1
The Company Is the Atomic Unit of Capitalism, and It Was Not Designed for Compute
The company is where intelligence meets reality. It is the structure through which ideas become products and where the economy generates most of its value. Whatever you believe about the trajectory of AI, the company, as opposed to the research lab or the government, is the place where that future will either be built or squandered.
Adam Smith's most famous insight was the power of division of labor. Fragment work into narrow, repeatable tasks, he said, and you’ll multiply output. For two and a half centuries, every major organizational structure has followed from that premise, and underneath it sits a single assumption: that human capacity is the binding constraint on what an organization can do.
These structures made sense when coordination was expensive and cognition was scarce. AI collapsed both conditions at once.
Right now, most organizations punish that collapse. If you double your output and the company responds by cutting your team in half, you will never touch AI again. As such, enterprise AI spending produces failure rates that would be treated as scandalous in any other category of investment. For example, a RAND study noted that 80% of enterprise AI initiatives fail against their stated goals, twice the rate of ordinary IT projects. Likewise, Gallup's February 2026 survey of 23,717 U.S. employees found that 65% of workers in AI-adopting organizations say AI has improved their productivity, but only about 10% strongly agree that AI has changed how work gets done across the organization.
The gains are real at the level of the task and absent at the level of the organization. Some workers have found leverage. The company has not yet redesigned itself around that leverage, because the organization was never designed for compute.
1
The Company Is the Atomic Unit of Capitalism, and It Was Not Designed for Compute
The company is where intelligence meets reality. It is the structure through which ideas become products and where the economy generates most of its value. Whatever you believe about the trajectory of AI, the company, as opposed to the research lab or the government, is the place where that future will either be built or squandered.
Adam Smith's most famous insight was the power of division of labor. Fragment work into narrow, repeatable tasks, he said, and you’ll multiply output. For two and a half centuries, every major organizational structure has followed from that premise, and underneath it sits a single assumption: that human capacity is the binding constraint on what an organization can do.
These structures made sense when coordination was expensive and cognition was scarce. AI collapsed both conditions at once.
Right now, most organizations punish that collapse. If you double your output and the company responds by cutting your team in half, you will never touch AI again. As such, enterprise AI spending produces failure rates that would be treated as scandalous in any other category of investment. For example, a RAND study noted that 80% of enterprise AI initiatives fail against their stated goals, twice the rate of ordinary IT projects. Likewise, Gallup's February 2026 survey of 23,717 U.S. employees found that 65% of workers in AI-adopting organizations say AI has improved their productivity, but only about 10% strongly agree that AI has changed how work gets done across the organization.
The gains are real at the level of the task and absent at the level of the organization. Some workers have found leverage. The company has not yet redesigned itself around that leverage, because the organization was never designed for compute.
1
The Company Is the Atomic Unit of Capitalism, and It Was Not Designed for Compute
The company is where intelligence meets reality. It is the structure through which ideas become products and where the economy generates most of its value. Whatever you believe about the trajectory of AI, the company, as opposed to the research lab or the government, is the place where that future will either be built or squandered.
Adam Smith's most famous insight was the power of division of labor. Fragment work into narrow, repeatable tasks, he said, and you’ll multiply output. For two and a half centuries, every major organizational structure has followed from that premise, and underneath it sits a single assumption: that human capacity is the binding constraint on what an organization can do.
These structures made sense when coordination was expensive and cognition was scarce. AI collapsed both conditions at once.
Right now, most organizations punish that collapse. If you double your output and the company responds by cutting your team in half, you will never touch AI again. As such, enterprise AI spending produces failure rates that would be treated as scandalous in any other category of investment. For example, a RAND study noted that 80% of enterprise AI initiatives fail against their stated goals, twice the rate of ordinary IT projects. Likewise, Gallup's February 2026 survey of 23,717 U.S. employees found that 65% of workers in AI-adopting organizations say AI has improved their productivity, but only about 10% strongly agree that AI has changed how work gets done across the organization.
The gains are real at the level of the task and absent at the level of the organization. Some workers have found leverage. The company has not yet redesigned itself around that leverage, because the organization was never designed for compute.
2
Compression Is a Means, Not the Goal
Compression is the reduction of time and effort that routine work demands, achieved through AI augmentation. Automation implies replacement. Compression implies concentration. When you compress a file, the information does not disappear; it occupies less space. When you compress human work, the routine recedes and what remains is the judgment, the relationships, and the creative decisions that the routine was consuming time and attention away from.
Compression is where our work begins. Left alone, it produces a smaller version of the same job, which is why so many AI deployments stall after the first efficiency gain. But the point of compression is to clear room for everything that has to happen next. What workflows will get redesigned? What operating models must change? The person doing the work is now freed up to do more human work, the work that matters more. Compression that opens into redesign is how an organization reinvents itself.
Someone who figures out how to do their work in half the time should be handed harder problems, not a layoff notice.
Consider a recruiter who can now manage 600 hires instead of 300. The structure that fragmented her work gets rebuilt around evaluating candidates and building relationships with hiring managers. AI handles the screening, the scheduling, and the data entry, and she keeps the parts of the job that require her to be a person. Her output doubles, and the work itself gets more interesting.
Or consider a nurse freed from end-of-shift charting spends those hours at the bedside instead of in front of a screen. Or a building superintendent whose tacit knowledge of a property gets captured in a system can finally take a week off without the building falling apart, because the institution no longer depends on what lives only in his head. Each person becomes more valuable the moment AI removes the low-value work that was crowding out their judgment and care.
Of course, maybe this work that we’re compressing wasn’t perceived as “low value” to begin with. Rapid advancements in LLMs over the last few years have forced us to reconsider what work humans should do in an age of abundant, raw intelligence.
