The Distillation
What happens when the system built to diffuse ownership meets technology that concentrates it.
The old way of working diffuses context and accountability. The new way concentrates them. That drives a reckoning of value and ownership.
Most organizations are running on workflows that have never been documented, data that has never been cleaned and processes that exist because someone figured it out years ago and everyone else just followed.
A shocking number of companies still have a critical spreadsheet named after the person who started it. That is not a failure of intelligence. It is the logical output of a system that never required anyone to see their own work from a bird’s eye view.
Knowledge was distributed to minimize dependency and maximize execution. Companies did the knowing. People did the executing.
That worked until the execution became the thing you were building an agent to do. The previous model of working rarely demanded a structural account of the work itself. Using an LLM to craft an email is not the same thing as directing an agent to do your job. One is a simple tool. The other requires you to understand your own work well enough to architect, manage, replicate and quality control it. A decade of using search engines over our memory has trained us to locate rather than grasp. We readily conflate the locating with the knowing. Just like we confuse the execution of a task with the architecture of a workflow.
If it were only a capability gap that’s trainable with better prompts and better interfaces. The bigger hurdle is clarity. Clarity requires understanding contextual tradeoffs and third order effects. The ability to know not just what to build but why to build that instead of something else. Most people are waiting for more information before they decide. Clarity means they already know which information matters. That is not a workflow problem. That is a judgment problem.
Judgment of that kind is the accumulated residue of decisions made and consequences observed until it’s instinct. The more tacit knowledge someone has the easier they make the work seem. That makes it dangerously easy to underestimate and tough to replace. Time is the true teacher and that’s a luxury we’re actively cutting. Median tenure in the private sector has been just under four years and declining. People move on with pattern recognition but not consequence memory.
Even with sharp capability and clarity, people are not attached above all to outcomes. They are most attached to what is theirs. Better branding beats better products because identity trumps results. Over half of workers with AI options available still opted to do their work manually. Replacing tasks means giving up not just a habit but a form of power and expertise. Most organizations are not set up yet to reward that trade. They are only set up to say they reward that trade.
The same system that concentrated knowledge at the institutional level concentrated risk at the individual level. Buying has usually been safer than building. The enterprise software contract comes with a vendor to blame if it breaks. The McKinsey recommendation comes with a firm to point to if it fails. That calculus doesn’t evaporate because the technology got more powerful. The operations director who automated order fulfillment and broke it for days pays a steep reputational price even if the CEO is “excited about AI.” The activation energy required to build something even through natural language, own it when it breaks and defend it is still real.
That assumes people would choose to build even if they could. People buy pre-chopped strawberries for 5x the price. Convenience is not a failure of ambition. It is a rational response to finite time and attention and it runs through every level of AI adoption. The question is never only "can we build this." It is "can we build this and should this take time right now and can we defend it when we hit an error?" Not because the building is hard but because the defending is expensive. Building concentrates accountability. Buying diffuses it.
Those excited to build do so locally. At scale when agents proliferate function by function and person by person they encode the same siloed logic that predated the tech. The agent managing cash flow tightened payment terms last Tuesday. The agent managing supplier relationships promised flexibility to the same vendor on Wednesday. Without a shared context each becomes a local solution to a local problem. The undocumented institutional knowledge becomes a collision point. The inability to see end to end becomes a liability.
For decades the system was designed so that no single person had to understand the full picture. That was a feature. It allowed organizations to scale by making individuals interchangeable and keeping the institution as the ultimate holder of knowledge and system. Now the people who cannot see their own work from above have nothing to direct. And the people who can see and direct it clearly become disproportionately valuable.
This is what adoption actually offers and why most organizations will flinch. Radical visibility into how the work actually fits together. Who owns decisions that cross boundaries. Where institutional knowledge lives and whether it is consistent or contradictory. Which judgment calls were deliberate and which were just inherited habits. The difference was hard to measure when both looked like productivity.
Most leadership teams will say they want clarity. Fewer will tolerate what clarity actually reveals. That concentration of context and accountability is not a side effect to manage. It is the point.