Sangeet Paul Choudary
Reshuffle is a book I had been looking forward to reading ever since I started seeing Sangeet Paul Choudary’s excerpts and related posts. And despite reading all those posts, the book had sufficient meat to make it an excellent read. One that offers great perspectives on dealing with the influx of AI in work – at an individual and business level.
I think I have a bias for the book because of a few reasons. One, it follows a clear systems thinking approach, and the logic and reasoning is robust. Two, it is a strategic and framework-driven exploration, which means it offers tools to think about this in one’s own contexts. Three, it’s not just theory. The book provides sufficient examples that can serve as starting points on how to think through this. That, as you’d notice is an elegant why-what-how approach.
I’ll attempt a broad summary without ‘spoiling’ your reading of the book. Admittedly, this is difficult, because this book probably has the highest density of dog-ears in a while! Reshuffle is about how generative AI and systems-level automation are not simply replacing tasks but remapping where value, power and expertise live inside organisations and markets/ecosystems.
“AI changes the structure of the system in which it is introduced.”
It moves tacit knowledge out of individual minds/ ‘tribal knowledge’ into composable/ almost autodidactic (after initial onboarding and tutoring) tools and platforms, which forces businesses and professions to be reorganised from first principles rather than merely sped up through automation and efficiency. An excellent framework here is how AI achieves coordination without consensus – the architecture of coordination: representation, decision, execution, composition, governance (superb example of how the British controlled the provinces through this coordination).
The overall narrative is like a consultant making a layered case study (which I think his profession is). The first section – “AI & the system” – is about reframing the mental models we use to understand AI. AI isn’t simply substituting/automating tasks but re-modularising knowledge. Its superpower is coordination across fragmented systems involving actors with divergent incentives.
This section explains how the ‘knowledge economy’ is built on a stack of activities (data, expertise, judgment, coordination) rather than simple task work, and shows how AI is modularising what used to be tacit, breaking up old unit-models of expertise, and thereby changing the anatomy of work and value creation.
When the word processor arrived, ‘typists didn’t disappear, but the job of the typist did’. In essence, value is shifting from ownership of assets/knowledge to orchestration of flows and coordination, and thus restructuring power. Great examples here of Amazon (warehouse), legal firms, Shein, Walmart, through a task – organisation system – competitive ecosystems framework that showed the redistribution of value through (eco)system thinking.
The second section is “Work & Organizations”, and it focuses on the strategic implications on individual roles, and the workflows and business architecture of organisations. For individuals, the future is less about racing to acquire skills – because an entire system is changing , and more about rebundling different modules and contexts to create leverage, value and relevance of a different kind (sommeliers).
Reshuffle makes a good point that though human-generated outputs might have higher intrinsic value (subjective, rooted in meaning), its economic value (what gets traded) and contextual value (value within the system of work) might be less because markets might reward ‘sufficiency at scale’. Taxi drivers (knowledge based on experience) becoming commodity in the era of GPS are an example.
So the idea that Reshuffle posits is to find new constraints. These could be based on scarcity (radio operators in ships vs modern communication infrastructure) , risk (anaesthesiologist making decisions based on patient response and taking on the risk) or coordination (movie producer).
‘Don’t follow the skill, follow the constraint’.
This goes for organisations too, with the additional layers of how roles and teams need to be redesigned, and how governance must evolve. A good example is Ramp using AI to convert customer support team conversations into podcasts that people across departments could hear and thus reduced the coordination tax (meetings, reports) that it pays. Another example is the work Palantir’s Foundry does. Whether individuals or organisations, it is about new business economics when AI can embed expertise into software. Systems over tools.
Reshuffle’s final section is “Competitive Advantage”, and it focuses on ecosystems and economies. The first two chapters are spent differentiating the scope and future of tools vs solutions. I understood/thought it was a little B2B digression, but they do have some excellent examples and frameworks. For instance, how TikTok used a constraint (lack of social graph) to create an advantage, inferring a behaviour graph from scratch. Now new creators didn’t need to create an audience (a barrier to them joining), a ‘good’ video would become viral irrespective.
Also, an interesting part on why the robotics industry hasn’t really scaled (access, usable, reliable framework). And a framing of differentiating Work/Result/Outcome as a a Service.
The next three chapters focus on how AI can be used by organisations to reorganise entire ecosystems, what this would mean for labour markets and society, and how winners are doing/will do differently. I found the idea of ‘control without consensus’ (dynamic linkages across components, humans and machines, illustrated with an example of Tractable vs CC in the insurance claims space) very intriguing and at a very broad level dovetailing into the blockchain philosophy. A good analogy I found here was ‘what Ford’s moving assembly line did for car production, AI powered systems are doing for knowledge workflows, coordinating activity around the user rather than forcing the user to coordinate activity’.
Overall Reshuffle is about how AI is reshaping systems, and how the restacking of knowledge work will redistribute power (winners vs. losers), alter career paths, require new skills (judgement, curiosity, integration) and create new structural bottlenecks (coordination, data access, platform control). In summation, value, power and work are being redesigned; and human skills need to shift toward orchestration and judgement.
The language is very accessible, and the key takeaways at the end of chapters is a nice touch. I had only a couple of thoughts on what could have been made better. One, an editor could have made it crisper (without repetitions) and provided a more elegant narrative flow (seamlessness and cohesion of concepts, as opposed to stacking what might seem like blog posts rather than chapters). And two, for a book like this, not having an index is criminal!
But those are minor things compared to the value of the book. Absolutely worth a read, and in my 2025 Bibliofiles
Notes & Quotes from Reshuffle
1. AI is better understood as a practical utility that integrates into workflows and changes how decisions are made. By ‘sensing’ the environment, creating a working model, reasoning and acting based on this, and learning using patterns. (e.g. Google Maps, ChatGPT)
2. The example of how containers changed shipping. Predictability and reliability through coordination. Airbnb being similar, through reviews and a reputation system. Trust is practically a ‘system’. Stripe, Tesla are other examples
3. Three factors in making algorithmic coordination important. Ability to collect and use data to manage economic activity (software sensors over physical presence). Growing adoption of smartphones. (production and consumption timescale difference between Nike and Shein), rise of cloud based services to access specialised operational capabilities instead of building it themselves (Uber).
4. AI works at three different levels – first order effects of automation, cascading effect of coordination, accelerated effects of learning, where AI shapes the system’s behaviour and this further trains the AI
5. Mr.Beast assembled a business from pre-made building blocks. He had an online audience and started a burger chain with no expertise using ‘blocks’ of getting orders (demand generation), cooking the food (Ops), and delivering it (logistics). The constraint because of which it failed – quality control and lack of someone who took ownership of the entire process.
6. Historically expertise and specialised knowledge were tightly bundled with human labour. Organisations paid a premium for this. Now AI had unbundled it. It becomes rentable, recombinable and scalable.
7. “Tools amplify performance but solutions absorb risk.” A good version of the latter has the ability to “deliver predictable performance in unpredictable environments.”
8. “Control is earned by solving coordination problems, not merely by owning interfaces.”
9. Dominance does not guarantee control, but dependence does.
10 AI eliminates the need for shared standards by enabling the interpretation of diverse inputs into a unified understanding.
Leave a Reply