the agent founder

notes for a new paradigm of human-agent interaction

abstract; in this piece, i draft an idea for an agent co-founder. i propose the "conceptual co-founder," as a new role within startups — an ai agent-human thought partner designed to support human founders via augmenting cognitive abilities, reducing coordination debt, and dampening the principle agent problem. the idea is to pair the best of ai— bias-free reasoning, scientific thinking, and impenetrable consistency—with the best of humaning— intuition, empathy, and trust. i outline seven core postulates behind the idea of a conceptual co-founder and sketch both the mvp and product vision.

---

co-founders and knowledge logistics

‘in a knowledge-rich world, progress does not lie in the direction of reading information faster, writing it faster, and storing more of it. progress lies in the direction of extracting and exploiting the patterns of the world… and that progress will depend on … our ability to devise better and more powerful thinking programs for man and machine.' — herbert simon, designing organizations for an information-rich world, 1969

startup wisdom tells us the hardest decision a founder faces is choosing a co-founder. should you go solo? how do you decide? is it about shared vision or work ethic or complementary skills? there are countless articles talking about co-founder dynamics. history = rags-to-riches and graveyard stories of successful and failed co-founder relationships. legendary partnerships — jobs and wozniak at apple, page and brin at google — on one hand. famous fallouts — zuckerberg and saverin at facebook, gates and allen at microsoft — on the other.

a great co-founder can be a 1000x amplifier or a -1000x headache. i’ve had both. a great co-founder > no co founder > bad co-founder.

ai will change the co-founder relationship, full stop. in an agent world, like with practically everything else, we can rethink the idea of a co-founder even is, because we can rethink what a company is. to understand the conceptual co-founder which i’m going to introduce, i’m going to interlude with a quick postulate on knowledge logistics.

postulate 0 — knowledge logistics as primary; in a llm driven era, the general constraint of a startup is knowledge logistics capabilities.

for the first time in history, founders can offload general cognitive output. the founder becomes less of a knowledge gather and more of a knowledge curator. things like manual research are now a product of ai driven systems. google gave the ability to access all knowledge of mankind, but now a founder can have it summarized, organized and contextualized. said differently, the human is no longer the organizational constraint, knowledge logistics — how information is organized and presented — are.

the most effective knowledge logistics tool is an interface — not just a repository of data, but the general ability to facilitate conversations and connect domains.

a good co-founder is an interface — someone who can bridge knowledge gaps, enabling rapid, context-sensitive decision-making by understanding and synthesizing information from diverse sources. if a co-founder can interface with an investor, a customer, the legal team, the product team and also say for instance, the history of complexity, that founder is going to be more valuable than one that can’t.

a major limiting step to integrating information (the challenge of knowledge logistics) is the absence of effective interfaces—individuals who can communicate across knowledge domains. different fields have distinct paradigms, tools, and mental models, making cross-disciplinary communication inherently challenging. — upcoming piece, a person trust system.

the best interface is always a human interface. the more general purpose the human is as an interface, the more valuable (and the harder to find). finding a founder that can incorporate complexity ideas into operations sounds great, but i’ve never met one. enter an llm and it’s hypothetical ability to (a) consume all of the information of history and (b) partner with a human co-founder (c) remove knowledge logistics constraints.

what if we could design a co-founder from history’s best? one could imagine having an intellectual giant like john von neumann as a co-founder, or an extremely disciplined individual like a jocko willink. how about a high growth mentality of a travis kalanick. or the operating genius of a george mueller? why not combine them all into one? with llm’s and agent based models, this all becomes possible.

this of course brings up question. what exactly is an agent co-founder? how would you use the co-founder in a day-to-day startup environment? what exactly would it do? is it just another ai agent or something different? next, i’ll introduce a few postulates to summarize introduce a conceptual co-founder as an ai-human agent hybrid.

agent co-founder (postulates)

‘we should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil. yet we should not pass up our opportunities in that critical 3%.’ — donald kuth father of analysis algorithms

‘getting the requirements right is a very difficult task, and therefore a task that is fraught with errors. an error that is caught during requirements development can be fixed for about 10% of the cost associated with an error caught during coding. errors caught during maintenance in the operation of the system cost about 20 times that of an error caught during coding and 200 times the cost of an error caught during requirements development.’ — dennis buede the engineering design of systems

1. postulate 1: the conceptual founder is non operational

   the conceptual founder’s value lies in improving thinking/ideas, rather than handling tasks. the conceptual founder does not have any direct responsibilities. this constraint focuses the product on exclusively being based on quality of thought. in a world in which every agent tool is trying to do a task, our focus is to avoid premature optimization and clarity of thought. practically, the only outputs are audio/message input or document creation.

   important; this is not to say that the conceptual cofounder is high-level and fancy thinking; they are equal parts in the weeds, commenting on docs and covering both professional and personal topics. the conceptual founder is as much about covering the obvious as it is about new ideas.
   ‘one lesson is that most accidents are not the result of unknown scientific principles but rather of a failure to apply well-known, standard engineering practices.’

