When I was reading, of course it crossed my mind the dilemma between an interesting piece or something else from an actual partner/investor. But then I was like, wow, the author/investor who knows a lot is actually exposing to everyone a lot. So it's not difficult to know that part - that a16z invested - but the value, I hope, is in the conversation that David set in the open. David wants to know more, and the only thing possible to do is to bring up the conversation. Through writing, he increases clarity, and through writing he opens up The Conversation too. My 0.02 here.
Well articulated piece. I very much do hope you're bullishness is well-founded as I use chatGPT as a paid subscriber multiple times per day for personal uses ranging from substitutes for Google search to weighing long-term pros and cons of various AI evolution scenarios (for proprietary work, my firm uses Gemini, Glean, and Claude). However, there is also a -- scary -- case in which OpenAI is AOL to the extent that the company gets stuck as the consumer company and they struggle to get material headway with corporate buyers. I acknowledge that it is great to see so many MAU for OpenAI, but like back in 2000, eyeballs that don't pay you money are worth ... um, nothing.
Interesting thesis. I am not up on the jargon used here but it seems like OpenAI is positioned to be the "Apple" of AI. Of course the article is going to be tendentious in the sense you are a GP in a very successful fund who is invested in OpenAI's success. One has to respect the hustle. I also note the research cited here is mostly from OpenAI itself: https://openai.com/signals/enterprise-data/
Will have to pay attention to the rates of change in your thesis to see if it bears fruit. Thanks for the perspective. Again...interesting.
Everybody talks about AI as if the models are the bottleneck.
They are not.
The bottleneck is the company-wide decision to actually use the thing.
I have spoken to a lot of large corporations over the last three months. Not one of them had a well-thought-out internal policy for how AI gets used inside the building.
Three reasons keep showing up.
1. The people with the power do not do the easy-to-automate work.
The people at the top live in strategy. Strategy is a poor place to discover what AI is good at. The work that is actually ripe for it sits lower: the admin, the reporting, the same five tasks that eat a Tuesday. The people who live inside those Tuesdays do not hold the decision. So the case never reaches the room that can say yes.
2. Picking a system/model feels like selecting your life-partner, saying yes to a marriage.
One vendor over another is not a software purchase. It is data, safety, proprietary work walking out the door. People are right to worry about that. The problem is what the worry produces: delay that looks like prudence. Delay with no policy is still a decision. It just does not get written down.
3. Leadership has no way to see what the workforce can actually do with it.
How do you test who can use this well? Where would a bot take hours off a job, and where would it make the work sloppier? Most companies have no map and no test. Without that, “not yet” is the only call that feels defensible.
The models are already here. The missing piece is a way for a leadership team to see, in their own shop, where this helps and where it does not — and then to write the policy instead of waiting for one to appear.
I like chatGPT and use it daily as a paying customer.
But the pure text interface sucks. We need some sort of UI, like a programmer's IDE that is better organized, makes it easier to find things, makes it easier to connect things.
Working with the current UI is like working with a a decades old green screen terminal!
This is an insightful paper with compelling business case talking points for enterprises lagging innovation to prioritize ethical RAI automation as an actionable objective.
As a Fintech startup nerd and strategic partner to financial services enterprises, even the larger innovative enterprises like PenFed (to compliment your OpenAI reference), that design menber/CX-centric tech emphasize managing their reputational, liability and systemic risks despite having solved for "Getting the security right and the access and data structures and constraints right..." with RAI governance rails and AI education across the organization.
The smaller enterprises, especially if they're not tech savvy, lag scaling AI even with 3rd parties (startups/Fintechs/their legacy core partner, etc) as they're concerned with the additional risks associated with vetting, oversight and assessment processes.
I am not sure I feel the same way about OpenAI, but I loved this post and I think it applies to any software company atm. Most will disappear if they don’t manage to create new user behaviours and win on distribution.
If the rare part is model behavior rather than raw capability, the useful test seems to be whether it can explain uncertainty and revise when evidence changes. What evaluation would convince you this is understanding rather than a polished pattern match?
Nice read! To the end customer, Claude vs ChatGPT might soon be so commoditised because of the Red Queen race that it’s a more ethereal brand play? We don’t know who’s going to be the Coke and who Pepsi but that may very well be how it pans out. Also given Google is already a platform why assume it’s a duopoly? What are the trade offs they are making would be interesting to find out.
At the moment, I don’t think the strongest case for safety is fear. It’s more about creating adoption that is stablized. People have and will use systems they can understand, well enough to use, supervise, challenge, and or recover from consequences of use. Institutions can and will deploy systems with confidence, when responsibility remains clear and understandable. Safety, security, and accountability are not brakes to be attached after the lawnmower is started. They are responsibly, part of the infrastructure that lets useful capability leave the laboratory, safely for adoption at scale.
Wonderful piece. I do miss a bit of a comparison on why exactly you think OpenAI nailed distribution and creation of a new customer, as I think Grok Bot is winning the "manage all your agents" game and Claude code became the default for many marketers.
"Anyone can do nearly anything, but most people aren't, yet." That line describes the inside of nearly every company I coach right now. The AI First CEOs pulling ahead didn't buy better tools than their peers. They got their own people to actually start from a Day One approach and widened the risk tolerance band to allow for go big thinking and action.
It’s nice to read something positive written about OpenAI for once, a truly visionary and innovative company (even if they have skimped on their safety and alignment budget). I think the switching costs are underestimated. Memory and personalization is increasing stickiness on the consumer side, and for enterprises you can’t just swap out model families all willy nilly if you care at all about responsible AI governance.
