Agentic Intelligence Book Review: Can Smarter AI Fix Our Data Silos?
A Book About AI That Begins With Something Less Glamorous: Data
What happens when organisations have plenty of data but very little usable intelligence?
That question stayed with me while reading Agentic Intelligence: Strategy at the Speed of Data by Manish Sood and Venkat Venkatraman. Rather than presenting AI as another shiny technological tool, the authors begin with a more fundamental problem: information remains scattered across systems that were never designed to work together.
The book's central argument is that organisations entering the Intelligence Age need connected, real-time data before they can expect intelligent agents to deliver meaningful results.
But for me, the most interesting part was not simply the promise of agentic AI. It was the tension between faster intelligence and imperfect foundations.
From Vertical Filing to Data in Motion
The book's discussion of older information architectures reminded me how deeply organisations inherited the logic of the Industrial Age.
The industrial model separated work into functions, departments and systems. Information could remain vertically organised because humans were expected to navigate those boundaries.
The Intelligence Age demands something different.
The authors repeatedly return to the idea of connected data and what happens when information can move across organisational boundaries. Their discussion of datagraph principles makes this shift easier to understand: data becomes more valuable when its relationships and context are preserved rather than trapped inside isolated repositories.
That also explains the book's emphasis on cognitive latency. The problem isn't always that data doesn't exist. Sometimes the problem is the time required to find it, connect it and understand it.
The idea of movable data therefore becomes crucial. Intelligence cannot move at the speed of an agent if the information it needs remains locked inside legacy systems.
The Mercury Insurance Example: When Systems Become Silos
The Mercury Insurance example particularly brought this problem to life for me.
Scattered systems are not merely inconvenient. They can affect how quickly an organisation understands a changing situation. The experience of businesses during the COVID-19 emergency made this issue especially visible: suddenly, organisations needed information, employees and processes to function across physical and technological boundaries.
This is where Agentic Intelligence becomes more than an AI book. It becomes a book about organisational design.
The authors' argument is not simply “replace old technology.” It is about creating an environment in which information can actually participate in decision-making.
Progress Over Perfection—But What Happens When the Machine Is Wrong?
Here is where I found myself questioning the book's optimism.
The technology world often celebrates progress over perfection. And experimentation is necessary.
But agentic systems introduce a difficult question:
What happens when the information entering the system is wrong?
An AI agent can be remarkably fast and still be wrong.
The book's discussion of AI hallucinations therefore caught my attention. The examples involving the digital Heart-o-meter and smarter workplaces raise an important issue:
understanding that an AI system produces an incorrect output is not enough. We also need to understand where the error entered the chain.
Was the data incomplete?
Was the context missing?
Was the prompt inadequate?
Or was the underlying information itself wrong?
This is where I think the book's discussion becomes particularly relevant to the real world. Better AI cannot compensate indefinitely for unreliable foundations.
Phantom Braking: When Human and Machine See Different Realities
The phantom-braking example was one of my favourites because it makes this distinction tangible.
A driver may see a clear road while an automated system interprets something in its environment as a potential obstacle. The result can be unexpected braking.
For me, this works as an excellent analogy for the difference between hallucination and distorted perception.
The machine does not necessarily need to invent something from nothing. It can also act upon a version of reality that differs from the human interpretation.
That raises a much bigger question for agentic AI:
What exactly does the agent see, and what are we failing to see about what it sees?
Coagency Needs More Than Speed
This brings me to the concept I found most compelling in the book: coagency.
The future described by Sood and Venkatraman is not simply one where humans disappear and agents take over. It is a future where humans and intelligent systems work together through connected data and shared context. The publisher similarly describes the book as a roadmap for organisations seeking to move beyond outdated systems and put real-time intelligence to work.
I like that vision.
But I also think coagency requires something the technology cannot manufacture by itself: trust.
Responsible data sharing, behavioural trust, systemic accountability and error management still need considerable evolution.
Industrial or Intelligent age?
Industrial
Intelligent!
Both!
What are they?!
My Takeaway
Agentic Intelligence is an ambitious and highly futuristic book, but its strongest contribution, in my view, is not predicting what AI agents will eventually do. It is asking organisations to examine whether their data infrastructure is ready for them.
I particularly appreciated the discussions of legacy-system isolation, cognitive latency, data in motion, hallucinations, coagency and the human-machine perception gap. The numerous real-world scenarios make complex technological ideas considerably easier to grasp. The publisher describes the book as a practical guide supported by real-world examples, and that is one of its strongest qualities.
Yet I finished the book with one lingering concern:
we may be moving towards intelligent agents faster than we are learning how to recognise their errors.
Agentic AI could revolutionise efficiency and decision-making. But before we allow intelligent systems to act at scale, we need controlled environments where their mistakes can be studied, traced and corrected.
Because in the Intelligence Age, having more intelligence is not enough.
Comments