Alex Imas, Director of AGI Economics at Google DeepMind, has argued that as AI compresses routine production, spending migrates toward goods and services where human involvement is itself part of the value. Drawing on a 2021 Econometrica paper by Comin, Lashkari, and Mestieri, Imas notes that income effects, not price effects, account for more than 75% of historical patterns of sectoral reallocation. This means that when people get richer, they do not simply buy more of the same things at lower prices. A larger share of their spending moves toward goods where the identity of the producer matters: education, healthcare, hospitality, craftsmanship, anything where you are buying the story and the humanity behind the product and not only its function.
AI pushes the economy toward a post-commodity shape. A growing share of expenditure flows to goods and services whose value is inseparable from the human who provided them.
2
Compression Is a Means, Not the Goal
Compression is the reduction of time and effort that routine work demands, achieved through AI augmentation. Automation implies replacement. Compression implies concentration. When you compress a file, the information does not disappear; it occupies less space. When you compress human work, the routine recedes and what remains is the judgment, the relationships, and the creative decisions that the routine was consuming time and attention away from.
Compression is where our work begins. Left alone, it produces a smaller version of the same job, which is why so many AI deployments stall after the first efficiency gain. But the point of compression is to clear room for everything that has to happen next. What workflows will get redesigned? What operating models must change? The person doing the work is now freed up to do more human work, the work that matters more. Compression that opens into redesign is how an organization reinvents itself.
Someone who figures out how to do their work in half the time should be handed harder problems, not a layoff notice.
Consider a recruiter who can now manage 600 hires instead of 300. The structure that fragmented her work gets rebuilt around evaluating candidates and building relationships with hiring managers. AI handles the screening, the scheduling, and the data entry, and she keeps the parts of the job that require her to be a person. Her output doubles, and the work itself gets more interesting.
Or consider a nurse freed from end-of-shift charting spends those hours at the bedside instead of in front of a screen. Or a building superintendent whose tacit knowledge of a property gets captured in a system can finally take a week off without the building falling apart, because the institution no longer depends on what lives only in his head. Each person becomes more valuable the moment AI removes the low-value work that was crowding out their judgment and care.
Of course, maybe this work that we’re compressing wasn’t perceived as “low value” to begin with. Rapid advancements in LLMs over the last few years have forced us to reconsider what work humans should do in an age of abundant, raw intelligence.
Alex Imas, Director of AGI Economics at Google DeepMind, has argued that as AI compresses routine production, spending migrates toward goods and services where human involvement is itself part of the value. Drawing on a 2021 Econometrica paper by Comin, Lashkari, and Mestieri, Imas notes that income effects, not price effects, account for more than 75% of historical patterns of sectoral reallocation. This means that when people get richer, they do not simply buy more of the same things at lower prices. A larger share of their spending moves toward goods where the identity of the producer matters: education, healthcare, hospitality, craftsmanship, anything where you are buying the story and the humanity behind the product and not only its function.
AI pushes the economy toward a post-commodity shape. A growing share of expenditure flows to goods and services whose value is inseparable from the human who provided them.
2
Compression Is a Means, Not the Goal
Compression is the reduction of time and effort that routine work demands, achieved through AI augmentation. Automation implies replacement. Compression implies concentration. When you compress a file, the information does not disappear; it occupies less space. When you compress human work, the routine recedes and what remains is the judgment, the relationships, and the creative decisions that the routine was consuming time and attention away from.
Compression is where our work begins. Left alone, it produces a smaller version of the same job, which is why so many AI deployments stall after the first efficiency gain. But the point of compression is to clear room for everything that has to happen next. What workflows will get redesigned? What operating models must change? The person doing the work is now freed up to do more human work, the work that matters more. Compression that opens into redesign is how an organization reinvents itself.
Someone who figures out how to do their work in half the time should be handed harder problems, not a layoff notice.
Consider a recruiter who can now manage 600 hires instead of 300. The structure that fragmented her work gets rebuilt around evaluating candidates and building relationships with hiring managers. AI handles the screening, the scheduling, and the data entry, and she keeps the parts of the job that require her to be a person. Her output doubles, and the work itself gets more interesting.
Or consider a nurse freed from end-of-shift charting spends those hours at the bedside instead of in front of a screen. Or a building superintendent whose tacit knowledge of a property gets captured in a system can finally take a week off without the building falling apart, because the institution no longer depends on what lives only in his head. Each person becomes more valuable the moment AI removes the low-value work that was crowding out their judgment and care.
Of course, maybe this work that we’re compressing wasn’t perceived as “low value” to begin with. Rapid advancements in LLMs over the last few years have forced us to reconsider what work humans should do in an age of abundant, raw intelligence.
Alex Imas, Director of AGI Economics at Google DeepMind, has argued that as AI compresses routine production, spending migrates toward goods and services where human involvement is itself part of the value. Drawing on a 2021 Econometrica paper by Comin, Lashkari, and Mestieri, Imas notes that income effects, not price effects, account for more than 75% of historical patterns of sectoral reallocation. This means that when people get richer, they do not simply buy more of the same things at lower prices. A larger share of their spending moves toward goods where the identity of the producer matters: education, healthcare, hospitality, craftsmanship, anything where you are buying the story and the humanity behind the product and not only its function.
AI pushes the economy toward a post-commodity shape. A growing share of expenditure flows to goods and services whose value is inseparable from the human who provided them.
2
Compression Is a Means, Not the Goal
Compression is the reduction of time and effort that routine work demands, achieved through AI augmentation. Automation implies replacement. Compression implies concentration. When you compress a file, the information does not disappear; it occupies less space. When you compress human work, the routine recedes and what remains is the judgment, the relationships, and the creative decisions that the routine was consuming time and attention away from.
Compression is where our work begins. Left alone, it produces a smaller version of the same job, which is why so many AI deployments stall after the first efficiency gain. But the point of compression is to clear room for everything that has to happen next. What workflows will get redesigned? What operating models must change? The person doing the work is now freed up to do more human work, the work that matters more. Compression that opens into redesign is how an organization reinvents itself.
Someone who figures out how to do their work in half the time should be handed harder problems, not a layoff notice.
