I think we are underestimating how strange work is about to feel.
The first stage of AI adoption is usually individual. Someone discovers that a report can be summarized in minutes. A product manager realizes they can synthesize customer interviews without waiting for the next research cycle. A consultant can pressure-test a strategy hypothesis before the first workshop deck is even finished. A customer-success team can detect a repeated product issue across dozens of conversations before it becomes a formal escalation.
At that level, the productivity story is real. AI gives people leverage. It reduces repetition. It makes unfinished thinking easier to revisit. It turns documentation from a static obligation into something closer to reusable working material.
But then the person hits the workflow.
The roadmap still waits for quarterly planning. The client still expects discovery theater. The signal from the field still has to pass through layers of interpretation before product leadership treats it as evidence. The manager still coordinates by assigning tasks, even when the better job is to define sharper objectives, constraints, and judgment gates.
That is where AI starts to become interesting.
Not because it makes people faster, but because it makes the old workflow look too small.
For years, companies have explained slow progress with phrases like “lange Entscheidungswege”, long decision paths. Decisions move through too many people, too many forums, too many approval steps, too many rituals that exist to create confidence before action.
But AI changes the scale of the mismatch. A long decision path is annoying when the underlying work is also slow. It becomes absurd when individual pieces of the work can be explored, tested, summarized, and improved at AI-assisted speed.
The issue is no longer simply that the organization is slow.
The issue is that the organization was designed around a different scarcity.
Individual productivity is not the whole story
A lot of AI productivity advice assumes the company around the worker stays the same.
The person becomes faster. The process remains intact. The meeting cadence remains intact. The approval logic remains intact. The reporting ritual remains intact. The handoff remains intact. The business case remains intact. The quarterly planning cycle remains intact.
That is a comforting version of AI because it does not ask leadership to change very much. It says: keep the organization, add intelligence to the edges, and enjoy the efficiency gains.
There are gains there. I use AI every day for exactly this reason. Repetitive work like reports, summaries, research preparation, synthesis, and first-pass structuring has changed dramatically. The first question is no longer only, “Can AI write this?” The better question is, “How should I structure my documentation so AI can output what I need, when I need it, in the format I need it?”
That sounds like a small shift. It is not.
Once documentation becomes reusable context, the nature of the work changes. Notes are no longer just memory aids. They become material for future synthesis. Meeting summaries are no longer administrative cleanup. They become inputs into decisions. Articles, research, applications, conversations, and performance data become a living body of evidence rather than scattered artifacts.
This is what I mean by Cognitive Leverage. AI is most useful when it changes the ratio between thought, context, and action. It is not just a tool for producing more words. It is a way to make thinking compound.
But once that starts working at the individual level, the bottleneck moves.
The person can now produce a better summary, faster. The organization still asks for the same status deck.
The person can now identify the pattern earlier. The organization still waits for the formal forum.
The person can now test the hypothesis cheaply. The organization still behaves as if the only legitimate path to confidence is a months-long process.
This is where individual productivity becomes the least interesting part of AI.
The workflow becomes visible
AI exposes workflows because it changes the speed of the work inside them.
A workflow can look reasonable when every step takes time. If research takes weeks, synthesis takes weeks, alignment takes weeks, and decision-making takes weeks, the whole system feels slow but coherent. Each step seems to justify the next.
But when one person can compress parts of that chain into days or hours, the workflow starts to reveal which parts were designed for value and which parts were designed for reassurance.
Status decks are a good example.
They have felt strange since business intelligence dashboards became normal more than a decade ago. If the data exists, and the state of the work can be seen, a deck should not be the primary way the organization learns what is happening. Yet status decks survive because they are not only information products. They are rituals of control. They help leaders feel that work has been translated into a form they can approve, question, or escalate.
AI makes that ritual even harder to defend.
