Nodding furiously over here. Today we were discussing all the myriad meanings that people in different roles/context hold in relation to phrases like 'speed to value'. When our measurement of speed doesn't take into account customer pain, rework, all sorts of things - it's a bad measure. We compensated for a time by focusing on predictability over pure velocity and that helped somewhat. Thanks for that, Toyota. The meaning of value is a problem we haven't really cracked the code on, though. It can be so many things... Talk to us about value, quality sometime? I like the way you skew your lenses.
Yes, the meaning of value is very context-sensitive and hard to determine sometimes, but... brining value into the conversation, even when its definition is not fully agreed upon, is always better than talk about cost absent any value discussion.
It's a little bit like saying "the cost of cars today" is much higher than "the cost of cars 50 years ago". Yes, even after adjusting for inflation, but the car you get today has AC/heat, power everything, airbags, ABS and a reliable GPS system. There was no amount of money 50 years ago that could buy you all those capabilities.
In other words, cars today allow you to do things that you couldn't even do 50 years ago.
Same for AI - it's not just "augmenting" human capabilities, it's allowing humans to do things that would have been basically impossible in any practical sense. Like me doing 2 minutes of inflation research to validate the car comparison above. (I was going to use computers, because the capability gap was more intuitive to me, but my AI assistant was able to correct me very quickly and show me computer prices have actually fallen considerably)
I think an interesting 'benchmark' equivalent topic is "how does one gage productivity with AI".
There are so many different levels of execution capability at this stage. "Was I effective in my use of AI there past few days vs. how others would have done it" is a question I constantly ask myself. And that links back to your cost question!
I think people either don't baseline their current productivity or lose track of their baseline quickly once they evolve from it.
It's also a function of what "effective" means in this context. Versus what baseline? Not having AI at all - or using AI based on how you used it 12 months ago?
One thing that's really helped our teams is coming up with a role-specific maturity scale. This varies a lot based on your activities, but everyone basically starts with "single chat window, with one-time prompts" - maturity level 1. Then re-usable prompts. Then single agents. Then multiple agents. Then MCPs. Etc... We can argue about the order and the exact definition of each maturity level, but tracking that evolution, and being intentional about progressing through it, is a good way to track "efficiency".
Nodding furiously over here. Today we were discussing all the myriad meanings that people in different roles/context hold in relation to phrases like 'speed to value'. When our measurement of speed doesn't take into account customer pain, rework, all sorts of things - it's a bad measure. We compensated for a time by focusing on predictability over pure velocity and that helped somewhat. Thanks for that, Toyota. The meaning of value is a problem we haven't really cracked the code on, though. It can be so many things... Talk to us about value, quality sometime? I like the way you skew your lenses.
Yes, the meaning of value is very context-sensitive and hard to determine sometimes, but... brining value into the conversation, even when its definition is not fully agreed upon, is always better than talk about cost absent any value discussion.
I am well out of my depth reading this. Instead I will share what i think is a relevant article. Is the cost per token being subsidized? Is there a higher real cost being hidden? https://www.entrepreneur.com/business-news/nvidia-vp-says-it-costs-more-to-use-ai-than-to-hire-humans
It's a little bit like saying "the cost of cars today" is much higher than "the cost of cars 50 years ago". Yes, even after adjusting for inflation, but the car you get today has AC/heat, power everything, airbags, ABS and a reliable GPS system. There was no amount of money 50 years ago that could buy you all those capabilities.
In other words, cars today allow you to do things that you couldn't even do 50 years ago.
Same for AI - it's not just "augmenting" human capabilities, it's allowing humans to do things that would have been basically impossible in any practical sense. Like me doing 2 minutes of inflation research to validate the car comparison above. (I was going to use computers, because the capability gap was more intuitive to me, but my AI assistant was able to correct me very quickly and show me computer prices have actually fallen considerably)
I think an interesting 'benchmark' equivalent topic is "how does one gage productivity with AI".
There are so many different levels of execution capability at this stage. "Was I effective in my use of AI there past few days vs. how others would have done it" is a question I constantly ask myself. And that links back to your cost question!
I think people either don't baseline their current productivity or lose track of their baseline quickly once they evolve from it.
It's also a function of what "effective" means in this context. Versus what baseline? Not having AI at all - or using AI based on how you used it 12 months ago?
One thing that's really helped our teams is coming up with a role-specific maturity scale. This varies a lot based on your activities, but everyone basically starts with "single chat window, with one-time prompts" - maturity level 1. Then re-usable prompts. Then single agents. Then multiple agents. Then MCPs. Etc... We can argue about the order and the exact definition of each maturity level, but tracking that evolution, and being intentional about progressing through it, is a good way to track "efficiency".