Systems of Agentic Leadeship

For most of history, our systems for organizing information have been passive and laborious. The notebook waits for us to write in it, while the note management system requires your dedication and effort. AI Agents change this completely. They allow us to create systems that work alongside us, actively administering our management systems for us, automatically cataloguing, sorting, summarizing, and surfacing the right stuff at the right time. A well-built agentic system for managing your team or organization is far more than just a tool: it is a genuine extension of your mind, integrating cognition into the fabric of the world around you.

In 1998, two philosophers, Andy Clark and David Chalmers, published a short paper called “The Extended Mind.” It contains a thought experiment with a somewhat startling conclusion: your mind extends into the world around you via the tools and systems you use.

Their idea starts with two friends, Otto and Inga. Both want to attend an upcoming event at the Museum of Modern Art. Inga hears about it, makes a mental note, and on the day of the event she recalls that the museum is on 53rd Street and heads over, arriving on time. Otto also wants to go, but he is forgetful. Instead of using his memory, he uses a notebook that he carries everywhere. He has recorded the date, time, and location of the event, and on the day he looks up the address, reads “53rd Street,” and walks there, arriving fashionably late.

The question Clark and Chalmers ask is deceptively simple: is there any real difference between what Inga did and what Otto did? Inga’s knowledge that the museum is on 53rd Street lived in her head; Otto’s lived in his notebook. But functionally, in the way it was formed, stored, retrieved, and acted upon, the notebook played exactly the role that biological memory played for Inga. This conclusion was considered radical for its time and has aged remarkably well. The argument is that Otto’s notebook is part of his mind, not just a tool his mind uses. Cognition, they argued, does not stop at the boundary of the brain. It extends out into the world around us, into the objects and systems we couple ourselves to, which become load-bearing parts of how we think.

One very important insight is that the notebook is only an extension of the mind insofar as there is a system in place to manage it. A notebook without a system is just a stack of paper. I have a drawer full of them that I will never use. Systems like Getting Things Done, Zettelkasten, and PARA create structure that makes your notes useful, extending your cognition into the real world, allowing you to offload the work of remembering, connecting, and organizing your thoughts. The system is what makes the notebook part of your mind.

But, perhaps like you, I have tried and abandoned many of these systems over the years. It is easy enough to get started, but difficult to maintain over time. They require effort, dedication, and discipline to be effective, and the moment you stop attending to the rules, the effectiveness collapses.

I have spent the last 10 years in Enterprise-scale Software Engineering leadership roles at one of the largest companies in the world. It is complex, varied, detail-oriented work, with attention needed involving technical details, operational systems, feature briefings, strategic planning, organizational issues, people operations, project management, quality management, and much more.

I absolutely love my work because each day the activities are diverse, engaging, and fascinating. But every person in a role like mine struggles with the sheer amount of information we have to process each day, minute to minute. Throughout the day, signals arrive from a dozen systems: Jira, Confluence, GitHub, chat, dashboards, mail, the hallway conversation, the triple-booked 9am meeting slot. A sort of paranoia develops, a constant fear of missing important info or tasks. With so many things happening around us, the bottleneck for a modern engineering leader is focus and attention. There is always more information to process than one can consume. Artifacts fly across the organization, and context evaporates the moment you close the tab.

For centuries, the solution to attention management has been the same: Write it down. Take notes. The Stoics kept their hypomnemata. Renaissance scholars kept “commonplace books.” John Locke thought the practice important enough to publish an entire method for keeping one in 1706, complete with an indexing scheme. These were external memories, Otto’s notebook avant la lettre. They worked because a page stays still while your mind wanders.

But every one of these extensions, from Locke’s commonplace book to Otto’s notebook to the Notes app on your phone, shares a limitation. They are passive. They remember only what you remember to write. They notice nothing. They connect nothing. Otto’s notebook still required Otto to open it, to have written the address in the first place, to know where to put the details, and to find it later when it is needed. The extended mind stored his cognition, but it never contributed anything. And for the system to work, it is on us to act according to its rules.

So here we arrive at what has fascinated me since the inflection point of powerful AI Agents last December: What happens to the extended mind when the notebook can think? When it can write on its own and act on your behalf? When it can manage the system rules for you? The answer is a massive opportunity for how we lead. With that in mind, I have been vigorously exploring it, along with other leaders on my team.

Clark and Chalmers were careful to define when a tool ascends to become an extension. Not everything you use is part of your mind. They offered conditions for deciding when an external resource earns the status of genuine cognitive extension. I want to lift three of them out and apply them here, because I think they are the clearest design criteria I have found for guiding my thinking about agentic management systems. Then I want to add a fourth criterion that Otto’s paper notebook could never satisfy.

1. Availability. The system must be a constant in your life, reliably followed and there whenever you need it. The notebook works because it is always in your pocket. A note you can only sometimes reach is not memory; it is archived. For a knowledge system, this means the substrate has to be durable and portable in the deepest sense. For us, the memory layer is organized simply around markdown files, grouped in folders, stored locally. No proprietary database, no cloud vault I can be locked out of, no interface that my agent can’t access (looking at you OneNote!). Markdown files are a simple and consistent way to store information, easy to jump in and out of, and simple to back up and share. Availability is not a feature you add; it is a property of the medium you choose to use. 2. Accessibility. The system must be directly and easily integrated into your environment, with minimal friction between thought and action. This is the condition that causes most systems to fail, and they fail by making you do something unnatural: switching context, recalling arbitrary rules, breaking your flow. Context switching is the attention killer; the damage is already done. The extended mind cannot create hurdles for you; it has to operate smoothly in the places you work. We have to invert the geometry: instead of you going to it, it comes to you. It lives in your tools, it watches your work alongside you, it acts on relevant signals, omnipresent and non-obnoxious. The big win comes from developing your own software to augment each surface you work in, extending your mind into your environment, meeting you where you are. 3. Reliability. When Otto checks his notes, he does not pause to interrogate whether he can believe it. He trusts it automatically, the same way you would trust your own memories. If you have to double-check what your system tells you (if you have to ask “is this up to date? is this formatted right? did this land in the correct place?”), it has ceased to function as an extension of the mind and has reverted to being just a tool. Reliability is what turns your system into extended cognition. In practice this means the system has to be disciplined: consistent formatting, correct structure, continuous verification. The right context in the right file, organized for the reader. You have to provide the rules to make the system a reliable partner.

