Building Out Loud: AI native Startup Journey | Faith Forster and Randy Silver

speaker-0: Next up, have a really innovative format, as you see from the activity that's happening around me. We're going do a podcast as a talk. for everyone who standing up there, ⁓ and also, like, if see all of these seats open here towards the front, just come on down, like move closer, because you're going to be in this particular podcast. Format here.

⁓ speaker-1: Live. So the full. speaker-0: is that about 10 minutes into it in this podcast that's co-hosted by Faith Forster and Randy Silver, 10 minutes into it, they're going to start taking questions from the And free to enter into Slido both comments and they'll start to integrate that into their And so let me take moment to introduce them and give you a little about background. speaker-1: the audience.

questions. speaker-0: But again, a little closer, because we're going to make lots of noise and participate in this podcast. So while you're coming closer, here, I'm going to introduce, let me start. speaker-1: with faith.

is... speaker-0: the Chief Product Officer of Discovery. she's built products in B2B, SAS, in FinTech. And been part of several successful acquisitions.

And Randy ⁓ is the of of Owls. And he has experience building high-performance teams, having done that at Amazon, Sainsbury, and HSBC. And he's the author of Do We Do Now? Together, ⁓ their co-hosts of the podcast, Building Out Loud, AI native startup journey.

I'm excited for this recording. Let's make lots of noise for Faith and Randy. speaker-1: And I'm live record. Thank you.

is weird. speaker-0: Normally sitting in my garden office while we do this. not You're your own office, yes. speaker-1: Okay, so let's start the way we always start.

So welcome back to another episode of Building Out Loud with me, Randy Silver, where every week we check in with Faith to see how everything's going as she builds out Discovery. this one's a little bit different. We're not doing it in our offices. We're doing this live in Lisbon and productized.

speaker-0: Faith Forster. speaker-1: So normally we would start with checking in and I would ask you what's been going on in the last week, but for the audience here who may not have listened to every episode, let's do a little bit of background. So what is Discovery? How did you get started on this journey?

Let's do the short potted history. speaker-0: So I was a Chief Product Officer at a legal tech company and this sort of late last year. I was trying to use agents to gather all the information we needed to help inform what we focus on in 2026. And I found it really hard.

There was just, as product leaders, there was so many different sources of information we need to refer to. And so I ended up... I ended up having this like beast of a word doc with different tabs where I sort of dumped all this information that the agents had given me. But I still had to find time between to go through it and work out what's important, what does this mean, what do do about it.

And I was like, ugh. And I had already reset our product strategy around building agentic workflows. And I had had a lot of pushback from the engineers on it, actually. And so when I left that organization, I really wanted to get hands on with AI coding platforms.

Because I think as leaders, it's really hard to understand properly the possibilities and the pitfalls of AI coding if you haven't actually done it yourself. And so I was kind of looking for an excuse to do that. really stuck with me and I thought there's it's actually like I've talked a lot about the fact that we an AI world we need to go back to first principles to the problem we're trying to solve for our customers and rethink how we solve that problem and I kind of thought this was a great opportunity to kind of go back to first principles.

I've rolled out so many different product management tools and different organisations I've led. None of them worked particularly well. did certain bits well, I of, I just thought there was an opportunity here to go back and rethink that whole problem using AI. speaker-1: So let's go into the problem that you were trying to solve.

was the thing that you wanted to dig into at this? speaker-0: So when I started, I was looking to build a series of agents that would gather that information and give you the so what for product. As I worked through problem, and I remember saying this to you on one of the episodes, it went so further than I thought possible the start. once you actually start to understand the technology and how to use it.

I had these agents that were gathering all that information, ⁓ is tasks, that ⁓ give each agent a particular type of information to go away and get and come back. But actually where it got much more interesting was then when I started building agents on top of that that did the analysis and then on top of that that did the recommendations. we're able to, and I remember showing you under the screens, like competitor intelligence, we gathered detailed profiles about all of your competitors.

But then when you have the so what around, like what does their pricing model say about their strategy? What does their integration say about their partnerships? We do sort of strategic assessment of that competitors and give it to suggest a threat level and we use that to prioritise feedback we gather on your competitors from the web. But then you can add on top of that again and use that to understand your competitive landscape and what's moving and that means for you strategically and the risks it creates and recommendations on that terms of what you then should be thinking about with your product.