   .

2. postulate 2: the agent is a human-agent hybrid.

   we keep the value of the human interface and supplement the human with the general capabilities of an llm and overtime, specific task execution abilities of an agent.

   

3. postulate 3: the conceptual co-founder should combine the greatest variety of ideas with the highest degree of order and fewest rules.¹

   the fewest rules = (a) the conceptual founder is non operational which generates the order. (b) the agent is not just an agent but a human-agent hybrid. the variety = founder should a (a) be very scientific like a von neumann (b) knows startups very well like a kalanick (c) be a genius at operational thinking (d) act as a true friend (e) be an equal parts an idea creator and idea synthesizer.

4. postulate 4: the frame is as important as ideas
   the conceptual founder brings clarity by being able to integrate ideas from a wide range of disciplines, and frame those ideas. framing is a level of abstraction above information presentation — and in a startup, we know how something is framed can dramatically affect the direction you take.²

5. postulate 5: early coordination for preventing costly errors
   the conceptual founder is highly focused on being a thought parter during initial conditions of things by providing lots of variety. the idea is to mitigate the cost of bad operational decisions but the long-term compounding effect of early misalignment.

6. postulate 6: all knowledge is grounded on scientific foundation

   as much as possible thinking has to be built on built on solid foundations (more on this later). as in detail in an upcoming essay, american business logic, there is a physics of organization and startup canonical axioms often violate them.

   

   this is crude and needs to be teased out, but that’s where the conceptual co-founder comes in

   startup advice (while awesome) a supermarket of facts without a foundation. disjointed tips rather than a cohesive framework. the conceptual founder would go through all of the science and understand to what extent startup advice is subject to the princial agent problem.

   the principal-agent problem occurs when one party (the principal) hires another party (the agent) to perform a task on their behalf, but the agent's interests don't align with the principal's. the agent might act in their own self-interest rather than in the best interest of the principal, leading to inefficiencies or conflicts. this often happens when the principal cannot perfectly monitor the agent's actions or when the agent has more information about the task than the principal. the idea is that the conceptual co-founder would not be subject to the general social dynamics of knowledge systems and have the capacity to able to process startup wisdom under a scientific frame.

7. postulate 7: a new abstraction layer: one could imagine that pairing a von neumann with a kalanick would create a co-founder capable of driving the kinds of thinking that lead to a new abstraction layer that focuses on what co-founders typically do not have the time for, specifically.

   

mvp and product

over the past several years, i completed the equivalent of a phd in coordination. i also have hands-on experience running and working for a high-growth organization. and more than anything, i am an idea-driven founder (which is hopefully obvious by this blog). i’m no von neumman-mueller-kalanick-jocko, but i think i have learned enough from each of them to provide a bare bones human as a service mvp.

the service could productize by (a) increasing how much the conceptual co-founder takes on and (b) how much of it is done by a human vs how much is done by ai. one could imagine an ai-human chatbot integrated into slack, just as one could imagine ingesting structured/unstructured data, or automatic document creation to operationalize ideas mentioned in the last postulate.

open questions / risks

• what does combining von neumann and kalanick actually look like?

• how does the conceptual co-founder to evolve as the startup grows?

• how do we measure success? how about compared to traditional human co-founders?

• how do we actually define what non operational means? (the lines seem very blurry)

• how do we not overwhelm the human co-founder with information?

• how do we think about customization needs of different startups?

1. the goal of your platform

2. ask chatgpt im sure u can get a million examples of how ths plays out. if i introduce a new product to market. the way the product is framed—whether as a new innovation or a slight improvement—sets the tone. if framed selling as service vs a challenge everything changes. a related book here is metaphors we live by.

related to this, i wrote about the idea that the data structure is more important then the data, in q2q towards a new data paradigm.