Consider a recruiter who can now manage 600 hires instead of 300. The structure that fragmented her work gets rebuilt around evaluating candidates and building relationships with hiring managers. AI handles the screening, the scheduling, and the data entry, and she keeps the parts of the job that require her to be a person. Her output doubles, and the work itself gets more interesting.
Or consider a nurse freed from end-of-shift charting spends those hours at the bedside instead of in front of a screen. Or a building superintendent whose tacit knowledge of a property gets captured in a system can finally take a week off without the building falling apart, because the institution no longer depends on what lives only in his head. Each person becomes more valuable the moment AI removes the low-value work that was crowding out their judgment and care.
Of course, maybe this work that we’re compressing wasn’t perceived as “low value” to begin with. Rapid advancements in LLMs over the last few years have forced us to reconsider what work humans should do in an age of abundant, raw intelligence.
Alex Imas, Director of AGI Economics at Google DeepMind, has argued that as AI compresses routine production, spending migrates toward goods and services where human involvement is itself part of the value. Drawing on a 2021 Econometrica paper by Comin, Lashkari, and Mestieri, Imas notes that income effects, not price effects, account for more than 75% of historical patterns of sectoral reallocation. This means that when people get richer, they do not simply buy more of the same things at lower prices. A larger share of their spending moves toward goods where the identity of the producer matters: education, healthcare, hospitality, craftsmanship, anything where you are buying the story and the humanity behind the product and not only its function.
AI pushes the economy toward a post-commodity shape. A growing share of expenditure flows to goods and services whose value is inseparable from the human who provided them.
3
The Silicon Valley Story Is the Wrong Story
This is why the “one-person billion-dollar company” that Silicon Valley has fascinated itself with is the wrong ambition. A single person has a hard ceiling on judgment and context, and a truly AI-native organization would need more humans in order to grow and succeed.
What happens when every person in an organization operates at a hundred times their current output? The organization becomes more powerful with each person you add. Each marginal member of the team brings judgment and relationships that AI cannot replicate, and compression makes all of it more valuable.
Another dominant account of AI and work runs in a straight line: AI produces productivity, productivity produces layoffs, and the leverage concentrates in the hands of a small elite who command the machines, forming a “permanent underclass.” Some actors have promised abundance so large that the transition costs stop mattering. Others warn of mass displacement to be managed through taxation and redistribution. In both of these stories, AI happens to people. The economy is a system to be steered from the outside, in a very Seeing Like a State-esque fashion.
AIVC tells a different story. AI compresses the routine, the routine compression opens into organizational redesign, the redesign produces expansion, and the expansion creates new work, new companies, and new opportunity. The first story ends with fewer people doing the same work for less. Our belief is that the story continues with more people doing better work than they could before. This is only possible if someone redesigns the organization so that the people inside it receive the upside of the compression.
Our core belief is that the view from inside the company is the most important.
The AI industry recognized this gap and decided to enter as a vendor. So far, the assumption has held that building the model was the hard part and everything downstream is distribution. Of course, the labs then ran into a deployment bottleneck, so in May 2026, the leading labs launched deployment companies and structured around forward-deployed engineers who embed inside portfolio companies.
The incentives of this approach are all wrong. These ventures are built so that the real revenue engine is the explosion of API calls and inference that follows each deployment. The consulting is the customer acquisition cost and the tokens are the margin.
Nobody in that chain is asking the question that matters: what should this organization become, and what do the people inside it need to align around compression?
3
The Silicon Valley Story Is the Wrong Story
This is why the “one-person billion-dollar company” that Silicon Valley has fascinated itself with is the wrong ambition. A single person has a hard ceiling on judgment and context, and a truly AI-native organization would need more humans in order to grow and succeed.
What happens when every person in an organization operates at a hundred times their current output? The organization becomes more powerful with each person you add. Each marginal member of the team brings judgment and relationships that AI cannot replicate, and compression makes all of it more valuable.
Another dominant account of AI and work runs in a straight line: AI produces productivity, productivity produces layoffs, and the leverage concentrates in the hands of a small elite who command the machines, forming a “permanent underclass.” Some actors have promised abundance so large that the transition costs stop mattering. Others warn of mass displacement to be managed through taxation and redistribution. In both of these stories, AI happens to people. The economy is a system to be steered from the outside, in a very Seeing Like a State-esque fashion.
AIVC tells a different story. AI compresses the routine, the routine compression opens into organizational redesign, the redesign produces expansion, and the expansion creates new work, new companies, and new opportunity. The first story ends with fewer people doing the same work for less. Our belief is that the story continues with more people doing better work than they could before. This is only possible if someone redesigns the organization so that the people inside it receive the upside of the compression.
Our core belief is that the view from inside the company is the most important.
The AI industry recognized this gap and decided to enter as a vendor. So far, the assumption has held that building the model was the hard part and everything downstream is distribution. Of course, the labs then ran into a deployment bottleneck, so in May 2026, the leading labs launched deployment companies and structured around forward-deployed engineers who embed inside portfolio companies.
The incentives of this approach are all wrong. These ventures are built so that the real revenue engine is the explosion of API calls and inference that follows each deployment. The consulting is the customer acquisition cost and the tokens are the margin.
Nobody in that chain is asking the question that matters: what should this organization become, and what do the people inside it need to align around compression?
3
The Silicon Valley Story Is the Wrong Story
This is why the “one-person billion-dollar company” that Silicon Valley has fascinated itself with is the wrong ambition. A single person has a hard ceiling on judgment and context, and a truly AI-native organization would need more humans in order to grow and succeed.
What happens when every person in an organization operates at a hundred times their current output? The organization becomes more powerful with each person you add. Each marginal member of the team brings judgment and relationships that AI cannot replicate, and compression makes all of it more valuable.
Another dominant account of AI and work runs in a straight line: AI produces productivity, productivity produces layoffs, and the leverage concentrates in the hands of a small elite who command the machines, forming a “permanent underclass.” Some actors have promised abundance so large that the transition costs stop mattering. Others warn of mass displacement to be managed through taxation and redistribution. In both of these stories, AI happens to people. The economy is a system to be steered from the outside, in a very Seeing Like a State-esque fashion.
