I keep seeing that AI has a cost problem: SaaS is expensive to build but cheap to scale; LLMs start cheap but costs go up as you grow and use them more.
I’m not convinced by this argument.
Say I’ve got to come up with an economic forecast.
There are a few ways I could do this.
Subscribe to a Bloomberg terminal for 30k that has all the data and analysis I need.
Hire someone to read and summarise material for maybe 100-400 a day.
Or run some PDFs through an LLM as a starting point for 20p.
You still need quality control so the real cost of an LLM is compute + review time costs.
But LLM + time is cheaper and possibly higher quality than time alone if you use the tools well.
What this means for me is that when I come across a business problem my first question is:
“Can an LLM help me do this better or faster?”
Before reaching for more expensive options.
Because the economics of how we deliver knowledge work are changing – and we’ve got to figure out where human judgement and input adds the most value.
Handcrafted insight 