And I think that's where it starts getting a lot more interesting when you add that level ⁓ insight with your insights and feedback, with your strategy, with your business goals. out your road map and pull it in from planning tools you're using and suggest road map items. ⁓ out where you're not doing the most valuable things and where you should focus your attention instead. There's just so much more value we can add.

speaker-1: So how does this live up to the promise and hype? We just saw a slide that talked about lovable getting to 100 million ARR in 20 minutes or something like that. been doing this for six months. It's been smooth sailing, right?

Everything went perfectly and you're on the road to millions and millions. Yeah. So what's actually happened? speaker-0: show.

Yeah, it's been a rocky journey, which was we wanted to do this podcast, because there was just so much, from the very start, was so much ⁓ I learning, and I knew everyone else needed to learn this. And so as I reached out to you, I'm like, let's do this out loud. Gosh, there's some ups and downs. It took me weeks to get the agents to actually do what I was trying to get them to do.

Opus 4.6 was an absolute game changer. Even like, so I built it in Replet. Replet hosted in the US.

a lot I've worked with 96 product leaders to get to this through Alpha and Beta, and of those, not all, but most of them are in the UK ⁓ or EU. And so we it hosted the EU, and that's created a whole bunch of challenges just trying to get two instances set up, a common code making sure that payments worked across both instances properly, the and the email notifications and things knew where to send people back. There's just been, yeah, been a lot of complexity. I did an incident a couple of weeks ago where, is that what you're fishing for?

Yeah. ⁓ speaker-1: One of those things, yeah. speaker-0: Just before the long weekend, I made some changes to one of the competitor intelligence agents and it started throwing an error, but I had run out the door by this point for my son's sports day. And it kept retrying and retried to its job every three seconds, every competitor profile, across every product, across every organisation.

So over 100,000 failed LLM calls later, it had drained my bank account. it hadn't been a long weekend, I would have realised much quicker. But unfortunately, I was very, very busy and yeah, hurt. Yeah.

speaker-1: you did learn some lessons from that one and yeah. ⁓ step away from that incident. It's not like it's all been a straight line. There has been discovery about discovery along the way.

Tell about the changes in and how you learned what the product actually was over time. How has it changed? speaker-0: I did, yeah. Yeah, gosh, a lot.

I feel like that's an evolution in itself. I said, it started out by just sort of doing the gathering of the information. We have so much information we use as inputs, and lot of it not very well, to help us think through. what we should focus on in pricing.

Like, you know, we're in lost reasons in the CRM, customer feedback is always scattered over lots of places. very, very poor generally at understanding our competitors. Very poor also at connecting the dots back to the financials and actually understand what drives ⁓ and margin within our product. And I thought there's a lot of value in just gathering that information.

But actually when I spoke to a bunch product leaders, they were like, yeah, kind of interesting. might save bit of time, but actually it's not really valuable enough. so that's where I started looking at sort of what else can we do. And I think one of the things that I got really excited about, so I was with AI, but I was still thinking about the product like a traditional SAS platform with boxes everywhere.

And when I redesigned it to look more like Markdown files, and that opened up a whole new set of possibilities. And one of the ones I got really excited about was the growth strategy, because it's just too complex to put into boxes. actually, when you have... a chat assistant who can gather understanding first and then create the interface accordingly, it completely changes the game on that.

And so we a general strategy assistant who will ask you a bunch of sort of common or basic questions about your growth strategy and how you're thinking about driving the business. We then have 20 specialist strategy assistants will then do a deep dive on things like pricing or new market entry or your optimization funnel or even raising money, funding. And so can do a deep dive with any of those specialist agents ⁓ on specific areas you're at for your business. And then we also have a partner who you can just have a really open conversation about anything you want with all the context that we have in the platform.

But one of the great things about this, think is quite cool, is you can chat to them about all the reasons why your strategy might fail. And then it uses that to update your strategy and work out what does this mean, what do we do differently. And I thought that is just so cool to have that partnership, ⁓ I But have it in a way that's you can share it with the whole team. It becomes the organisational context that everyone else uses to prioritise, to work out what they should be focusing on.

So it's not you sitting with Claude over in the corner and then coming out and saying, is it, without any understanding the team. It actually becomes a living, breathing part of the way that your organisation works, which think is super powerful. speaker-1: I want to ask you about the pronouns you were using in that last question. Because you kept referring to we and the team and the organization.

But discovery is you. artificial intelligence enabled things. speaker-0: Yeah. speaker-1: You've done a in Scale Up before, but you had a team around you, actual people.

