Rasmus Ähtävä

where is the value when code is free

i have this mental model from what the current LLMs can do and go maximally, meaning if they get better and better, how far can they go? they predict the next token -> next best token comes from training data -> training data is midwit -> if you don't know exactly what to ask, model will give you midwit -> no differentiating business value since all businesses get this midwit program -> all software that is known will get commoditized and cheap -> what differentiates a business then is what they are uniquely good at, their experiences and domain knowledge -> how to build software that multiplies their unique advantage -> midwit software is not enough when code is infinite leverage and marginally better program produces half the better product -> best product wins in teh market -> to get the best product, programmer has to understand code deeply and business case deeply in order to come up with a unique software multiplier for each business -> this is artistry and AI midwit coding won't be able to product the best software due to training data and customers don't know what they need and would just ask what they think they need which is not the same since they don't think in code. -> someone could make argument that coding agents can be trained to unique business cases -> really? unique experience of unique business throughout years? no way. -> need human to understand code deeply in order to best get this marginally best software for business case -> make business win -> huge value in there