Science and education concept. AI (Artificial Intelligence).
getty
There’s an interesting book on AI about to come out, and if you’re looking for deep insights into how the technologies in play are impacting business, you might want to check it out. Satish Viswanathan has written “The Weight of Intelligence,” a weighty tome with a table of contents that looks like an indexed survey on LLMs in enterprise.
Viswanathan has AI bonafides, as former Managing Director at Accenture, and MIT ties, with a history of working with our academic labs in Boston. Now he’s scrutinizing the landscape of AI, and coming up with some applicable ideas about technology in our times.
The Pace and Nature of Change
In the early pages of the book, Viswanathan goes into some of the ways that neural nets are bringing us a new reality.
“Artificial intelligence is no longer advancing in isolated bursts,” Viswanathan writes. “Its capabilities are stacking. What once looked like separate breakthroughs now appears as a reinforcing sequence: better architectures, more data, more compute, broader interfaces, and then new layers of reasoning and action built on top of them. The story is no longer about one breakthrough at a time. It is about a system of breakthroughs that increasingly amplify one another.”
This concept, of concentric circles, rocks dropped in a great body of water, is instructive in helping us to see not only the scope of AI, but the shape of it, and how things will be different moving forward.
A Timeline
Viswanathan also presents a timeline of historic change, with the following: deep learning breakthroughs in perception tasks in 2012, transformer architecture in 2017, GPT progress circa 2020 and more successive GPT versioning through 2022–2024. He notes the rise of generative AI in public and enterprise experimentation over the past few years, and how reasoning-centric and agentic systems have begun combining language, planning, and action.
A Theory of AI
Another major pillar of Viswanathan’s evaluation of the AI era is that the analysis of AI that matters is not limited to token expansion, specifically, in the author’s words, that “it’s not capability, but absorption,” also noting the “metabolization” of AI in enterprise.
That is, the ability of LLMs to absorb intelligence, and work with it, agentically, you might say.
Viswanathan describes three categories: implementers, consumers, shapers, and illustrates how they differ.
“Each layer optimizes for something different,” Viswanathan writes. “Producers optimize for capability and market momentum. Implementers optimize for translation and deployment. Consumers optimize for reliability, governance, and economic value. Shapers influence how the field measures progress, legitimacy, and seriousness. The resulting tension is structural, not accidental.”
The Jagged Foundation as a Consequence
I found this to be another compelling thought from Viswanathan:
“The frontier is not smooth because organizational absorption is not smooth. Once framed this way, the jagged frontier becomes more than a description of uneven outcomes. It becomes a diagnostic lens. It helps reveal where capability is being translated effectively, where institutions are not yet ready, and where the ecosystem is producing more possibility than enterprises can realistically metabolize.”
That addresses that fundamental idea that AI is not just “good at” everything to the same generic degree: it produces magnificent successes and abject failures, all together.
Then, too, Viswanathan theorizes an “enterprise” flavor of AI analysis that differs from what people often talk about when they reference artificial general intelligence or the singularity, an alternative based on that absorption principle.
“Artificial General Intelligence asks whether machines can become broadly intelligent. Artificial Enterprise Intelligence asks whether enterprises can become intelligently organized around AI. AGI concerns the outer frontier of machine capability. AEI concerns the inner architecture of institutional value realization.”
And, if I can present a further sneak peek from the book, the ultimate conclusion:
“The current AI economy has been built largely around isolated solutions. Each solution can solve a bounded problem and create local movement, much like a spoke can carry part of the load. But to create shared direction, continuity of motion, and coordinated scale, those spokes must be bound into a wheel. Compounded enterprise cognition is that wheel. If a spoke-led AI era could already build a multi-trillion-dollar economy, a wheel-led era of compounded enterprise cognition could build the next economic age. The wheel is now in motion. The next AI era belongs to those who build the future with it.”
I found this to be a great read, and a real asset to those who must wrestle with the big, thorny problems of AI. The technology does, as Viswanathan suggests, have “weight.”
Stay tuned.
