You usually emphasise on 3x improvements in productivity gains to drive real adoption.
Transformer-based LLMs / agents can potentially do tasks/part of a workflow "autonomously".
How does this trend make you think about the "Productivity gains"?
1. **In absolute sense** , Is it still 3x?
2. This trend may potentially free up "significant time" of the knowledge workers. This means, the workers may use this time on other outcomes. In this case, is it also useful think beyond "productivity gains with respect to that workflow" and see how agent adoption for a particular workflow **relatively** influences the "quality of work/ outcome" of the workers in the "other outcomes" which they focus now?
Agents are going to be big when they become capable. They will deliver over 10x gains - which will be felt enough that people will pay for them. But, if they are not capable enough, they will claw back some of that time with customers cleaning up after them and tweaking (the bad representation problem). So they have to be competent to deliver on the promise.
Faster workflows always free up time for either leisure or other more complex higher-order workflows that were too hard to attempt economically. So this will be no different.
As a general principle to make sure you’re offering sufficient value, I agree but the precision of 3x seems arbitrary. I’d instead refer to the research around risk aversion which pegs it at 10x.
Yes it’s arbitrary for the most part. Most of the research is not precise anyhow. I usually say 3-10x. The most important thing is to illustrate that ‘doubling’ is not enough.
In addition I think the threshold is different for different problems. At work, I’d hazard that the magnification can be smaller - people are really focused on productivity @work. For entertainment and life, I’d say the threshold is much higher than 3x for sure.
Hey Oji, Bringing this thread up.
You usually emphasise on 3x improvements in productivity gains to drive real adoption.
Transformer-based LLMs / agents can potentially do tasks/part of a workflow "autonomously".
How does this trend make you think about the "Productivity gains"?
1. **In absolute sense** , Is it still 3x?
2. This trend may potentially free up "significant time" of the knowledge workers. This means, the workers may use this time on other outcomes. In this case, is it also useful think beyond "productivity gains with respect to that workflow" and see how agent adoption for a particular workflow **relatively** influences the "quality of work/ outcome" of the workers in the "other outcomes" which they focus now?
Agents are going to be big when they become capable. They will deliver over 10x gains - which will be felt enough that people will pay for them. But, if they are not capable enough, they will claw back some of that time with customers cleaning up after them and tweaking (the bad representation problem). So they have to be competent to deliver on the promise.
Faster workflows always free up time for either leisure or other more complex higher-order workflows that were too hard to attempt economically. So this will be no different.
Great read!
Currently trying to decide on a beachhead for www.quabbly.com
For a product with different use cases for different industries, which do you think is best at early stage?
1. Picking a specific use case for a specific industry
2. Picking a specific use case for multiple industries
1, then 2; when you see some PMF.
As a general principle to make sure you’re offering sufficient value, I agree but the precision of 3x seems arbitrary. I’d instead refer to the research around risk aversion which pegs it at 10x.
Yes it’s arbitrary for the most part. Most of the research is not precise anyhow. I usually say 3-10x. The most important thing is to illustrate that ‘doubling’ is not enough.
In addition I think the threshold is different for different problems. At work, I’d hazard that the magnification can be smaller - people are really focused on productivity @work. For entertainment and life, I’d say the threshold is much higher than 3x for sure.