An actual And this time mostly you in the garden office. So have you started of agents as personalities, as people in some ways? How were you, let's go back into we a little How were you actually feeling about this? ⁓ speaker-0: Yeah, is interesting.

I do talk about they and we and my team of agents. I haven't entirely done this on my own. I have had an engineer helping me. So he's helped set up the EU instance and hardened the data security and the other really, really critical things.

I have an amazing VP of product marketing that I worked with previously. He's been helping me. You've kept me accountable every week. I honestly don't think I would have gotten this far without you.

the agents. It's interesting because it's, I had this question the other day, they're like, you know, do they work? Do they perform? It's like, well, it's like any other team.

Like sometimes they do some really brilliant work, sometimes they do some really terrible work. Most of the time it's in between. it's like, that's the same as any team. No.

And I haven't given them these. speaker-1: Do have one-to-ones with them? So the one question I ask you every single week on this is are you charging yet? So you charging yet?

speaker-0: Well, it's set up. It's working. The payments is now in place. I a presentation on this at Mind the Product early this week, and we've got this ⁓ today.

so it's effectively become launch week by default, despite trying to avoid a launch as such. But yeah, I was working 2 AM on night trying to get it all ready. speaker-1: Okay, that's wonderful. Let's take a little bit of a retrospective look at this as well.

you've built that has a lot of components, a lot of functionality, a lot of features to it. you were, you done product for a long time, I've seen some really great frameworks and advice that you've given to other people. Would you have advised anyone to start this way? speaker-0: to build out such a fully-fiction platform.

I was chatting to Matt Lemay about this last night. I can't see if he's in the audience. honestly think AI has completely changed the concept of MVP, because MVP was designed for an era where it was expensive to build. And so you to build the least amount and get it out and validate it first.

In this the tools I'm, guess, competing against. speaker-1: As your MVP essentially, speaker-0: are established tools, they're established practices and the people who using this are tech savvy and they have certain level of expectations. so, yeah, I've used a lot of AI type tools and the quality often isn't there and it's really frustrating and actually really unimpressive. And so didn't, I'm used to working and delivering products at certain level of quality and I didn't...

I didn't want to not do that. in theory, I could have rushed this ⁓ a couple of months ago. didn't want to do that because I wanted to make sure it was a quality product, worked, it did what it was supposed to do and actually delivered value. So it been a super intensive exercise.

The building is absolutely not the hard part. The hard part the moving target. Because I would talk to these product leaders, and month to month, their thinking, their understanding of... their organization and their process and what they want to do and how they want to use AI would shift.

that's, you it's one of the reasons why we've done the Makers Manifesto, which we'll talk about. actually the hard part was understanding how to this in a way that supported where these teams are heading, where they want to and to get enough understanding of that across organizations to know what was possible and therefore, you know, help people on that journey, I guess. speaker-1: Okay, so that was a perfect segue. Thank you.

So first off, there is the Slido QR code up on the screens. If you do have questions that you'd like to ask us, please do get them in. We'll start checking that in just a minute or two. the other question I ask every week is, so what's happened in the last week?

Has it been a busy one? What's going on? ⁓ speaker-0: my god, it's been insane. Yeah, in such a great way.

So yeah, as I said, I was madly trying to get the product to, so it all worked. One of the challenges with the fact that it is so fully-fictioned is there's a lot to test. And so there was a mad dash over the last week to make sure it all fully worked and was at a level that I was comfortable with releasing into the wild, so to speak. We've also had mine the product.

event straight after we're now in Lisbon. I love Lisbon, I'm so glad to be here. Yeah, it's been a crazy week on top of, you know, also personal things. My husband had surgery and I was in a rowing race and, I have three kids and all that kind of stuff.

So been super fun. There's nothing like a hard deadline to really focus the mind. speaker-1: Okay, but let's focus on the thing that you talked about at Mind the Product. the last few months, ⁓ been working together with a bunch of other fantastic people.

Some of them are here, Rich Mironov, Martin Erickson, Matt Lemay, Radhika, Andre, Joao, doing an version ⁓ of the manifesto called the Maker's Manifesto. How did this about? speaker-0: Yeah, so as I said, spoke 96 product leaders in the process of building discovery. And the thing that really struck me about those conversations was how wildly different people's understanding was of AI and the implications it would have.

you put in touch with some of the conference organizers. Andre was one them. And I'd put my pitch together on these are all the things will be different in the new world for product managers. And I thought it was on in particular in Australia who runs leading the product.

