Scaffolding and Intelligence Entrepreneurship
What’s the difference between “businessman” and “entrepreneur”? Or between “starting a business” and “doing a startup”? I think these are often cultural/social distinctions. If you’re inspired by the silicon valley style of software and tech business, you probably like the term “entrepreneur”. If you are more traditional, then the word “business” is fine.
There is a simple economic definition of the word “entrepreneur” that I find illuminating. It comes from Jean-Baptiste Say, who was one of the first to theorise the concept. My interpretation of it is as follows:
Economic output refers to goods and services.
Factors of production are the “inputs” that are combined to make economic outputs.
The three factors of production are:
Land (incl. raw materials)
Labour
Capital (i.e. tools, machinery, etc.)
The entrepreneur is the one who combines the factors of production to create economic outputs.
For the recent past, although entrepreneurship has been growing in popularity, the main strategy of talented youth has been to specialise in some domain, or generally be smart, and make the bulk of your money working with that brainpower. The focus has been on being valuable labour, and not on doing entrepreneurship.
AI is becoming a very powerful tool (capital). In the short term, it is boosting the productivity of skilled labour, making the old strategy even more successful. My friends who are “labour” in tech and finance are enjoying the productivity boost of AI tools.
However, soon, AI will become so intelligent that in many cases instead of being capital that boosts the productivity of labour, it will start to act as labour itself. This transition of AI from capital to labour is marked by the current transition of AI from “assistants” to “agents”.
We are presented with an opportunity to “rise up the value chain”. In some industries this is an existential matter because your job is at risk, in others it is a megatrend that you can choose to join.
We have the opportunity to start applying intelligence rather than being intelligence.
Key to applying intelligence is to view yourself not as the person doing the work, but the person orchestrating intelligence so that the work gets done. This was a lot harder in the past: you had to work with human people, who are expensive and require some charisma to motivate. Today you can apply intelligence by orchestrating AI agents from your bedroom.
Scaffolding is the activity of creating processes that use LLMs in such a way that gets them to act more intelligently than the LLM could all by itself.
A simple example of scaffolding is called RAG: Retrieval Augmented Generation. When ChatGPT first launched, it would hallucinate a lot. This is not surprising as it was brave enough to answer any question about anything even though it only had about a terabyte of total memory. There were unsurprisingly some things it didn’t know, so it just made them up. RAG fixed this problem and made the GPT model act “more intelligently”. In RAG, you first perform a websearch based on the user’s prompt and then you let ChatGPT answer the question with all the search results in its context window. This combines the better accuracy of document search with ChatGPT’s ability to navigate language reasoning tasks.
Other successful scaffolding ideas are:
- chain of thought prompting, where the model reasons by “thinking out loud” before responding to the prompt
- deep research, which is a process that the model follows to do a thorough literature review
- LLM tool calls, where the agent has access to APIs and tools like calculators or programming environments to help it think
Scaffolding is like building a workflow for your employees to follow to ensure that standards are maintained. In fact, organisations can also be seen in this way: creating processes that lead to better outputs than what any individual person in the organisation could have created by themselves.
Most realistic businesses will probably scaffold AI into human organisations.
Beyond scaffolding to make more intelligent AI agents, one will also need to find a customer, and embed this intelligence into their personal/business processes to deliver some outcomes for them.
I think we are in a golden era of intelligence entrepreneurship - where intelligence is becoming free, and the goal is to find ways to orchestrate and apply this intelligence.