AIVC tells a different story. AI compresses the routine, the routine compression opens into organizational redesign, the redesign produces expansion, and the expansion creates new work, new companies, and new opportunity. The first story ends with fewer people doing the same work for less. Our belief is that the story continues with more people doing better work than they could before. This is only possible if someone redesigns the organization so that the people inside it receive the upside of the compression.
Our core belief is that the view from inside the company is the most important.
The AI industry recognized this gap and decided to enter as a vendor. So far, the assumption has held that building the model was the hard part and everything downstream is distribution. Of course, the labs then ran into a deployment bottleneck, so in May 2026, the leading labs launched deployment companies and structured around forward-deployed engineers who embed inside portfolio companies.
The incentives of this approach are all wrong. These ventures are built so that the real revenue engine is the explosion of API calls and inference that follows each deployment. The consulting is the customer acquisition cost and the tokens are the margin.
Nobody in that chain is asking the question that matters: what should this organization become, and what do the people inside it need to align around compression?
3
The Silicon Valley Story Is the Wrong Story
This is why the “one-person billion-dollar company” that Silicon Valley has fascinated itself with is the wrong ambition. A single person has a hard ceiling on judgment and context, and a truly AI-native organization would need more humans in order to grow and succeed.
What happens when every person in an organization operates at a hundred times their current output? The organization becomes more powerful with each person you add. Each marginal member of the team brings judgment and relationships that AI cannot replicate, and compression makes all of it more valuable.
Another dominant account of AI and work runs in a straight line: AI produces productivity, productivity produces layoffs, and the leverage concentrates in the hands of a small elite who command the machines, forming a “permanent underclass.” Some actors have promised abundance so large that the transition costs stop mattering. Others warn of mass displacement to be managed through taxation and redistribution. In both of these stories, AI happens to people. The economy is a system to be steered from the outside, in a very Seeing Like a State-esque fashion.
AIVC tells a different story. AI compresses the routine, the routine compression opens into organizational redesign, the redesign produces expansion, and the expansion creates new work, new companies, and new opportunity. The first story ends with fewer people doing the same work for less. Our belief is that the story continues with more people doing better work than they could before. This is only possible if someone redesigns the organization so that the people inside it receive the upside of the compression.
Our core belief is that the view from inside the company is the most important.
The AI industry recognized this gap and decided to enter as a vendor. So far, the assumption has held that building the model was the hard part and everything downstream is distribution. Of course, the labs then ran into a deployment bottleneck, so in May 2026, the leading labs launched deployment companies and structured around forward-deployed engineers who embed inside portfolio companies.
The incentives of this approach are all wrong. These ventures are built so that the real revenue engine is the explosion of API calls and inference that follows each deployment. The consulting is the customer acquisition cost and the tokens are the margin.
Nobody in that chain is asking the question that matters: what should this organization become, and what do the people inside it need to align around compression?
4
Compression Is Already Working
Compression-driven flourishing sounds great in theory, but is it actually happening in the real world?
Healthcare seems to be the best example so far. A JAMA study of ambient AI scribes across five academic medical centers found that they cut documentation time by 16 minutes per clinician per day and were associated with 0.49 more visits per week. Mass General Brigham saw burnout prevalence fall by 21.2 percent after 84 days. Obviously, efficiency is the headline here. But the deeper story is that clinical attention moves back toward the patient and the practice can spend recovered capacity on follow-up and services it previously could not justify staffing. The human becomes more available for real care.
The same is true beyond medicine. In a field experiment across 5,172 support agents, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that a generative AI assistant raised issues resolved per hour by about 15%. The best customer service is the right routing of human attention, and compression lets the function stop spending people on repeated friction and reserve them for the moments where a customer needs a person.
These examples are early, and they do not prove that AI automatically produces better companies. The good outcome appears only where the routine layer is compressed and the organization knows what to do with the capacity that comes back. AI can make human work more valuable, but only if the company is redesigned so the human gets the upside of the compression.
4
Compression Is Already Working
Compression-driven flourishing sounds great in theory, but is it actually happening in the real world?
Healthcare seems to be the best example so far. A JAMA study of ambient AI scribes across five academic medical centers found that they cut documentation time by 16 minutes per clinician per day and were associated with 0.49 more visits per week. Mass General Brigham saw burnout prevalence fall by 21.2 percent after 84 days. Obviously, efficiency is the headline here. But the deeper story is that clinical attention moves back toward the patient and the practice can spend recovered capacity on follow-up and services it previously could not justify staffing. The human becomes more available for real care.
The same is true beyond medicine. In a field experiment across 5,172 support agents, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that a generative AI assistant raised issues resolved per hour by about 15%. The best customer service is the right routing of human attention, and compression lets the function stop spending people on repeated friction and reserve them for the moments where a customer needs a person.
These examples are early, and they do not prove that AI automatically produces better companies. The good outcome appears only where the routine layer is compressed and the organization knows what to do with the capacity that comes back. AI can make human work more valuable, but only if the company is redesigned so the human gets the upside of the compression.
4
Compression Is Already Working
Compression-driven flourishing sounds great in theory, but is it actually happening in the real world?
Healthcare seems to be the best example so far. A JAMA study of ambient AI scribes across five academic medical centers found that they cut documentation time by 16 minutes per clinician per day and were associated with 0.49 more visits per week. Mass General Brigham saw burnout prevalence fall by 21.2 percent after 84 days. Obviously, efficiency is the headline here. But the deeper story is that clinical attention moves back toward the patient and the practice can spend recovered capacity on follow-up and services it previously could not justify staffing. The human becomes more available for real care.
The same is true beyond medicine. In a field experiment across 5,172 support agents, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that a generative AI assistant raised issues resolved per hour by about 15%. The best customer service is the right routing of human attention, and compression lets the function stop spending people on repeated friction and reserve them for the moments where a customer needs a person.
