Building an agentic AI system is a software engineering problem. Shipping it inside a large organisation is an organisational engineering problem. Most teams underestimate the second one.
What “agentic” actually means
An agentic system is one that takes multi-step actions in the world, using tools, making decisions, calling external services, in pursuit of a goal. It is not a chatbot. It is not a copilot. It is a system that acts.
The engineering implications are significant. Failures are non-deterministic. Side effects may be irreversible. The blast radius of a bug is larger than in a traditional CRUD system. Standard software engineering discipline is necessary but not sufficient.
The organisational surface
Inside a large organisation, the hardest problems are rarely the AI ones. They are:
-
Stakeholder confidence - Executives who have approved the budget want to see demos, not architectures. Demos are lies. Architectures are truth. You must translate one into the other continuously.
-
Data access - Agentic systems need access to internal tools and data. In a large org, that means internal APIs, permissions, rate limits, and political territory. The map of what the agent can call is different from the map of what it should call.
-
Observability- A traditional service call either succeeds or fails. An agent succeeds, fails, or produces a plausible-looking wrong answer with confidence. Your observability posture needs to distinguish all three.
-
Human-in-the-loop - The right abstraction is not “how do we remove the human” but “where does the human need to remain.” For high-stakes actions, the answer is often “everywhere, for now.”
What works in practice
Building incrementally. Narrowly scoped: one class of task, one data source, one action. Instrumented everything so you can monitor as much as possible. Run it in shadow mode and test it against deterministic questions.
The results will be incremental, classic ‘boring’ deployments. No incidents. No catastrophic failures. Incremental scope expansion.
Boring is the goal.