Says the investment firm that poured a lot of money into OpenAI.
yea but this is simply the world we live in ;-) got to accept that in everything we see online
When I was reading, of course it crossed my mind the dilemma between an interesting piece or something else from an actual partner/investor. But then I was like, wow, the author/investor who knows a lot is actually exposing to everyone a lot. So it's not difficult to know that part - that a16z invested - but the value, I hope, is in the conversation that David set in the open. David wants to know more, and the only thing possible to do is to bring up the conversation. Through writing, he increases clarity, and through writing he opens up The Conversation too. My 0.02 here.
Well articulated piece. I very much do hope you're bullishness is well-founded as I use chatGPT as a paid subscriber multiple times per day for personal uses ranging from substitutes for Google search to weighing long-term pros and cons of various AI evolution scenarios (for proprietary work, my firm uses Gemini, Glean, and Claude). However, there is also a -- scary -- case in which OpenAI is AOL to the extent that the company gets stuck as the consumer company and they struggle to get material headway with corporate buyers. I acknowledge that it is great to see so many MAU for OpenAI, but like back in 2000, eyeballs that don't pay you money are worth ... um, nothing.
Interesting thesis. I am not up on the jargon used here but it seems like OpenAI is positioned to be the "Apple" of AI. Of course the article is going to be tendentious in the sense you are a GP in a very successful fund who is invested in OpenAI's success. One has to respect the hustle. I also note the research cited here is mostly from OpenAI itself: https://openai.com/signals/enterprise-data/
Will have to pay attention to the rates of change in your thesis to see if it bears fruit. Thanks for the perspective. Again...interesting.
Everybody talks about AI as if the models are the bottleneck.
They are not.
The bottleneck is the company-wide decision to actually use the thing.
I have spoken to a lot of large corporations over the last three months. Not one of them had a well-thought-out internal policy for how AI gets used inside the building.
Three reasons keep showing up.
1. The people with the power do not do the easy-to-automate work.
The people at the top live in strategy. Strategy is a poor place to discover what AI is good at. The work that is actually ripe for it sits lower: the admin, the reporting, the same five tasks that eat a Tuesday. The people who live inside those Tuesdays do not hold the decision. So the case never reaches the room that can say yes.
2. Picking a system/model feels like selecting your life-partner, saying yes to a marriage.
One vendor over another is not a software purchase. It is data, safety, proprietary work walking out the door. People are right to worry about that. The problem is what the worry produces: delay that looks like prudence. Delay with no policy is still a decision. It just does not get written down.
3. Leadership has no way to see what the workforce can actually do with it.
How do you test who can use this well? Where would a bot take hours off a job, and where would it make the work sloppier? Most companies have no map and no test. Without that, “not yet” is the only call that feels defensible.
The models are already here. The missing piece is a way for a leadership team to see, in their own shop, where this helps and where it does not — and then to write the policy instead of waiting for one to appear.
I like chatGPT and use it daily as a paying customer.
But the pure text interface sucks. We need some sort of UI, like a programmer's IDE that is better organized, makes it easier to find things, makes it easier to connect things.
Working with the current UI is like working with a a decades old green screen terminal!
This is an insightful paper with compelling business case talking points for enterprises lagging innovation to prioritize ethical RAI automation as an actionable objective.
As a Fintech startup nerd and strategic partner to financial services enterprises, even the larger innovative enterprises like PenFed (to compliment your OpenAI reference), that design menber/CX-centric tech emphasize managing their reputational, liability and systemic risks despite having solved for "Getting the security right and the access and data structures and constraints right..." with RAI governance rails and AI education across the organization.
The smaller enterprises, especially if they're not tech savvy, lag scaling AI even with 3rd parties (startups/Fintechs/their legacy core partner, etc) as they're concerned with the additional risks associated with vetting, oversight and assessment processes.
I am not sure I feel the same way about OpenAI, but I loved this post and I think it applies to any software company atm. Most will disappear if they don’t manage to create new user behaviours and win on distribution.
If the rare part is model behavior rather than raw capability, the useful test seems to be whether it can explain uncertainty and revise when evidence changes. What evaluation would convince you this is understanding rather than a polished pattern match?
Nice read! To the end customer, Claude vs ChatGPT might soon be so commoditised because of the Red Queen race that it’s a more ethereal brand play? We don’t know who’s going to be the Coke and who Pepsi but that may very well be how it pans out. Also given Google is already a platform why assume it’s a duopoly? What are the trade offs they are making would be interesting to find out.
totally disagree the winner will ba an open system like Co-pilot that includes Open AI on its platform but pricing will be cutthroat...care to bet?
At the moment, I don’t think the strongest case for safety is fear. It’s more about creating adoption that is stablized. People have and will use systems they can understand, well enough to use, supervise, challenge, and or recover from consequences of use. Institutions can and will deploy systems with confidence, when responsibility remains clear and understandable. Safety, security, and accountability are not brakes to be attached after the lawnmower is started. They are responsibly, part of the infrastructure that lets useful capability leave the laboratory, safely for adoption at scale.
Wonderful piece. I do miss a bit of a comparison on why exactly you think OpenAI nailed distribution and creation of a new customer, as I think Grok Bot is winning the "manage all your agents" game and Claude code became the default for many marketers.
"Anyone can do nearly anything, but most people aren't, yet." That line describes the inside of nearly every company I coach right now. The AI First CEOs pulling ahead didn't buy better tools than their peers. They got their own people to actually start from a Day One approach and widened the risk tolerance band to allow for go big thinking and action.
It’s nice to read something positive written about OpenAI for once, a truly visionary and innovative company (even if they have skimped on their safety and alignment budget). I think the switching costs are underestimated. Memory and personalization is increasing stickiness on the consumer side, and for enterprises you can’t just swap out model families all willy nilly if you care at all about responsible AI governance.
Steve Jobes did the same to computers, and Andreesen did the same to the Internet