These examples are early, and they do not prove that AI automatically produces better companies. The good outcome appears only where the routine layer is compressed and the organization knows what to do with the capacity that comes back. AI can make human work more valuable, but only if the company is redesigned so the human gets the upside of the compression.
4
Compression Is Already Working
Compression-driven flourishing sounds great in theory, but is it actually happening in the real world?
Healthcare seems to be the best example so far. A JAMA study of ambient AI scribes across five academic medical centers found that they cut documentation time by 16 minutes per clinician per day and were associated with 0.49 more visits per week. Mass General Brigham saw burnout prevalence fall by 21.2 percent after 84 days. Obviously, efficiency is the headline here. But the deeper story is that clinical attention moves back toward the patient and the practice can spend recovered capacity on follow-up and services it previously could not justify staffing. The human becomes more available for real care.
The same is true beyond medicine. In a field experiment across 5,172 support agents, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that a generative AI assistant raised issues resolved per hour by about 15%. The best customer service is the right routing of human attention, and compression lets the function stop spending people on repeated friction and reserve them for the moments where a customer needs a person.
These examples are early, and they do not prove that AI automatically produces better companies. The good outcome appears only where the routine layer is compressed and the organization knows what to do with the capacity that comes back. AI can make human work more valuable, but only if the company is redesigned so the human gets the upside of the compression.
5
The Knowledge Inside Companies Cannot Be Accessed From Outside
We believe that we should be seeing many more examples like the above, though. The technology has not unlocked the abundance we were promised. This is because companies contain decades of accumulated context: tacit knowledge embedded in workflows and relationships and workarounds and institutional memory that is poorly documented.
The CFO's spreadsheet contradicts the ERP and everyone knows which one to trust. The scheduling coordinator memorizes every clinician's preference because the system never captured them. Three downstream reports depend on a field corrupted in a 2019 data migration, and one person knows why. The senior account manager has a relationship with the client's procurement lead that has saved three contracts over a decade.
No API call retrieves any of this. No model trained on public data contains it. The people who carry this knowledge are the documentation, and they will not hand it to a vendor whose incentive is to automate them out of a job.
Timothy Williamson, in Knowledge and Its Limits, argues that knowing is a fundamental state rather than a composite that can be assembled from more basic ingredients like belief plus justification. You are not always in a position to know whether you are in a state of knowing. A condition can fully obtain while remaining invisible to the person standing closest to it, and no outside observer can reliably recover what has not been made legible. The knowledge inside a company behaves this way. Intelligence is not the same thing as usable context. A model can reason over what it can see, but the facts that matter most inside a company stay invisible until trust and proximity and operational contact make them available. Context is a living account of how work actually gets done.
There is a reason the labs started with code: code has cheap, open verifiers in compilers and test suites, so you can tell whether the output is correct. Business operations have no public equivalent, and the verification environment is the company itself. Parachuting engineers in for ninety days, deploying a model, and leaving won’t work. An outside team optimizing for inference revenue on a short clock has no way to earn what the work requires.
All of these observations make the firm more important. If cognition gets cheaper, the scarce asset is trusted access to the context that makes an answer useful and the accountability to own what happens after the answer becomes systematic action. The company is the container for the tacit knowledge, institutional memory, liability, and human judgment that turn intelligence into value.
The knowledge that makes a company work lives in the people who do the work, and unlocking it requires earning their trust. That means being on their side. It means designing the compression so that the person who reveals how the work actually gets done is rewarded with a better role, not a termination meeting. The traditional consulting model cannot do this because the vendor's incentive is perpendicular to the employee's interest.
5
The Knowledge Inside Companies Cannot Be Accessed From Outside
We believe that we should be seeing many more examples like the above, though. The technology has not unlocked the abundance we were promised. This is because companies contain decades of accumulated context: tacit knowledge embedded in workflows and relationships and workarounds and institutional memory that is poorly documented.
The CFO's spreadsheet contradicts the ERP and everyone knows which one to trust. The scheduling coordinator memorizes every clinician's preference because the system never captured them. Three downstream reports depend on a field corrupted in a 2019 data migration, and one person knows why. The senior account manager has a relationship with the client's procurement lead that has saved three contracts over a decade.
No API call retrieves any of this. No model trained on public data contains it. The people who carry this knowledge are the documentation, and they will not hand it to a vendor whose incentive is to automate them out of a job.
Timothy Williamson, in Knowledge and Its Limits, argues that knowing is a fundamental state rather than a composite that can be assembled from more basic ingredients like belief plus justification. You are not always in a position to know whether you are in a state of knowing. A condition can fully obtain while remaining invisible to the person standing closest to it, and no outside observer can reliably recover what has not been made legible. The knowledge inside a company behaves this way. Intelligence is not the same thing as usable context. A model can reason over what it can see, but the facts that matter most inside a company stay invisible until trust and proximity and operational contact make them available. Context is a living account of how work actually gets done.
There is a reason the labs started with code: code has cheap, open verifiers in compilers and test suites, so you can tell whether the output is correct. Business operations have no public equivalent, and the verification environment is the company itself. Parachuting engineers in for ninety days, deploying a model, and leaving won’t work. An outside team optimizing for inference revenue on a short clock has no way to earn what the work requires.
All of these observations make the firm more important. If cognition gets cheaper, the scarce asset is trusted access to the context that makes an answer useful and the accountability to own what happens after the answer becomes systematic action. The company is the container for the tacit knowledge, institutional memory, liability, and human judgment that turn intelligence into value.
The knowledge that makes a company work lives in the people who do the work, and unlocking it requires earning their trust. That means being on their side. It means designing the compression so that the person who reveals how the work actually gets done is rewarded with a better role, not a termination meeting. The traditional consulting model cannot do this because the vendor's incentive is perpendicular to the employee's interest.
5
The Knowledge Inside Companies Cannot Be Accessed From Outside
We believe that we should be seeing many more examples like the above, though. The technology has not unlocked the abundance we were promised. This is because companies contain decades of accumulated context: tacit knowledge embedded in workflows and relationships and workarounds and institutional memory that is poorly documented.
