Book Review: Reshuffle by Sangeet Paul Choudary

I have been wondering how AI will change jobs. Most of what I've read on this has been disappointing.

Most research goes like this:

Now, many researchers conclude that there is a 48% chance that Data Scientists are redundant, or that 48% of Data Scientists are not needed anymore. I would argue that this does not work.

• What if you only have one Data Scientist? Will you fire half of them?

• What if you need the full expertise, but only 52% of the work?

• What if your Data Scientist was not cleaning that much data anyway? Or if their strength was in other areas?

• What if the freed-up time doesn't disappear but gets filled with new tasks that didn't exist before? Prompt engineering, evaluation pipelines, AI governance. The role doesn't shrink, it reshapes.

Breaking jobs into tasks and checking which ones AI can do is just Taylorism with a new label. I was looking for a more holistic view. How will roles look in the future? Which ones split apart, which new ones emerge, which new tasks fill the gaps?

Sangeet Paul Choudary's Reshuffle is the first book that gave me a framework I actually found useful. Choudary's earlier book Platform Revolution (2016, co-authored with Geoffrey Parker and Marshall Van Alstyne) was one of the foundational works in platform economics, so I had high expectations.

What the Book Is About

The Big Reframe: Coordination, Not Just Automation

The central argument of Reshuffle is simple: we've been thinking about AI's economic impact through the wrong lens. Most discussions focus on AI as a tool for automation, taking over tasks and processes. Choudary argues that the far bigger impact will come from AI as a tool for coordination, enabling components of an economy to specialize, align, and scale in ways that weren't previously possible.

His anchor metaphor is the standardized shipping container that didn't just make loading ships faster. It enabled outsourcing to low-wage countries, allowed companies to specialize, created entirely new port economies, for example Singapore, and fundamentally reshaped global trade. The container wasn't primarily an automation, but a coordination technology.

Choudary argues AI will do something similar for knowledge work, by taking over some of aspects of coordination:

• Unified representation: AI can establish a structured view of a situation across fragmented information sources.

• Decision support: Analyzing trade-offs and either making decisions (agentic) or supporting human decision-makers (assisted).

• Execution: AI is able to exectute certain workflows or agentic scenarios.

• Composition: In the past, agreeing on common standards between organizations was tedious and expensive. AI can bridge those gaps dynamically.

• Governance: Setting up shared governance frameworks used to require painful negotiations. AI can do this more flexibly, based on preferences.

This has real strategic implications. If you only see AI as automation, you optimize existing processes. If you see it as coordination, you start asking different questions entirely: what new organizational forms become possible? What previously impossible collaborations can now work?

Unbundling and Rebundling — But Around Constraints, Not Tasks

The second hypothesis is that the two most common frameworks are too simplistic to explain what will happen. He argues for the following process:

• Existing roles unbundle as AI erodes the value of their component tasks.

• They rebundle around new constraints: for example, where things are scarce or where someone needs to be accountable.

The unbundling is based on a value decrease. The role-based view says AI will replace entire roles, like programmers, designers, analysts. The task-based view says roles are bundles of tasks, and AI will replace specific tasks within them, for example writing unit tests or cleaning data.

Choudary goes further than both. He introduces a distinction between economic value and contextual value of tasks. Economic value is straightforward supply-and-demand: if AI can do a task, the scarcity premium drops. Even when AI can't fully replace a task, enabling more people to do it "good enough" will change the value. One of his examples is how GPS navigation depleted the value of knowing your way around a city and taxi drivers took a hit.

Contextual value is more subtle and, I think, more important. It's the value a task carries because of where it sits in a workflow. Great example: a student taking notes has lower contextual value than a CEO taking notes. Same task, vastly different value. Contextual value can arise from internal company knowledge, expensive intra-role communication, or simply the decision-making context surrounding the task. (Although Choudary is not so clear here.)

Choudary's key insight: the drop in economic value is well captured by the task-based view. But contextual value is often overlooked, and it can be much more impactful. The question is where contextual value will concentrate. Choudary's answer: look at the constraints.

He identifies three types of constraints that will shape where roles rebundle:

• Scarcity-based constraints: Something remains genuinely scarce (supply and demand still applies).

• Risk-based constraints: Someone needs to be responsible. Accountability doesn't automate easily.

• Coordination-based constraints: Communication between teams, organizations, systems still needs human judgment.

From My Personal Experience: Future of Software Engineering

To make this concrete, I've been applying Choudary's framework to software engineering, a domain I know well from my own work.

Previously, a developer would look at specs, design a solution, write code, and test it (among other stuff). The full bundle. Now, AI coding tools are restructuring that bundle. AI can understand large codebases, generate solutions, and handle much of the routine coding work.

So where do the constraints land?

• Scarcity: We can now build far more things than we can actually verify. Security audits, best-practice reviews, deployment validation. These become the bottleneck, not the coding itself.

• Risk: We still want a responsible, knowledgeable human in the loop, especially for production systems where failures have real consequences.

The effect is a general trend toward specialization. We no longer need a single developer who holds the entire codebase and context in their head. But constraints are building around security, responsibility, platform engineering, and fixing things at scale. The coding itself becomes less valuable; the surrounding constraints become more so.

A lot of developers I talk to think the smart move is to go "up the stack": become an architect, focus on system design. Others double down on becoming the developer who is best at steering AI to write code. Both feel logical from a task-based perspective. But AI is getting increasingly good at both: understanding how large systems fit together and generating the code itself. The constraint-based move is the opposite: go toward verification, security, reliability. The things that become scarce precisely because building gets easy.

This isn't from the book, it's my own extension of Choudary's framework. But I think it illustrates the power of thinking in constraints rather than tasks.

What This Means for Organizations

The third major thread in Reshuffle addresses how companies themselves will change. Organizations face a constant tradeoff between autonomy and coordination. Right now, coordination costs are incredibly high. A huge number of employees in any large organization are essentially moving information around. When AI lowers those costs, teams can specialize more, organizations can become more modular, and the overhead of alignment shrinks.

The strategic question becomes: where are the new constraints in your industry, and can you build a position around them that others can't easily replicate? Choudary cautions against over-relying on external AI tools, since they can lead to lock-in, give away learning effects, and shift pricing power to your vendor. Owning the tools that give you an advantage matters. And if the constraints are shifting, the org chart should follow, not the other way around.

My Verdict: Brilliant, but Could Be Clearer

I want to be honest about one thing: the book's style can be tough to follow. Choudary jumps between ideas from shipping containers to Shein to Formula 1 pit stops and the connections, while often brilliant, aren't always easy to track on a first read. The depth of thinking is impressive, but the density means you'll probably need some time to digest. Some chapters could have benefited from tighter structure or clearer signposting.

That said, intellectual ambition more than compensates. Most AI books give you one idea stretched across 300 pages. Reshuffle gives you a full framework, coordination vs. automation, economic vs. contextual value, constraint-based rebundling and then applies it across work, organizations, and industries.

I still don't know how AI will change our job landscape, but I got a lot of inspiration and found a good framework to think through it in a more structured way.

AI Disclosure: The research, structuring and notetaking was done by myself, Claude wrote a draft that I improved iteratively. No links or sources were added by Claude. Image done by Gemini.