Industries are under pressure to turn Generative AI from experimentation into tangible business impact. This article introduces the concept of the Atomic Delivery Team as a proven model for delivering scalable, enterprise-grade Generative AI solutions end-to-end. It outlines the critical roles, engineering capabilities and operational principles required to move beyond prototypes and achieve sustainable value creation.
In the context of Generative AI projects, success does not come from the technology itself, but from the team building the AI system. We therefore explore the roles and skills required within an Atomic Delivery Team and highlight how these capabilities evolve when moving from classic GenAI systems to more complex Agentic AI architectures.
What Is an Atomic Delivery Team?
Boyan Angelov introduces the concept of the “Atomic Team” in his book Elements of Data Strategy as an abstraction representing the smallest functional unit of roles required to complete all tasks in a data project.
We extend this concept to the delivery of Generative AI projects and define an Atomic Delivery Team as a cross-functional, self-contained group that possesses all the skills required to deliver a discrete, valuable AI product or service end-to-end.
Core Roles in an Atomic Delivery Team for Generative AI Projects
At HMS, we typically identify the following core roles in our Generative AI projects, each of which is usually represented within an Atomic Delivery Team:
AI Architect: Designing Scalable, Enterprise-Ready AI Systems
AI Architect – Defines the AI system architecture, ensures scalability and system quality, and leads early prototyping.
Rationale for the role:
- A recent MIT report observed that most GenAI projects fail. Primary reasons include brittle workflows, flawed implementation, lack of scalability, and poor integration with IT systems.
- We believe successful GenAI projects require architectural foresight. This role demands not only excellent AI expertise but also experience in cloud infrastructure, software and code quality, data engineering, requirements engineering, and user adoption.
- A testament to this approach is our very first GenAI project launched in spring 2023, which remains productive and scalable. Its ability to incorporate new features and model updates, and to integrate with existing systems, is due to the strong architectural foundation established at the outset.
A I Engineer: Building Reliable GenAI Workflows & Model Orchestration
AI Engineer – Implements the system by combining prompt engineering, model orchestration, and data integration.
- The "workhorse" of the project, responsible for implementing the AI workflows.
- Strong development skills are essential to translate well-designed architecture into high-quality code and avoid technical debt, ensuring the project does not join the 95% of failed GenAI initiatives noted in the MIT report.
AI Ops / Cloud Engineer: Operating and Scaling AI Systems in Production
Cloud-/AI Ops Engineer – Builds cloud infrastructure, CI/CD pipelines, and ensures the system operates efficiently and is scalable.
- AI systems typically run in the cloud.
- Responsibilities include managing model endpoints (updates, cost, performance), maintaining supporting infrastructure such as vector stores and UI components, and automating deployments and development workflows.
Frontend Engineer: Driving User Adoption Through UX
Frontend Engineer – Designs intuitive user interfaces and integrates AI capabilities into business applications.
- One key difference between traditional ML and GenAI projects is the higher frequency of use cases requiring direct user interaction and UI components.
- MVPs can often be built using frameworks like Streamlit, Chainlit, or Gradio by AI Engineers. However, as applications scale in complexity and UX expectations increase, dedicated frontend technologies and expertise are required.
- Customers may be hesitant to onboard frontend developers, but we consistently observe that a well-designed, user-centric feature can have a greater impact than minor backend optimizations.
How Cross-Functional Teams Improve AI Delivery
Given this, a natural question arises:
Does every GenAI project require four separate individuals to cover these roles (or even more, if extended roles are considered)?
Obviously not. Ideally, team members have overlapping skill sets.
For example:
- For example, an AI architect can typically take on the role of an AI engineer and, ideally, also cover the basic responsibilities of an AIOps engineer. To be an architect, one should possess a broad spectrum of experience and skills anyway, having gotten their hands dirty in the past - and likely still doing so from time to time.
- A Frontend Engineer with broader training can contribute to AI system code or UI infrastructure.
- For smaller projects (e.g., 2 FTEs), staffing four people at 0.5 FTE each may cause unnecessary friction. A “Swiss Army Knife” AI Engineer is more effective - not implying the AI role is superficial, but that a successful AI Engineer must look beyond the AI component alone.
We believe this cross-functional approach contributes to the high production rate of our GenAI projects - our consultants bring the technical skills required to deliver not just MVPs, but full production-ready systems.
Depending on project scope, extended roles may include:
- Data Scientist – for analytics and hybrid AI approaches
- Data Engineer – for building and managing data pipelines
- Software Engineer – for broader system integration
- Project Lead – for stakeholder management and coordination
What Changes with Agentic AI Systems? Architectural & Team Implications
For Agentic AI systems, the above roles remain relevant for each individual agent. However, additional considerations arise:
- Communication and coordination between agents
- Standardization and API design
- Telemetry and monitoring
- Event-driven architectures
- (Role-based) access control
- Security and fallbacks
- Increased infrastructure demands
- Infrastructure as Code
Key Takeaways for AI Delivery in Enterprise Environments
This increases the importance of the Engineering and Architecture elements in the AI Engineer/Architect role, as well as the Ops component in the AIOps role. While AI expertise remains essential, the range of potential “full stack” implementation capabilities (or even better experience) a developer can offer is becoming increasingly important
Important distinction: These observations refer specifically to the roles of AI Engineer/Architect in implementation projects, not in AI research.
