AI Product Strategy

"You need to be deliberately starting in places where there is minimal impact and more human control so that you have a good grip of what are the current capabilities and what can I do with them and then slowly lean into the more agency and lesser control."

"LLMs allow writing shitty software to be significantly cheaper, not necessarily good software, but good enough in certain contexts. And also it means that there's certain software now that isn't plain old computing that can be run cheaply. It's relatively expensive marginal cost."

"Because these models are so effective at this point, you want to start to tune them to certain types of outcomes. All of a sudden, these are these living organisms that just get better with the more interactions that happen. I think this is the new IP of every single company products that think and live and learn."

"If you talk to the users who understand what they want or they don't want, look into the feedback, then you can actually improve the application way, way, way more. Why do you need to keep up to date with the latest AI news?"

"If you could choose the perfect model behavior, which model would you want? Do you want a model that says, "You're absolutely right. There are definitely 20 more ways to improve this email," and it continues for 50 more iterations or do you want a model that's optimizing for your time and productivity and just says, "No. You need to stop. Your email's great. Just send it and move on"?"

"These things take 6 to 12 months to get them truly robust enough where an important process can be automated. Like with any of these major tech revolutions, headlines tell one story and then on the ground, laying broadband means you need to dig up every single road in America to lay it."

"We're not going to launch some of these varied verticalized products. We're not going to launch an AI sales agent. That's just not what we're building towards. And companies who are and have some domain specific knowledge and they're really excited about that problem space, they can go into that and leverage our models and end up continuing to be on the cutting edge without having to do all that R&D effort themselves."

"There is something called the shiny object trap, and I'm always telling people, "Hey, don't do AI for the sake of doing AI." Make sure there is a problem there. Make sure there is a pain point that needs to be solved in a smart way."

"We definitely didn't expect to be doing any of our own model development. And at this point, every magic moment in Cursor involves a custom model in some way."

"I think in the AI space, we're trying to hear from customers, what do you wish Slack could do if it had these new superpowers? Let's incubate a couple teams or prototype, give them space to run and pilot and then get something to launch that's amazing. Blows people away. That's the formula that we've seen."

"But if there were one technique that I could recommend people, it is few-shot prompting, which is just giving the AI examples of what you want it to do. So maybe you wanted to write an email in your style, but it's probably a bit difficult to describe your writing style to an AI. So instead, you can just take a couple of your previous emails, paste them into the model, and then say, 'Hey, write me another email.'"

"Studies have shown that using bad prompts can get you down to 0% on a problem, and good prompts can boost you up to 90%. People will always be saying, 'It's dead,' or, 'It's going to be dead with the next model version,' but then it comes out and it's not."

"My main advice there is choose a common format. So XML, great. If it's, I don't know, I don't know, question, colon, and then you input the question, then answer, colon, and you input the output, that's great too. It's a more research-y approach."

"I actually think that the role of AI going forward will be to have applications increasingly meet us where we are. To this day, we've always had to generalize onboarding experiences for the most part for everyone. And I'm really excited about the day when kind of products meet us where we are based on what type of user we are."

"I think it's a 50/50 of both. I think the capabilities obviously have improved a ton and we've seen these get better and get measurably better. But I think the other side of it is everything to do with yeah, really the product interface and the tools and so on."

"Well, you've got this synthesis machine, which is this LLM thing that's going to help you do synthesis, but if it hasn't got all that data to do synthesis on top of, it's got nothing. And so that means that LLMs can only be as good as the data they are given and how recent that data is."

"The field and the models themselves are just changing so, so quickly. They tend to disrupt themselves. The models will eat your scaffolding for breakfast."

"To enable a one person billion dollar startup, there might be a hundred other small startups building bespoke software. So I think we might actually enter into a golden age of B2B SaaS."

"So it wasn't like, 'Oh, we have this cool technology, what can we do with it?' It was like, 'Let go of what you've built. Go back to the objectives you were trying to solve and now with this technology, how can you do that objective better?'"

"So now our hypotheses are just wrong. So at this point then, most of the value is probably not going to accrue at purely the, at least this is our belief, at the infrastructure layer. It's going to accrue somewhere else. Where is the layer that you can actually differentiate on? And we believe the application layer is a very, very deep layer to go out and differentiate on, right?"