If a project space, CRM, product system, research repository, and meeting transcript can be summarized into an executive-ready view, then the question is no longer whether humans can prepare the status deck. Of course they can. The question is whether the status deck is still the right object.
The same is true for handoff meetings.
Communication between humans should not disappear. That would be a terrible reading of the AI opportunity. The point is not to automate human contact out of the company. The point is to stop using human meetings as the default compensation mechanism for bad information design.
Information can now be summarized better than ever. Context can be preserved better than ever. Decisions can be prepared better than ever. Open questions can be extracted better than ever. If the organization still insists that every handoff requires another meeting because the summary is not perfect, then it is choosing a very expensive version of imperfection. AI hallucination should be a bullet point on the agenda, not an excuse to revert to manual handoffs.
That compromise will be hard for many companies to learn.
The old standard was human effort. If someone spent hours assembling the story, the output felt legitimate. The new standard has to become usefulness under judgment. AI-assisted summaries will not be perfect. Human summaries were never perfect either. The difference is that we have been trained to forgive human imperfection when it arrives wrapped in effort.
AI removes some of that effort, which makes the remaining judgment more visible.
That is uncomfortable, but useful.
Field signal is moving faster than product leadership can absorb
The clearest example for me is the distance between customer signal and product decision-making.
Most companies say they want to be customer-led. Many of them are sincere. But the actual operating model often treats customer signal as something that must be filtered, packaged, prioritized, normalized, and translated before it becomes legitimate.
Sales hears objections. Customer success hears repeated friction. Support sees failure modes. Consultants see where clients misunderstand value. Account teams notice when the story that worked last quarter no longer works.
Individually, these signals are often dismissed as anecdotes.
That used to be understandable. Qualitative signal was expensive to consolidate. It took time to gather. It took judgment to interpret. It took people to compare patterns across calls, notes, tickets, emails, transcripts, and commercial conversations. By the time the signal reached product leadership, it had often become either too filtered to be useful or too anecdotal to be trusted.
AI changes that equation.
A team can now consolidate layers of qualitative data much faster. It can look across customer calls, support notes, sales objections, onboarding friction, renewal risks, product feedback, and internal commentary. It can identify recurring themes. It can separate loud anecdotes from repeated patterns. It can prepare sharper questions for product and leadership.
But what happens if there is no way to act on it?
That is the real bottleneck.
The field can learn faster than the product system can absorb. The organization can generate evidence faster than the governance model can interpret. The customer-facing edge of the company can become more intelligent while the center still waits for the old planning cycle.
This is why the old workflow is too small for AI.
It is not that AI magically solves product strategy. It does not. Human judgment becomes more important, not less. Someone still has to decide which signal matters, which segment it represents, which customer pain deserves roadmap attention, which pattern is commercially meaningful, and which request should be ignored.
But the shape of the work changes.
The scarce resource is no longer only the ability to gather and summarize signal. The scarce resource becomes the ability to absorb signal, judge it, and change direction without turning every decision into organizational theater.
That is a very different kind of company.
Some workflows only existed because cognition was expensive
Here is the uncomfortable part: some workflows only made sense because cognition used to be expensive.
They were built for a world where analysis took longer, synthesis was scarce, comparison across sources was manual, documentation was static, and summarization required human time. In that world, it made sense to batch work. It made sense to create stages. It made sense to limit who could access what. It made sense to filter information upward. It made sense to turn evidence into decks because the deck was the only practical way to compress complexity into a leadership-consumable object.
AI does not eliminate the need for structure. It does not remove the need for accountability. It does not mean every employee should improvise around the operating model.
But it does make the design premise visible.
- If cognition is less scarce, why are so many workflows still built as if every act of synthesis must be rationed?
- If analysis can happen earlier, why wait for the formal phase?
- If qualitative data can be consolidated continuously, why treat customer insight as a quarterly artifact?
- If strategy hypotheses can be tested cheaply, why sell certainty before the evidence is strong?