Availability, accessibility, reliability. Your system has to have all three, or it is nothing but a stack of paper.

Here is what the agentic revolution adds: the thing no notebook in the history of notebooks could do. The system can now act.

• It can observe. When a customer issue enters the chat, the system can notice (genuinely, without being asked) and update your customer profile before you’ve finished thinking about it.
• It can write. You open a Jira ticket. A review is immediately kicked off, running in the background, with a durable artifact explaining the ticket while you read.
• It can connect. The thing you’re looking at right now has the three old notes that relate to it, the ones you had forgotten you wrote but are now right beside it.
• It can enrich, while you sleep, weaving links and metadata through your hasty captures so the graph stays connected even when you were writing too fast to connect it yourself.

I have started calling this fourth property “bounded initiative,” and the two words carry equal weight. Initiative, because the extension of the mind is no longer passive; it does cognitive work on its own, the noticing and formatting and filing and cross-referencing that used to be pure overhead and difficult to maintain. Bounded, because the whole thing collapses without limits. An extension of your mind that acted without boundaries would not be an extension; it would be a second, unaccountable mind wearing your memory as a costume. The instruction files, the defined skills, the action triggers, and the information architecture are not bureaucracy. They are the rules that make the system safe to trust, which brings us right back to condition three. Initiative makes the system self-managing, but initiative without bounds kills consistency and makes it unreliable. You need both, held in balance.

This condition, or perhaps we should say capability, is the difference between an extended memory and an extended mind. Locke’s commonplace book, Otto’s notebook, your Notes app: these extend memory. But they hold it still and wait. Extending your mind with an agentic system integrates cognition; the system participates in the thinking and takes care of routine administrivia. The notebook, incredibly, writes itself. Locke could never even have imagined it.

The smallest act in our system is a single captured thought: a quick note, a few keystrokes, perhaps a sentence about a customer. The system ensures that the note is immediately processed, routed, and connected to relevant artifacts. But zooming out, the same four conditions that govern my agentic management system can be extended to the organizational level.

For example, each of my team leads has been developing their own agentic management system. Their weekly activities roll into a record that is shared among us. Every Friday the raw material summarizing activities across my whole organization arrives automatically, accumulated at the team-leader level throughout the week by systems they have implemented to attend to their work. These reports give me a comprehensive view that I can use to generate a weekly report for my boss. These are the best weekly reports I have ever created in my life, because I now spend all my time thinking, analyzing, contemplating, and introspecting, instead of searching, collating, aggregating, and formatting.

The judgment is still mine: what matters, what to escalate, what to celebrate. That never offloads. But the collection does, and collection was never a good use of anyone’s time. The agentic system does the remembering, connecting, and filing. The extended mind is not a replacement for judgment; it is a replacement for the overhead of administering a system that would otherwise require way too much care and feeding to exist.

The extended mind, it turns out, does not stop at our notes. It can be integrated with all the systems around it. We have built custom software to integrate it into our tools and applications. We even link our individual systems together to create an organizational hive mind: a living, evolving socio-technical system that spans people, artifacts, and time. It remembers what no single person could ever remember, and it amplifies our effectiveness beyond what we could ever have hoped to achieve in the past.

Nevertheless, we should be realistic about where we are: it has only been 6 months since the coding agent inflection point, which gave us the agents we need to manage our system, and made it trivial to build the tools needed to power it: desktop apps, browser extensions, server infra, and more.

We are still in the “What is this? What is possible?” phase. It is an exciting and fascinating time, similar to where we were with Agile twenty-odd years ago and DevOps ten years back, when the practices were real but the shape of them was still emerging through our experiences. I do not know what our agentic management systems will become, but I am having an incredible time experimenting with different ideas and seeing what works.

Clark and Chalmers gave us the first three criteria for extended mind systems decades ago, but now we can unleash that potential with the advent of agents and “bounded initiative.” The question is no longer a personal matter of whether we can follow through on a commitment to a management system like GTD. The question is what we can build to operate that system for us and integrate it into our lives.

Over the coming weeks I will share more on how we are approaching building these systems into the organization I work in, and the lessons we are learning along the way.

I have taken to calling the idea of agentic leadership systems and the extended mind “AMOS,” the Agentic Management Operating System. I have been building mine for 6 months now, and it is still a constant work in progress. It has already had a big impact on the way I think and lead. I am excited to share more about it, and I hope to hear from others who are exploring similar ideas.

Have you built systems that help you lead your team or organization? Have you felt the moment a tool stopped being something you use and became an extension of you, one that thinks with you? I would love to hear about it. Reach out and tell me what your extended mind has started to notice on its own.

---

Cover Art: “The Librarian,” by Giuseppe Arcimboldo, c. 1566. A portrait of a person composed entirely of books: no face beneath the volumes, just the volumes arranged into a face. Arcimboldo painted it as satire. Four and a half centuries later it reads as prophecy: a mind made of its records, a self assembled from what it chose to keep.

Thanks for reading! Subscribe for free to receive new posts and explore Agentic Management Systems with me!

Subscribe