Great article, as always! Couldn't agree more and, in my case, I had used the marketing technology and the Brazilian wetlands analogy so marketers can understand they will need to have adaptive systems as well...
Yes to complex adaptive systems! Have you read Kathleen Schaub's book Marketing in the Great, Big, Messy real world? Also about marketing in these complex systems.
The Babbage insight is the most important reframe in here. Smith gets quoted constantly. Babbage almost never does but he's the one who identified that the division of labor was fundamentally a pricing mechanism, not just a productivity one. When AI absorbs the cheap steps, the economic logic of the lane structure disappears. What we're left with is the habit of lanes, not the reason for them.
"Redrawing the grid is still grid thinking" That's the line that should end most org redesign discussions. A new chart with the same underlying logic is just tomorrow's legacy system, and you've burned the transition cost to get there.
I write about production AI systems and distributed backends. Worth a subscribe here too.
As cited in you article, Adam Smith introduced the "Division of Labor" in his 1776 book "The Wealth of Nations" as workers in distinct, repetitive tasks within a production process. "Concrete" examples of functional specialization in a modern sense preceded Smith by thousands of years in large construction projects.
Since the metal ages, smelters, miners, metal smiths all had specific jobs. Specialization overlap occurred when talent and product economics could not support overheads of (hyper-)specialization. We find famous, practical successes from the middle ages, such as the Venetian Armory, which Smith would have known about.
Within specialized tasks, often we find workers using common, general purpose tools, such as hammers. Just because a hammer can be applied to multiple specializations, does not imply better products or specialization overlap. Rarely do specialized craftsman use one tool. Commonly used general purpose tools still require craftsmanship to actuate tools and produce products.
As tasks and products become increasingly specialize, tools also specialize to improve production efficiency.
Despite management beliefs, not everything is a same sized a nail. Selecting the wrong tool, such as one that's too general purpose for labor's skill set or poorly aligned to raw materials, creates a sub-optimal production system which either decreases product quality or increases production times and expense.
General purpose AI such as LLMs are the new hammer of the business environment. History shows us companies require craftsmanship with a portfolio of general purpose tools, specialized tools and tool skill sets to produce high quality products. Today's AI craftsman are 'the' integration between tool selections, workpieces and products. Craftmanship (skills) control product quality.
Today's AI are not and cannot replace craftsmanship. Technical (competences and integrations) and procedural barriers, such as quality processes prevent AI from replacing craftsman. Analysis of multiple AI project failures clearly expose the short comings of using a general purpose hammer to thread a needle.
Tools, raw materials, and craftmanship build product competency to provide valued solutions. Price, product quality, relationships and dynamics (tempos) determine customer segment. Respond too quickly, quality drops below customer expectation causing reputational damage and non-monetizable expenses. Respond too slowly, an run the risk of losing customers to competitors. Companies rate of adaptation is driven by multiple factors. Tools are selected based on rates of change required to support markets and profitability.
A living, prosperous company knows its goldie locks environment. It understands optimal rates to pivot and grow. Tools facilitate change. Tools do not execute commerce or business relationships
Love the observation that functional specialization long predates Smith, thanks for adding in that layer. And I'm with you that craftsmanship is the integration. The tool doesn't replace the person who knows which one the job needs and how to wield it. AI amplifies that person.
Where I'd add a little nuance is on who gets to be that person. General-purpose tools also lower the barrier to entry. A YouTube how-to now lets someone fix their own dishwasher, and AI lets people step onto the lower rungs of a trade they couldn't have entered before. If you'd been banging on things with a stone, a hammer suddenly lets you do more, and different, work. It widens the base of the pyramid rather than erasing the master craftsman at the top.
And these tools tend to create whole new trades rather than just filling old ones. Electricity gave us electricians, and then electrical engineers, roles that never existed to be replaced. I suspect AI does the same.
The Venetian Armory is an often overlooked example due to lost records from Napoleon completely destroying the place to prevent other nations from replacing damaged, captured and sunken military ships.
There are multiple stakeholder view points and perspectives to analyze discuss impacts of change. From the the workforce perspective, tools skills can be considered a type labor product diversification. Granted, it is a very Functionalist (neutral) view of workforce, but allows comparison to and possibly leverage well known other success strategies for their scales.
