Notes and links - 8th February
"Where we are with agentic AI feels a lot like the cloud computing boom of 20 years ago. Marketed on claims of efficiency and flexibility, its big innovation was to remove friction between a company’s operating expenses and revenue lines of Amazon, Microsoft and Google.
Whereas project purchasing decisions once had to go through Procurement, virtual machines can now be spun up in moments at a cost of a few dollars an hour. It’s impossible to guess how much rented hardware is left idling, never to be noticed again within the IT department’s Rube Goldberg machine, but experience tells us it’s likely to be a lot. Better waste management is one plausible explanation for why companies have been abandoning the cloud and bringing servers back on premises.
Office management by agentic AI promises to be all of this but much, much worse. For example, this chap slop-coded a reminder to buy milk and found it had a running cost of more than $13,000 a year. Extrapolate that level of misspend across dozens of workflow bots running on millions of poorly monitored corporate networks, with everyone being charged by one non-itemised monthly bill, and those AI return-on-investment projections start to make a little more sense."
The race to the bottom continues
