Deep-Dive into AI
A Step-by-Step Guide on Designing Agentic AI Systems
What will you learn?
Agentic AI represents a fundamental shift in how systems are conceived, built, and operated. Instead of designing static applications that follow predefined logic, organisations must now engineer dynamic ecosystems where autonomous components interpret goals, make decisions, and act within governed boundaries.
This transition requires rethinking architecture, control, validation, security, and scale at every layer of the stack. Our 16-part guide, written by Alison Hodivala, AI & Data Portfolio Strategy Director of Bell Integration, will guide you through the essentials of designing agentic AI systems – from frameworks and design principles to structural elements that make autonomous behavior reliable and safe.
Skip to:
Part 1: Understanding the Agentic AI Shift
Organisations that move quickly into agentic AI without the right foundations tend to encounter a familiar set of problems. These are not edge cases. They are the predictable result of deploying systems that act, make decisions, and operate across time without the structures required to guide them.
In the article, we dive into:
- The shift from AI Assistants to AI Agents
- What makes agentic AI systems different
- What breaks without the right foundations
- A different kind of engineering challenge
Part 2: Designing the Agent Control Plane
The simplest way to think about the Control Plane is that it sits between the agent’s raw capability and the real world. The agent may be capable of many things. The Control Plane determines what it is permitted to do, in what sequence, under what conditions, and with what level of oversight. A common concern when designing Control Planes is that governance will constrain capability. If every action requires approval, autonomy disappears.The goal is not maximum control. It is appropriate control. A well-calibrated Control Plane applies oversight where the risk warrants it and steps back where it does not.
In this article, we dive into:
- What the Control Plane does
- Why this matters in practice
- The relationship between control and capability
- What good looks like
Part 3: Designing the Agent Data Plane
The Agent Data Plane is the architectural response: retrieval and knowledge access (across vector/semantic and structured sources), grounding to verified sources, state/memory for long-horizon tasks, and structured validation against domain rules. It ensures that the agent is working with information that is current, relevant, and reliable. When the Data Plane is poorly designed, the problems are predictable. Agents act on stale information because retrieval is not connected to live systems. A well-designed Data Plane ensures that an agent always knows where its information came from, how current it is, and how confident it should be in it.
In this article, we dive into:
- The problem with model knowledge alone
- What the Data Plane provides
- The role of specialised models
- What breaks without a Data Plane
- What good looks like
Part 4: Small and Domain-Specific Language Models
One of the most common assumptions in early agentic AI deployments is that more capability is always better. Bigger models increase capability. They do not guarantee better systems. Small and domain-specific language models are not a compromise. They are a deliberate architectural choice that makes agentic systems more accurate, more efficient, and easier to validate.
In this article, we dive into:
- What we mean by small and domain-specific
- Where they fit in an agentic pipeline
- The trade-offs worth understanding
- What this means for organisations
Part 5: Designing Multi-Modal Agentic Systems
Multimodal agentic systems extend an agent’s perceptual capability. An agent that can see as well as read, interpret a chart as well as a spreadsheet, or process an audio recording as well as a transcript operates in a fundamentally richer context. This expands what is possible. It also introduces a new class of architectural challenge.
In this article, we dive into:
- The case for multimodal agents
- What changes architecturally
- The failure modes specific to multimodal systems
Part 6: Scalable Agent Architectures
Most organisations begin their agentic AI journey with a single agent solving a single problem. Simply building more agents does not produce an ecosystem. It produces a proliferation. More agents without architecture create fragmentation, not capability. Scaling agentic AI is not just a technical challenge. It is an architectural one. And it needs to be approached earlier than most organisations expect.
In this article, we dive into:
- Why more agents is not the same as greater capability
- The architectural patterns that enable scale
- The governance challenge at scale
- What good looks like
Part 7: Tooling Architecture for Agentic Systems
The capability that makes agentic AI genuinely transformative is the ability to act: to query live systems, update records, trigger workflows, send communications, and interact with the applications and services that organisations run on. Tools are what give agents that capability.
In this article, we dive into:
- What a tool actually is
- The components of a well-designed tooling architecture
- The testing dimension
- What this means for organisations
Part 8: Testing Agentic Systems
Confidence in an agentic system comes from a combination of structured testing, visibility into behaviour, and the ability to detect and respond to failure. Organisations need to think more carefully about what testing is intended to achieve and what confidence means in this context.
In this article, we dive into:
- Why agent testing is different
- The components of an agentic testing framework
- Building confidence in practice
- The Maturity of Testing Practice
- What confidence actually looks like
AI Glossary:
What is AI Observability?
In simple terms, AI observability is the ability to see inside your AI systems, understanding not just whether they are running, but whether they are working as intended, making sound decisions, and delivering the outcomes your business and customers expect. It is the difference between knowing your AI is on, and knowing your AI is working. Learn more about AI observability.
What is Agent Control Plane (ACP)?
The Control Plane is the layer that governs how an agent interprets goals, makes decisions, and takes action within defined boundaries. It is not a single component, but a set of mechanisms that together shape how an agent behaves within the limits an organisation has set. Learn about Agent Control Plane.
What is Data Plane?
The Data Plane is the layer that governs what an agent knows and how it uses information. It manages how data is retrieved, validated, grounded, and maintained throughout a task. It ensures that:
- the agent works with current and relevant information
- outputs are anchored in real data, not assumptions
- context is maintained across multi-step workflows
In simple terms, the Data Plane determines what the agent knows and how reliable that knowledge is. Learn more about Data Plane.
Navigation
Navigate through the individuals sections of the full series here.
- 1. Understanding the Agentic AI Shift
- 2. Designing the Agent Control Plane
- 3. Designing the Agent Data Plane
- 4. Understanding Small and Domain-Specific Language Models
- 5. How to Design Multi-Modal Agentic Systems
- 6. Scalable Agent Architectures
- 7. Tooling Architecture for Agentic Systems
- 8. Testing Agentic Systems
