Open Organizations and End-to-End Product Generators
(Unrelated photo - the northern lights have just been amazing this year)
What’s a “Product?” How are products developed, and how do they create value?
Well, if I tried to describe this to 10 year old me, I would say something like: a group of people work together to try to identify needs and opportunities, then they build something that addresses those needs or opportunities. Then that group tries to put the things they built in front of people, and if they do this effectively others will tend to pay money for it, and this then supports our group and allows them to do this over and over again.
With that framing, maybe we would call a Job “some agreement to contribute to a project together with a group.” People collaborate to create “more than the sum of their individual labors” and then find some way to distribute the value they create together. It’s a little unintuitive that people can come together to create “more than the sum of their individual labors” — How’s that work? I like to think of it like many small creeks and streams coming together to form big rivers. Small creeks and streams (like uncoordinated labor) have lots of friction, they weave their ways around many rocks and sticks and flow in a chaotic, messy way. Big rivers on the other hand are extremely streamlined, as they insulate themselves from the rocks and hills and quickly moves obstacles like fallen logs. When we work for big organizations, we tend to benefit from similar kinds of efficiencies, and we call it “Economy of Scale.”
But big organizations also have lots of downsides - they can struggle with bureaucracy, making it difficult to innovate or respond to changes. They can be “inhuman systems” that alienate individuals from their impact of their efforts. They can be bloated and wasteful and have misaligned incentives. Organizations are vulnerable to the rampant egos of poor leadership and can create multi-polar traps and short-term biases. Organizations aren’t perfectly optimized systems - they often fail catastrophically, create products that suck, extort or harm users, pollute the environment, mistreat employees, and a vast swath of other issues that everyone complains about. Despite these downsides, organizations stick around as useful entities.
I’ve thought about alternatives to these kinds of systems for most of my life, being a compulsive systems thinker. One idea that has really stuck in my head over the years is the “Open Source” movement and Yochai Benkler’s writing on commons-based peer production systems like Wikipedia. Functional methods from open-source software development are still just beginning to ripple out into other ecosystems - many of the same lessons from “Open Source” movements may be successfully applied to physical products, recipes, business models, marketing strategies, management, or other fields. If we extrapolate some of these ideas, what do “Open Source” organizations look like? What are some incentive structures for “Open Source Value Creation” that allow us normal humans to be as supported (or more supported!) than traditional organizations? And I mean actually supported, in the sense of “being able to buy/live in a house.” I don’t think it’s enough that open-source movements should survive purely on the donated labor of individuals - contributing to these systems should give normal people resources that make their lives easier.
One of my early, messy causal loop diagrams (read from bottom to top) exploring stages of developmental pipelines and feedback dynamics between various agents in (and out of) organizations
One idea in this space is “Open Organizations” which are starting to appear experimentally - their operating principle is usually some shade of “what if at each stage of organizational development, anyone could step in, contribute improvements or work, and push projects or initiatives forward?” There are some Decentralized Autonomous Organization (DAO) frameworks exploring possibilities in this space to further economize and modularize these dynamics of economy of scale and Value Creation. We all have firsthand experience with the inefficiency of businesses - what if there are more frameworks in niches like “Wikipedia” that cause systems to be more efficient? Or more profitable? More broadly, what other systemic or management frameworks are available now that just haven’t been possible before?
Bringing in New Tools
I was out on a walk the other day thinking about AI systems (which happens all the time now) and it seems like we are already outsourcing sections of organizational pipelines to AI. Many of us are already using AI tools to help us in our day jobs, where we handle some part of this “Value Creation” pipeline for our respective organizations.
One thing I’ve seen lately are experiments where AI agents are given ‘roles’ in organizations composed of other AI agents - for example this project that simulated a game development company made up entirely of AI Agents. This is all fun and interesting, some evidence that people are starting to experiment with which organizational roles might be offloaded to AI systems.
I think an important element here is that we have already modularized organizations into “Job Titles” where humans step in to handle that segment of the Value Creation pipeline. Something to call attention to here is that each of these modularized segments poses an “optimization problem” — when we step into a job as a ‘salesperson’ we have some complex landscape of problems (how to reach customers, what they care about, understanding what you’re selling, how to coherently and/or persuasively present it, etc) that people can perform better or worse at. Each of these roles represent landscapes of expertise that humans or agents can be good or bad at doing.
When we think about organizations as a modular set of optimization problems we can start to examine what the requirements are to perform each role effectively. Certain roles may be quite hard for an AI to perform well (It’s hard to imagine successful AI salespeople as the role depends so much on connection, trust, and rapport, and I already feel immediately and strongly disappointed by automated sales reps) whereas automated coding is feeling quite possible and advancing rapidly.
