Sprig hosted an AI + Product Community event at our headquarters in San Francisco, hosting local product managers to learn from and connect with AI experts Kevin Mandich and Jonas Lavoie.
Mandich, Head of AI and ML at Sprig, is an engineering PhD turned AI expert with over nine years of industry experience building state-of-the-art machine learning applications. Lavoie led product development efforts at Shopify, Elastic, and most recently Notion, where he spearheaded the integration of AI and ML into product roadmaps.
The evening was an amazing opportunity to learn from these industry leaders and hear about some of the benefits, exciting opportunities and cautious optimism around AI.
“We're at this moment where we're trying to figure out the boundaries of it, or what will be useful long term. I think right now, some people are pushing the envelope in ways that are super interesting. They're not super finished or very polished, but we're figuring this out,” said Lavoie.
“There's going to be the next generation of applications that are AI first, and that is really where the real power will be.”
Here’s a recap of the Q&A with our Head of Partnerships Lilly Fast:
What teams do you admire right now that are taking advantage of AI and really innovating on this front?
Lavoie: Adobe just came out with Firefly, and is starting to really leapfrog into the actual practical uses of image generation within their products, which is already in a place where you're familiar. It is allowing you to select parts of an image and turn that part of the image into something else. It's like you're actually describing what you want, but not from a blank canvas. And it's things like this, that I think over time are going to be incredibly successful.
Mandich: I think the folks that have the platform companies are going to reap the most benefits in the short term. They have the users and they have the ability to really deploy this technology in a meaningful way. I do believe that, because the playing field is becoming more and more level, that advantage may shift over time and become lesser, or it may just be easier for new companies to quickly build a platform that allows them to do the same things.
You guys have had incredible experiences. Can you share how you incorporated and prioritized AI into your roadmaps?
Lavoie: Elastic is one of the world's most popular data store, and it is wonderful at dealing with structured data, semi-structured data, and unstructured data – with a little bit of help.
There, the exploration into AI and machine learning was really born out of actual need with making sense of the unstructured information and meeting the increasing consumer expectations with regards to search. We had to go and meet professors and academics and try to capture the knowledge and bring it into practical environments. It required some amount of technological knowledge - you have to be tech-savvy to operate in that.
Fast forward to Notion, the story there was very different. Our founders' visionary thinking proved that this was something they'd been waiting for. They started pushing pretty hard internally for generative AI well before the hype cycle because Notion is full of data that can be taken, then modified into a way that could augment not only the output, the individuals working in that workspace, but also all the collaboration around that.
AI allows people to augment the overall quality standard and elevate the actionability of the content that is found in a workspace. Very naturally, generative AI became something of great use because Notion is a context- and content-heavy environment.
How do you see this AI evolution impacting product maintenance?
Lavoie: There will be added “burden." You’re now introducing a non-deterministic layer of something across your product that has the impact to change every single workflow that exists currently in your application. And every net-new implementation will most likely change the behavior of the same thing in the future. Over time, you will probably have to retrain your own users on how to use the thing. It might be one of those things where AI is so good that it ends up kind of solving its own change management along the way. I think that's TBD.
You're putting a black box in your workflow. And sometimes that black box’s behavior changes or produces slight variations in output.. As your product evolves, your input is also going to change, your distribution of data is going to change. Any ML product has always required a good amount of maintenance. And now we're adding a little extra complexity, but it is well worth it.
Mandich: I think that's where we talked briefly about open source here. I think what you're referring to here is the foundational layers or foundational models, and I think there is going to be at some point a rotation towards having both the foundational layer in some of our controls as you as you put it, but those are going to show as the technology commodifies - which it is already doing so rapidly.
If you are an open AI user here and a moderation endpoint go down, you know that you're now exposing yourself to a lot of bad stuff that could happen. You are at the mercy of someone else's technology.
And if they decide GPT 4 needs an update, and you’ve optimized your workflows for the 4 version and then they push out 4.1 – then it's a scramble to go and try to engineer your way into something that is more of what you wanted in the first place. So it's a very good point that you're making, they're being at the mercy of another system.
For a PM who's interested in learning more about AI, what are some tips or guidance that you would recommend for starting to immerse yourself?
Mandich: Sign up for newsletters, stay up to date, I used to consider myself to be somebody that kept up with the forefront in the state of the arts. But now I'm even struggling because things change day to day. But if you have an idea of what's possible with this technology, then you’re ahead of 90% of the game right there.