- If summaries can be produced on demand, why do humans still spend so much time performing context transfer?
These are not tool questions. They are workflow questions.
And eventually they become value-chain questions.
This is also why I think the conversation around AI adoption is often too polite. It focuses on training, usage, licensing, prompt quality, and productivity. Those topics matter, but they avoid the more difficult implication: if individual use cases start proving that work can move differently, the organization has to decide whether it will let the value chain change.
Microsoft’s 2026 Work Trend Index describes a similar tension as a “Transformation Paradox”: employees are ready to reinvent how they work, but metrics, incentives, and norms continue to reinforce the old way. The same report found that only 26% of surveyed AI users say their leadership is clearly and consistently aligned on AI, while 45% say it feels safer to focus on current goals than redesign work with AI. That is not a tool adoption problem. It is a system problem. Microsoft’s framing is useful because it names the mismatch between individual agency and organizational conditions.
McKinsey’s 2025 State of AI research points in the same direction from another angle. AI use is widespread, but enterprise-level impact remains uneven. The high performers are not merely using more tools. They are much more likely to redesign workflows, embed AI into business processes, and define when human validation is required. That distinction matters because it separates AI activity from operating impact.
Asana’s 2025 State of AI at Work report puts it even more directly: AI becomes another layer of complexity when it is applied to broken systems, while the organizations getting more value are redesigning work around it. That is the point. The workflow around the tool determines whether AI becomes leverage or just more noise.
The gate still feels important
There is another reason this will be hard: many approval loops are not only about risk. They are about identity.
Being the gate feels important.
I have seen good template work slowed down by approval loops that were not really improving the work. I have seen methods such as delegation poker struggle to gain traction because the theoretical logic was less powerful than the emotional reality: some leaders did not want to give up the feeling of being the place where permission happens.
That is not a personal criticism. It is an organizational pattern.
Most leaders were trained in systems where control was a proxy for responsibility. If information was scarce, if synthesis was expensive, if mistakes were hard to detect early, then approval gates felt like responsible management. They created a sense that risk was being held by someone senior enough to carry it.
But AI changes the timing of evidence.
If a team can show working examples faster, test variations faster, compare customer signal faster, draft alternatives faster, and surface risks earlier, then the approval point should move. It should not disappear. It should become more precise.
The human role shifts from being the default gate to being the validation point that matters.
That distinction is crucial!
AI should not remove judgment from work. It should stop wasting judgment on things that no longer deserve so much human coordination. The best leaders will not be the ones who approve every step. They will be the ones who decide where judgment genuinely changes the outcome.
That is why I like the idea of human judgment as a validation gate.
The gate should not exist to preserve hierarchy. It should exist where the cost of being wrong is high, where ethical judgment is needed, where tradeoffs are real, where accountability matters, or where the organization has to choose between competing versions of value.
Everything else deserves to be questioned.
Be the change inside the workflow
It is easy to make the old workflow look ridiculous. It is harder, and more useful, to build a better one in miniature.
That is how I think about my own work at Open Source Consulting.
In May, the content system was still fairly manual. There was essentially one thread for content generation, and I updated CSV files to improve context. That worked, but it carried administrative overhead. The system depended too much on me moving context around.
In June, that changed materially. The admin overhead dropped by around 90%. The work now runs through three MCPs and five individual agents. The agents do not replace my judgment. They depend on it, they are instructed to ask for it. They need me for pattern recognition, past stories, prioritization, and opinions on polarizing topics. But they no longer need me to do every piece of context transfer by hand.
That is a small example, but it matters because it is not theoretical.
The improvement did not come from asking, “How do I make the existing workflow a little faster?” It came from asking, “What would this workflow look like if the context did not have to be manually reconstructed every time?”
That is the deeper AI question.
The point is not to automate everything. The point is to make the workflow worthy of the leverage now available. The point was to design it knowing that AI existed and investing in becoming knowledgeable of the potential it brings.