The CFO's spreadsheet contradicts the ERP and everyone knows which one to trust. The scheduling coordinator memorizes every clinician's preference because the system never captured them. Three downstream reports depend on a field corrupted in a 2019 data migration, and one person knows why. The senior account manager has a relationship with the client's procurement lead that has saved three contracts over a decade.
No API call retrieves any of this. No model trained on public data contains it. The people who carry this knowledge are the documentation, and they will not hand it to a vendor whose incentive is to automate them out of a job.
Timothy Williamson, in Knowledge and Its Limits, argues that knowing is a fundamental state rather than a composite that can be assembled from more basic ingredients like belief plus justification. You are not always in a position to know whether you are in a state of knowing. A condition can fully obtain while remaining invisible to the person standing closest to it, and no outside observer can reliably recover what has not been made legible. The knowledge inside a company behaves this way. Intelligence is not the same thing as usable context. A model can reason over what it can see, but the facts that matter most inside a company stay invisible until trust and proximity and operational contact make them available. Context is a living account of how work actually gets done.
There is a reason the labs started with code: code has cheap, open verifiers in compilers and test suites, so you can tell whether the output is correct. Business operations have no public equivalent, and the verification environment is the company itself. Parachuting engineers in for ninety days, deploying a model, and leaving won’t work. An outside team optimizing for inference revenue on a short clock has no way to earn what the work requires.
All of these observations make the firm more important. If cognition gets cheaper, the scarce asset is trusted access to the context that makes an answer useful and the accountability to own what happens after the answer becomes systematic action. The company is the container for the tacit knowledge, institutional memory, liability, and human judgment that turn intelligence into value.
The knowledge that makes a company work lives in the people who do the work, and unlocking it requires earning their trust. That means being on their side. It means designing the compression so that the person who reveals how the work actually gets done is rewarded with a better role, not a termination meeting. The traditional consulting model cannot do this because the vendor's incentive is perpendicular to the employee's interest.
5
The Knowledge Inside Companies Cannot Be Accessed From Outside
We believe that we should be seeing many more examples like the above, though. The technology has not unlocked the abundance we were promised. This is because companies contain decades of accumulated context: tacit knowledge embedded in workflows and relationships and workarounds and institutional memory that is poorly documented.
The CFO's spreadsheet contradicts the ERP and everyone knows which one to trust. The scheduling coordinator memorizes every clinician's preference because the system never captured them. Three downstream reports depend on a field corrupted in a 2019 data migration, and one person knows why. The senior account manager has a relationship with the client's procurement lead that has saved three contracts over a decade.
No API call retrieves any of this. No model trained on public data contains it. The people who carry this knowledge are the documentation, and they will not hand it to a vendor whose incentive is to automate them out of a job.
Timothy Williamson, in Knowledge and Its Limits, argues that knowing is a fundamental state rather than a composite that can be assembled from more basic ingredients like belief plus justification. You are not always in a position to know whether you are in a state of knowing. A condition can fully obtain while remaining invisible to the person standing closest to it, and no outside observer can reliably recover what has not been made legible. The knowledge inside a company behaves this way. Intelligence is not the same thing as usable context. A model can reason over what it can see, but the facts that matter most inside a company stay invisible until trust and proximity and operational contact make them available. Context is a living account of how work actually gets done.
There is a reason the labs started with code: code has cheap, open verifiers in compilers and test suites, so you can tell whether the output is correct. Business operations have no public equivalent, and the verification environment is the company itself. Parachuting engineers in for ninety days, deploying a model, and leaving won’t work. An outside team optimizing for inference revenue on a short clock has no way to earn what the work requires.
All of these observations make the firm more important. If cognition gets cheaper, the scarce asset is trusted access to the context that makes an answer useful and the accountability to own what happens after the answer becomes systematic action. The company is the container for the tacit knowledge, institutional memory, liability, and human judgment that turn intelligence into value.
The knowledge that makes a company work lives in the people who do the work, and unlocking it requires earning their trust. That means being on their side. It means designing the compression so that the person who reveals how the work actually gets done is rewarded with a better role, not a termination meeting. The traditional consulting model cannot do this because the vendor's incentive is perpendicular to the employee's interest.
6
The Company's Side of the Table
Every institution entering this space has a business model that points away from the company's interest:
Labs earn on token consumption.
Consulting firms earn on billable hours.
Platform vendors earn on adoption metrics.
This would never pass in other industries. A commission-based advisor earns more when you trade more, while a fee-only fiduciary's return is tied to your outcome. Every firm in this space today is structurally a commission-based advisor: it earns more when you consume more compute, cut more cost, or exit at a higher multiple. We believe that return should only be tied to the company itself getting better over years.
The companies that most need to evolve, the ones with deep institutional knowledge and real workforces and genuine room for compression, will not hand the keys to a partner whose incentive is to maximize consumption on someone else's exit timeline.
Transformation of this kind requires trust, and trust requires that your return depend on the company's outcome.
6
The Company's Side of the Table
Every institution entering this space has a business model that points away from the company's interest:
Labs earn on token consumption.
Consulting firms earn on billable hours.
Platform vendors earn on adoption metrics.
This would never pass in other industries. A commission-based advisor earns more when you trade more, while a fee-only fiduciary's return is tied to your outcome. Every firm in this space today is structurally a commission-based advisor: it earns more when you consume more compute, cut more cost, or exit at a higher multiple. We believe that return should only be tied to the company itself getting better over years.
The companies that most need to evolve, the ones with deep institutional knowledge and real workforces and genuine room for compression, will not hand the keys to a partner whose incentive is to maximize consumption on someone else's exit timeline.
Transformation of this kind requires trust, and trust requires that your return depend on the company's outcome.
6
The Company's Side of the Table
Every institution entering this space has a business model that points away from the company's interest:
Labs earn on token consumption.
Consulting firms earn on billable hours.
Platform vendors earn on adoption metrics.
