineer who can deploy, monitor, and maintain models in production. Below three, you have single points of failure and no peer review. You can supplement with one AI product manager and share data engineering resources with the broader platform team. If you cannot afford three dedicated ML engineers, you are not ready for a custom AI investment — use third-party APIs and pre-trained models until you reach that scale.
What is an AI embedded liaison?
An AI embedded liaison is a senior ML engineer who sits in a product squad full-time but reports to the centralized AI team. They attend the product squad standups, understand the product context deeply, and serve as the bridge between product requirements and AI capabilities. They write feature specs that the centralized team can execute, translate model performance metrics into product impact language, and flag when a product team is asking for something that is technically infeasible. The liaison model prevents the centralized AI team from becoming an ivory tower disconnected from product reality while maintaining technical consistency and shared infrastructure.
When should a company transition from embedded AI to a centralized AI team?
Three signals indicate it is time to centralize: (1) You have more than 5 ML engineers across different product teams who are independently building similar infrastructure — data pipelines, model serving, evaluation frameworks. The duplication is wasting resources. (2) AI model quality is inconsistent across teams because there are no shared standards for evaluation, testing, or deployment. (3) You are hiring senior AI talent who want to work with other AI specialists — they will not join a team where they are the only ML engineer surrounded by product engineers. When two of these three signals are present, centralize.
·
Thomas Prommer
Thomas Prommer Technology Executive — CTO/CIO/CTAIO

These salary reports are built on firsthand hiring experience across 20+ years of engineering leadership (adidas, $9B platform, 500+ engineers) and a proprietary network of 200+ executive recruiters and headhunters who share placement data with us directly. As a top-1% expert on institutional investor networks, I've conducted 200+ technical due diligence consultations for PE/VC firms including Blackstone, Bain Capital, and Berenberg — work that requires current, accurate compensation benchmarks across every seniority level. Our team cross-references recruiter data with BLS statistics, job board salary disclosures, and executive compensation surveys to produce ranges you can actually negotiate with.

Design the team. Ship the AI.

Org design is the foundation. The AI Center of Excellence guide covers how to operationalize governance and standards once the team structure is in place.