Access to low cost educational sources such as youtube definitely lowers the barrier to skill development (knowledge + experience = fungible skill). Youtube and other non-interactive knowledge tools are still using the "lecture format" motif that been applied for thousands of years. Lecture formats provide access to knowledge, but fails to provide the "experience" elements (understanding and tool manipulation), critical components of marketable skills. Lecture was a necessary, interactive format when reference books were expensive and scarce. Instructors read aloud to student cohorts who took notes, copying the information they heard. Students also had the opportunity to ask questions to fill gaps in understanding, a feature missing from non-interactive formats. Motivated workforces will invest in experiential development, moving themselves up the pyramid, as long as they can incur the costs eg. not having a dishwasher until parts show up and correct parts had been ordered.
We can forgo the impacts of mutual information ...
Knowledge based technologies are not new. The earliest foundations were lookup tables for mathematical operations including multiplication, division squares, cubes, and reciprocals @2000BC. Logarithms, invented in the 1600s. The application of lookup tables improved accuracy, reduced computation times, and provided a means to quickly audit results for correctness. Lookup tables are common, domain databases. Skills are required to apply the data they provide.
Modern knowledge based tools increased the complexity of lookup table with the introduction of "comparison mechanisms" and "good enough" statistical results. Despite the complexity, the lookup table is a prominent, required element of knowledge tools. Meaning, skills are still required to locate and apply the information.
Jumping a head to LLM technologies, encoding are no more than databases. Combined with the degree of complexity attempting to be all things to all people, and the "good enough" statistical results, skills are required to both extract information and audit results to detect and correct errors caused by the statistical nature of the calculations.
IMHO, the nature and structure of knowledge tools have not changed much in the last 4500 years. Complexity and technologies have changed, the underlying premise remains the same. The same gaps and skills are require to ensure accuracy and correctness whether using mathematical lookup tables to predicted seasons, creating story arcs, or analyzing customer sentiments.
Great article, as always! Couldn't agree more and, in my case, I had used the marketing technology and the Brazilian wetlands analogy so marketers can understand they will need to have adaptive systems as well...
Yes to complex adaptive systems! Have you read Kathleen Schaub's book Marketing in the Great, Big, Messy real world? Also about marketing in these complex systems.
Not yet! Thank you for the recommendation, I am going to get it now!
The Babbage insight is the most important reframe in here. Smith gets quoted constantly. Babbage almost never does but he's the one who identified that the division of labor was fundamentally a pricing mechanism, not just a productivity one. When AI absorbs the cheap steps, the economic logic of the lane structure disappears. What we're left with is the habit of lanes, not the reason for them.
"Redrawing the grid is still grid thinking" That's the line that should end most org redesign discussions. A new chart with the same underlying logic is just tomorrow's legacy system, and you've burned the transition cost to get there.
I write about production AI systems and distributed backends. Worth a subscribe here too.
Sorry for the length and rambling comment...
As cited in you article, Adam Smith introduced the "Division of Labor" in his 1776 book "The Wealth of Nations" as workers in distinct, repetitive tasks within a production process. "Concrete" examples of functional specialization in a modern sense preceded Smith by thousands of years in large construction projects.
Since the metal ages, smelters, miners, metal smiths all had specific jobs. Specialization overlap occurred when talent and product economics could not support overheads of (hyper-)specialization. We find famous, practical successes from the middle ages, such as the Venetian Armory, which Smith would have known about.
Within specialized tasks, often we find workers using common, general purpose tools, such as hammers. Just because a hammer can be applied to multiple specializations, does not imply better products or specialization overlap. Rarely do specialized craftsman use one tool. Commonly used general purpose tools still require craftsmanship to actuate tools and produce products.
As tasks and products become increasingly specialize, tools also specialize to improve production efficiency.
Despite management beliefs, not everything is a same sized a nail. Selecting the wrong tool, such as one that's too general purpose for labor's skill set or poorly aligned to raw materials, creates a sub-optimal production system which either decreases product quality or increases production times and expense.