There are also circumstances where the bar feels… low, and even slightly better systems could result in better experiences for everyone. These would be things like Customer Service, where many humans are already being treated like robots with fixed scripts, inflexible corporate processes, dehumanizing workplaces, and disappointing results. In cases like this, it seems like well structured, well-operating AI systems can’t come fast enough.
The Value Creation Pipeline
So now we’ve laid out the spirit behind this project — thinking about products as “value creation” in the abstract, breaking down the “Value Creation Pipeline” of a business into modular stages of “Job Titles,” and thinking of each of these stages as optimization problems. For a simple example I broke this pipeline into 8 stages to roughly reflect the development process for new products:
Market Needs and Positional Awareness
Product Ideas that Address Needs and Market Changes
Product Scoping and Feature Selection
Technical Implementation Plan
Prototyping
Content Generation and Marketing Materials
Product Validation and Simulated Market Feedback
Launch Strategy
And then created a simple system where each stage is handled by a different “Agent” who responds to the outputs of agents from earlier in the chain. Since these stages are modular there’s room for expansion, for example connecting output tools or integrations, pulling in outside data, or even contracting humans to handle specific aspects of implementation that robots can’t do. Each time this system is run, it creates a ~ten-page document representing a complete end-to-end product analysis. But what are the outputs like?
Here are ten examples (which you can switch between with the dropdown on the top left):
Note that the system allows for custom instructions at each stage of the pipeline. In the first example I asked for a market analysis of plant-care and greenery industries. Some of these examples I asked for markets for physical goods and products that could be created using 3D printing. In others I asked about changes in the sphere of artisanal or handmade goods. Many of these examples also ask for products that can be developed by a small business with limited capital. In three cases, I ask the system to look at the market for software for Senior Citizens, and each time it has different findings and ideas.
You could ask the GTM agent to specifically consider network effects as promotional strategies, or the designer agent for products that might be laser-cut, etc — This makes the system easily tailorable to assist existing businesses in identifying opportunities for their current market or which suit their existing capital constraints or equipment.
Another northern lights photo :-)
My Thoughts
I think the results are interesting! Something in this direction will be valuable to a certain kind of person or business. I don’t know if the current form of these systems is really a “complete package” that has direct utility, but I am optimistic about derivative projects. One thing that this project highlighted was how often human organizations get “lost in the sauce” of long games of “telephone” that cause us to forget the broader market dynamics or systemic reasons for why we are doing what we’re doing. Creating these clear chains of reasoning that start to fluidly and quickly flow through total systems from end-to-end strikes me as something that AI might be significantly better at than humans, where each agent acts in good-faith and can communicate their findings effectively to the rest of the “organization.”
There are also some natural next steps of taking these broader product plans and tasking each stage to more specific AI or Human agents who actually implement the plans. This is why I think modularization of each agent/developmental stage is particularly important, as it seems likely that these systems will rapidly balloon in complexity as they compete with entire organizations.
A nice element of building in this sphere is that when we extract or structure information from LLMs, any development to those underlying LLM systems will tend to improve the performance of what’s built on top of them, meaning it takes no effort to “ride the wave” of LLM development. LLM weights may already represent (or rapidly become) the largest dataset of “business,“ “market,” or “product” information ever assembled, so while we could enhance this pipeline with additional data inputs (and it’s likely these sanity-checks would be valuable,) unless you have access to private data reflecting your industry, it may be difficult to outpace the current LLM developmental timelines. So while the current system (GPT-4o) might hallucinate or get certain details wrong, this may be a problem that resolves itself as model weights rapidly integrate and compress the other available data sources.
I do see major opportunities for connecting these sorts of generalized systemic frameworks to existing agentic tools, which already seems to be a rich and rapidly developing ecosystem with companies like Cursor, Zapier, and some of the “give an AI a mouse and keyboard” projects.
Something else that’s worth noting is that generating the above 100 pages of analysis cost about .50 cents! Even if turns out that 1/1000 ideas or plans are actually good, 50$ is not a bad price to pay for a business plan that may lead to a successful product line or development. I don’t think anyone really wants to read through 10,000 pages of AI generated business plans, but there are interesting opportunities to map out these model landscapes of market dynamics, user needs, and opportunities through stochastic processes. My current hunch is that complex integrated “Chain of Thought” reasoning (particularly at scale) is something that may lead to interesting emergent insights.
Anyways this was a fun project and it’s likely I’ll continue to work on this! If you’re interested in this stuff or working on something similar, of course feel free to contact me to connect, chat, or collaborate.