This would never pass in other industries. A commission-based advisor earns more when you trade more, while a fee-only fiduciary's return is tied to your outcome. Every firm in this space today is structurally a commission-based advisor: it earns more when you consume more compute, cut more cost, or exit at a higher multiple. We believe that return should only be tied to the company itself getting better over years.
The companies that most need to evolve, the ones with deep institutional knowledge and real workforces and genuine room for compression, will not hand the keys to a partner whose incentive is to maximize consumption on someone else's exit timeline.
Transformation of this kind requires trust, and trust requires that your return depend on the company's outcome.
6
The Company's Side of the Table
Every institution entering this space has a business model that points away from the company's interest:
Labs earn on token consumption.
Consulting firms earn on billable hours.
Platform vendors earn on adoption metrics.
This would never pass in other industries. A commission-based advisor earns more when you trade more, while a fee-only fiduciary's return is tied to your outcome. Every firm in this space today is structurally a commission-based advisor: it earns more when you consume more compute, cut more cost, or exit at a higher multiple. We believe that return should only be tied to the company itself getting better over years.
The companies that most need to evolve, the ones with deep institutional knowledge and real workforces and genuine room for compression, will not hand the keys to a partner whose incentive is to maximize consumption on someone else's exit timeline.
Transformation of this kind requires trust, and trust requires that your return depend on the company's outcome.
7
AIVC: Rebuilding the Company From the Inside
AIVC rebuilds companies so that every person inside them does the most valuable work of their career.
AIVC partners with a company, redesigns its organizational architecture so that compression is rewarded and the work on the other side is better, and ties its own return to whether the transformation actually works. Advice alone is fee-for-service, the same incentive structure as everyone else.
Our transformation work generates knowledge about the market. The upside is captured in the company's own outcome. One loop.
Three structural features make this work.
Economically unaffiliated. AIVC has no inference revenue to protect, no model ecosystem to feed, and no platform to lock anyone into. It has strong views about how the work should change, but it is economically non-aligned with any vendor, which means that if the right answer for a company is less compute and better organizational design, that is the answer AIVC gives.
Accountability through real downside. Consultants get paid to advise, and the lab-backed ventures guarantee returns to their financial sponsors rather than to the companies being transformed. AIVC's position is structured so that its own return depends on the company getting better over years. When the transformation fails, AIVC absorbs the consequence.
A compounding world model for work. Every company AIVC goes inside becomes a structured data point in a continuously updated model of how organizations actually create value: where the constraints sit, what changes when you redesign around compression, what breaks, and what the downstream effects look like over months and years. Better models help, but the accumulated understanding of how companies actually change, and where the design has to flex and where it has to hold, is what widens the gap between AIVC and anyone who starts later. The world model for work is an evolving understanding of the relationship between organizational architecture and human capacity, built one company at a time.
7
AIVC: Rebuilding the Company From the Inside
AIVC rebuilds companies so that every person inside them does the most valuable work of their career.
AIVC partners with a company, redesigns its organizational architecture so that compression is rewarded and the work on the other side is better, and ties its own return to whether the transformation actually works. Advice alone is fee-for-service, the same incentive structure as everyone else.
Our transformation work generates knowledge about the market. The upside is captured in the company's own outcome. One loop.
Three structural features make this work.
Economically unaffiliated. AIVC has no inference revenue to protect, no model ecosystem to feed, and no platform to lock anyone into. It has strong views about how the work should change, but it is economically non-aligned with any vendor, which means that if the right answer for a company is less compute and better organizational design, that is the answer AIVC gives.
Accountability through real downside. Consultants get paid to advise, and the lab-backed ventures guarantee returns to their financial sponsors rather than to the companies being transformed. AIVC's position is structured so that its own return depends on the company getting better over years. When the transformation fails, AIVC absorbs the consequence.
A compounding world model for work. Every company AIVC goes inside becomes a structured data point in a continuously updated model of how organizations actually create value: where the constraints sit, what changes when you redesign around compression, what breaks, and what the downstream effects look like over months and years. Better models help, but the accumulated understanding of how companies actually change, and where the design has to flex and where it has to hold, is what widens the gap between AIVC and anyone who starts later. The world model for work is an evolving understanding of the relationship between organizational architecture and human capacity, built one company at a time.
7
AIVC: Rebuilding the Company From the Inside
AIVC rebuilds companies so that every person inside them does the most valuable work of their career.
AIVC partners with a company, redesigns its organizational architecture so that compression is rewarded and the work on the other side is better, and ties its own return to whether the transformation actually works. Advice alone is fee-for-service, the same incentive structure as everyone else.
Our transformation work generates knowledge about the market. The upside is captured in the company's own outcome. One loop.
Three structural features make this work.
Economically unaffiliated. AIVC has no inference revenue to protect, no model ecosystem to feed, and no platform to lock anyone into. It has strong views about how the work should change, but it is economically non-aligned with any vendor, which means that if the right answer for a company is less compute and better organizational design, that is the answer AIVC gives.
Accountability through real downside. Consultants get paid to advise, and the lab-backed ventures guarantee returns to their financial sponsors rather than to the companies being transformed. AIVC's position is structured so that its own return depends on the company getting better over years. When the transformation fails, AIVC absorbs the consequence.
A compounding world model for work. Every company AIVC goes inside becomes a structured data point in a continuously updated model of how organizations actually create value: where the constraints sit, what changes when you redesign around compression, what breaks, and what the downstream effects look like over months and years. Better models help, but the accumulated understanding of how companies actually change, and where the design has to flex and where it has to hold, is what widens the gap between AIVC and anyone who starts later. The world model for work is an evolving understanding of the relationship between organizational architecture and human capacity, built one company at a time.
7
AIVC: Rebuilding the Company From the Inside
AIVC rebuilds companies so that every person inside them does the most valuable work of their career.
AIVC partners with a company, redesigns its organizational architecture so that compression is rewarded and the work on the other side is better, and ties its own return to whether the transformation actually works. Advice alone is fee-for-service, the same incentive structure as everyone else.
Our transformation work generates knowledge about the market. The upside is captured in the company's own outcome. One loop.