General purpose AI such as LLMs are the new hammer of the business environment. History shows us companies require craftsmanship with a portfolio of general purpose tools, specialized tools and tool skill sets to produce high quality products. Today's AI craftsman are 'the' integration between tool selections, workpieces and products. Craftmanship (skills) control product quality.
Today's AI are not and cannot replace craftsmanship. Technical (competences and integrations) and procedural barriers, such as quality processes prevent AI from replacing craftsman. Analysis of multiple AI project failures clearly expose the short comings of using a general purpose hammer to thread a needle.
Tools, raw materials, and craftmanship build product competency to provide valued solutions. Price, product quality, relationships and dynamics (tempos) determine customer segment. Respond too quickly, quality drops below customer expectation causing reputational damage and non-monetizable expenses. Respond too slowly, an run the risk of losing customers to competitors. Companies rate of adaptation is driven by multiple factors. Tools are selected based on rates of change required to support markets and profitability.
A living, prosperous company knows its goldie locks environment. It understands optimal rates to pivot and grow. Tools facilitate change. Tools do not execute commerce or business relationships
Cool, comment. Thanks for weighing in.
Love the observation that functional specialization long predates Smith, thanks for adding in that layer. And I'm with you that craftsmanship is the integration. The tool doesn't replace the person who knows which one the job needs and how to wield it. AI amplifies that person.
Where I'd add a little nuance is on who gets to be that person. General-purpose tools also lower the barrier to entry. A YouTube how-to now lets someone fix their own dishwasher, and AI lets people step onto the lower rungs of a trade they couldn't have entered before. If you'd been banging on things with a stone, a hammer suddenly lets you do more, and different, work. It widens the base of the pyramid rather than erasing the master craftsman at the top.
And these tools tend to create whole new trades rather than just filling old ones. Electricity gave us electricians, and then electrical engineers, roles that never existed to be replaced. I suspect AI does the same.
Appreciate the discussion.
The Venetian Armory is an often overlooked example due to lost records from Napoleon completely destroying the place to prevent other nations from replacing damaged, captured and sunken military ships.
There are multiple stakeholder view points and perspectives to analyze discuss impacts of change. From the the workforce perspective, tools skills can be considered a type labor product diversification. Granted, it is a very Functionalist (neutral) view of workforce, but allows comparison to and possibly leverage well known other success strategies for their scales.
Access to low cost educational sources such as youtube definitely lowers the barrier to skill development (knowledge + experience = fungible skill). Youtube and other non-interactive knowledge tools are still using the "lecture format" motif that been applied for thousands of years. Lecture formats provide access to knowledge, but fails to provide the "experience" elements (understanding and tool manipulation), critical components of marketable skills. Lecture was a necessary, interactive format when reference books were expensive and scarce. Instructors read aloud to student cohorts who took notes, copying the information they heard. Students also had the opportunity to ask questions to fill gaps in understanding, a feature missing from non-interactive formats. Motivated workforces will invest in experiential development, moving themselves up the pyramid, as long as they can incur the costs eg. not having a dishwasher until parts show up and correct parts had been ordered.
We can forgo the impacts of mutual information ...
Knowledge based technologies are not new. The earliest foundations were lookup tables for mathematical operations including multiplication, division squares, cubes, and reciprocals @2000BC. Logarithms, invented in the 1600s. The application of lookup tables improved accuracy, reduced computation times, and provided a means to quickly audit results for correctness. Lookup tables are common, domain databases. Skills are required to apply the data they provide.
Modern knowledge based tools increased the complexity of lookup table with the introduction of "comparison mechanisms" and "good enough" statistical results. Despite the complexity, the lookup table is a prominent, required element of knowledge tools. Meaning, skills are still required to locate and apply the information.
Jumping a head to LLM technologies, encoding are no more than databases. Combined with the degree of complexity attempting to be all things to all people, and the "good enough" statistical results, skills are required to both extract information and audit results to detect and correct errors caused by the statistical nature of the calculations.
IMHO, the nature and structure of knowledge tools have not changed much in the last 4500 years. Complexity and technologies have changed, the underlying premise remains the same. The same gaps and skills are require to ensure accuracy and correctness whether using mathematical lookup tables to predicted seasons, creating story arcs, or analyzing customer sentiments.