My own constraint is also obvious. I do not yet have the coding skills to build every system I can imagine. I also have to allocate time deliberately to improve the workflow instead of only running it. So every week needs time for improvement. Not because improvement is a side hobby, but because the workflow is now part of the work. And with every week, the iteration gives me additional capacity back to improve it further.
This is where I think many professionals can begin.
Do not wait for the whole company to redesign itself. Start by making one workflow more intelligent. Make documentation reusable. Make summaries dependable enough to reduce unnecessary handoffs. Make customer signal easier to compare. Make your own work legible enough that AI can help you retrieve, reshape, and apply it later.
Then pay attention to where the organization resists.
That resistance is information.
- If the improvement works locally but gets stuck in approval, the approval model is part of the problem.
- If the signal becomes clearer but nobody can act on it, the decision rights are part of the problem.
- If the summary is good enough but people still demand another meeting, the trust model is part of the problem.
- If the hypothesis can be tested cheaply but the organization still wants to sell or fund a large program before evidence exists, the value chain is part of the problem.
That is where personal productivity turns into organizational learning.
Vertical scaling and the new worker
I don’t claim credit for the idea of Vertical Scaling and Polymath thinking through AI, but I think it connects strongly to this question and I feel strongly inspired by it.
If AI expands what one person can absorb, the bottleneck shifts toward people who can move vertically across tool, workflow, operating model, and value. The valuable person is not only the one who can use the tool; it is the person who can see what the tool implies for the work around it.
That is different from being a generalist in the shallow sense.
The point is not to know a little about everything. The point is to connect layers that used to be separated: the customer conversation, the product implication, the workflow constraint, the operating model, the commercial logic, and the human judgment required to make the whole thing useful.
AI makes that kind of vertical movement more realistic. It also makes it more necessary.
When a sales or customer-success team can consolidate field signal faster, someone has to understand what that means for product. When a consultant can validate hypotheses faster, someone has to understand what that means for the economics of professional services. When an individual can produce more output, someone has to understand whether the output deserves to exist.
This is where the Human Premium becomes clearer. AI makes ordinary cognitive production cheaper, but it raises the value of people who can judge, connect, prioritize, and redesign. I wrote about this more directly in The Human Premium Enterprise, but the workflow layer is where the idea becomes practical.
Human value does not disappear when AI accelerates work.
It moves closer to judgment, synthesis, trust, and design.
The organization of today was not designed for AI
The organization of today was not designed for AI.
That sentence sounds dramatic, but I think it is increasingly practical.
Most organizations were designed around scarce cognition. They batch decisions. They filter information upward. They separate product teams from customer signal. They coordinate through meetings. They protect leadership attention by compressing reality into decks. They use approval gates to manage risk. They fund large programs because small evidence was historically hard to generate.
AI fundamentally challenges that design premise.
The organization of tomorrow will be designed differently: some of those organizations will be new from the beginning. They will not have to unwind the old value chain because they never inherited it. Think of Tesla being the first “software-first” car manufacturer, creating self evidence towards topics like “smartphones as car keys” or “over the air software updates”. Others will be existing companies willing to take the leap of faith early enough to learn before the market forces them.
That does not mean reckless transformation. It means evidence-led redesign.
Start where AI is already showing that work can move differently. Look for the individual use cases that are not just saving time, but revealing a broken handoff, a delayed signal loop, a pointless approval, a reporting ritual, or a decision path that made sense only when cognition was expensive.
Then ask the uncomfortable question: is this workflow still necessary, or was it just the shape work took when thinking was slower?
That question is not anti-management. It is a better management question.
It asks leaders to protect what still deserves human judgment and remove what merely consumes it. It asks companies to stop treating AI as a productivity layer on top of the old organization and start treating it as evidence that the workflow itself may need to change.
That is where the real opportunity begins.
The old workflow is too small for AI.
The useful response is not cynicism. It is redesign.
