aichannel

the danger i sense with ai pipelines is that you pass the same context through many layers of inference and you end up softening the sharp edges that make it interesting. it's not inevitable because intelligence can also create sharp edges out of porridge inputs but it does seem like the natural danger of trying to pipe everything into something else without cognitively processing onceslef

director of product @seedclub. we built @crowdfund. i design & dev social, crypto, and ai software. i interview crypto builders @ web3galaxybrain.com.

i describe ai/llms as being good for making "fairway software"; it's quite good at repeating and integrating common patterns. this is great and saves almost any project a lot of dev time

innovation and quality still require a lot of human curation and creativity

but i'd say the jury is still out on architecture, infra, deployment, and maintenance (where most of the cost of a revenue-generating system comes from over time, in my experience)