Three structural features make this work.
Economically unaffiliated. AIVC has no inference revenue to protect, no model ecosystem to feed, and no platform to lock anyone into. It has strong views about how the work should change, but it is economically non-aligned with any vendor, which means that if the right answer for a company is less compute and better organizational design, that is the answer AIVC gives.
Accountability through real downside. Consultants get paid to advise, and the lab-backed ventures guarantee returns to their financial sponsors rather than to the companies being transformed. AIVC's position is structured so that its own return depends on the company getting better over years. When the transformation fails, AIVC absorbs the consequence.
A compounding world model for work. Every company AIVC goes inside becomes a structured data point in a continuously updated model of how organizations actually create value: where the constraints sit, what changes when you redesign around compression, what breaks, and what the downstream effects look like over months and years. Better models help, but the accumulated understanding of how companies actually change, and where the design has to flex and where it has to hold, is what widens the gap between AIVC and anyone who starts later. The world model for work is an evolving understanding of the relationship between organizational architecture and human capacity, built one company at a time.
8
What the Company Becomes
The corporation was designed by the logic of the division of labor. For two and a half centuries the human at work was a unit of capacity slotted into a process, and the corporation's job was to coordinate those units efficiently. That relationship has been fraying for decades. The shift from lifetime employment to gig work, from institutional loyalty to personal brand, from office to remote, all signals the same movement: people compressing their dependence on any single organization. Work is already compressing, but the corporation has not caught up as a form capable of capturing that energy.
AI accelerates the transition by compressing the routine, the repetitive, and the administratively burdensome all at once. A company designed around that compression accelerates everyone who walks in the door rather than shrinking its workforce. The AIVC response to someone automating half their job is harder problems and broader scope, more human than what came before.
The labs proved intelligence is abundant. The deployment ventures proved the gap between model capability and enterprise adoption is real. Neither is asking the question that matters: what does a company look like when its people actually want to compress themselves? Answering it requires being inside the company, redesigning how work is structured, giving people a reason to embrace compression, and owning the results.
The company is the atomic unit of capitalism.
If cognition is cheap, what should the company become?
The answer will not come from a lab or a platform or a financial sponsor. It will come from the companies that figure out how to make every person inside them do the most valuable work of their career. AIVC exists to build those companies from the inside.
8
What the Company Becomes
The corporation was designed by the logic of the division of labor. For two and a half centuries the human at work was a unit of capacity slotted into a process, and the corporation's job was to coordinate those units efficiently. That relationship has been fraying for decades. The shift from lifetime employment to gig work, from institutional loyalty to personal brand, from office to remote, all signals the same movement: people compressing their dependence on any single organization. Work is already compressing, but the corporation has not caught up as a form capable of capturing that energy.
AI accelerates the transition by compressing the routine, the repetitive, and the administratively burdensome all at once. A company designed around that compression accelerates everyone who walks in the door rather than shrinking its workforce. The AIVC response to someone automating half their job is harder problems and broader scope, more human than what came before.
The labs proved intelligence is abundant. The deployment ventures proved the gap between model capability and enterprise adoption is real. Neither is asking the question that matters: what does a company look like when its people actually want to compress themselves? Answering it requires being inside the company, redesigning how work is structured, giving people a reason to embrace compression, and owning the results.
The company is the atomic unit of capitalism.
If cognition is cheap, what should the company become?
The answer will not come from a lab or a platform or a financial sponsor. It will come from the companies that figure out how to make every person inside them do the most valuable work of their career. AIVC exists to build those companies from the inside.
8
What the Company Becomes
The corporation was designed by the logic of the division of labor. For two and a half centuries the human at work was a unit of capacity slotted into a process, and the corporation's job was to coordinate those units efficiently. That relationship has been fraying for decades. The shift from lifetime employment to gig work, from institutional loyalty to personal brand, from office to remote, all signals the same movement: people compressing their dependence on any single organization. Work is already compressing, but the corporation has not caught up as a form capable of capturing that energy.
AI accelerates the transition by compressing the routine, the repetitive, and the administratively burdensome all at once. A company designed around that compression accelerates everyone who walks in the door rather than shrinking its workforce. The AIVC response to someone automating half their job is harder problems and broader scope, more human than what came before.
The labs proved intelligence is abundant. The deployment ventures proved the gap between model capability and enterprise adoption is real. Neither is asking the question that matters: what does a company look like when its people actually want to compress themselves? Answering it requires being inside the company, redesigning how work is structured, giving people a reason to embrace compression, and owning the results.
The company is the atomic unit of capitalism.
If cognition is cheap, what should the company become?
The answer will not come from a lab or a platform or a financial sponsor. It will come from the companies that figure out how to make every person inside them do the most valuable work of their career. AIVC exists to build those companies from the inside.
8
What the Company Becomes
The corporation was designed by the logic of the division of labor. For two and a half centuries the human at work was a unit of capacity slotted into a process, and the corporation's job was to coordinate those units efficiently. That relationship has been fraying for decades. The shift from lifetime employment to gig work, from institutional loyalty to personal brand, from office to remote, all signals the same movement: people compressing their dependence on any single organization. Work is already compressing, but the corporation has not caught up as a form capable of capturing that energy.
AI accelerates the transition by compressing the routine, the repetitive, and the administratively burdensome all at once. A company designed around that compression accelerates everyone who walks in the door rather than shrinking its workforce. The AIVC response to someone automating half their job is harder problems and broader scope, more human than what came before.
The labs proved intelligence is abundant. The deployment ventures proved the gap between model capability and enterprise adoption is real. Neither is asking the question that matters: what does a company look like when its people actually want to compress themselves? Answering it requires being inside the company, redesigning how work is structured, giving people a reason to embrace compression, and owning the results.
The company is the atomic unit of capitalism.
If cognition is cheap, what should the company become?
The answer will not come from a lab or a platform or a financial sponsor. It will come from the companies that figure out how to make every person inside them do the most valuable work of their career. AIVC exists to build those companies from the inside.