I to her, was like, this this is a market shift. really what we need is people to come together and say these are all the things you now need to think about in this AI world. And I said, it's almost like we need the next agile manifesto. And she was like, I love it.

And so I reached out to few of the names you mentioned, some of the guys I know in London. And I was like, I've got this crazy idea. What do you think? And they're like, absolutely, we need to do this.

in, let's do it. And so Andre very kindly and Adrienne ⁓ me contact a bunch of people around the world because... It was very clear to me from the start, if we were going to do this, it had to be properly representative. It had to be cross-disciplinary.

It had to be global. had have ⁓ all the different perspectives involved in building great products. And so ⁓ pulled together experts and operators globally. Amazing group of I feel really privileged to be part of group.

was a crazy ambitious Everyone, if saw LinkedIn yesterday, the comment made on there was the most common I got was, I'm honored to be asked to be part of this. speaker-1: And the second most common response was, have you got Rich Myrnau in yet? speaker-0: Yeah, yeah, because the way it kind of worked, said, look, we started with such a great group of people. Anyway, said, look, anyone you think, anyone you respect and think should be involved in this is very welcome.

⁓ most nominations I had was for Rich. Yeah. speaker-1: So the I'll put it up on the screen here or at least the link to it. ⁓ So similar to the agile manifesto.

It's got a preamble, it's got four basic values and then, sorry, 16 principles, ⁓ values, four per value alongside it. You get it all here. ⁓ There is a text human readable version. There's also an MD file so you can ⁓ download it and it into your CodMD.

I've done it, a few other people have done it, and you had an interesting experience trying to apply it as well. speaker-0: I did. So we have skills files within Discovery. can put whatever way your organisation is choosing to work in, can add it to Discovery and it uses that context as well.

So I uploaded the Makers Manifesto skills MD file to be embedded in each organisation within Discovery by default. And again, was one of those ones that was like... It was definitely at the bottom end of the spectrum in terms of team performance. took quite a number of attempts to get it to work.

But then when it did it, added it as a skills file, it had made up a completely new manifesto. speaker-1: all humans. speaker-0: gave it to ⁓ You need to follow word by word. So yeah, anyway, it's there now.

speaker-1: you And one of the key principles in there is humans assigned the work. You can't delegate autonomy and authority. So let's move to some of the questions from the audience. Faith, what was the structure of your process of iterating on the agents?

It can be very time consuming to make such a workflow create an output you can truly find complete. speaker-0: Gosh, that's a big question. The process of iterating on the agents. Actually, mentioned Opus 4.

6 was a game changer. There was, because I had a lot of challenges with this over a number of weeks. And then Ed Biden from Hustle Badger, do amazing product management training, was advising me. He joined one of the episodes ⁓ to talk about And so I use ⁓ Lengfuse, an eval's now.

Yeah, it was a lot of pain. A lot of ⁓ trying capture that, put it in. Even the more recent issue I had with when the LLMs joined my bank account, went back through and made some changes and obviously put some guardrails, circuit breakers in. But even I found, because I did at one point completely change the architecture of the agents.

So I had originally set them up as quite separate things. One of the things I wanted to do was make it so that customers could put their own LLM keys in because if they've already got an enterprise license for we use Claude open AI perplexity and I'm one Anyway, it's full. they've already got their enterprise licence for any of those that they've gotten approved through IT screw, it's far easier for them to put their keys in and it's then within their own environment, the data.

so I needed to change the whole architecture so it was all routed through one router that they could put their own LM keys in. But also that's what I used to track, to do the tracing for the eval's platform and also to track the usage for billing. And it out from this incident a couple weeks ago, even though I had distinctly a number of times said, all the agents through this platform? it's also reading off some skills files around how they're supposed to operate.

despite being told multiple times that they were, they weren't. And again, I've to go back through and sort of clean that up and make sure that all of the are actually operating off that through sort of router now. So it has been... It's been a challenging exercise.

are now better tools to agents available more recently. But it like so many people talk about, I just whip up this agent to do this thing. It really is not that simple. To get it to the point where it's actually performing the task you want at the quality that you want, it takes a lot of investment.

speaker-1: Creating an agent is easy, training it is not. next question. As a solopreneur, how do you prioritize what to work on? How do you decide between building your product and doing activities to market it and get new users?

speaker-0: Yeah, it's a very good question. This is a habit I've had for years now actually. ⁓ I'll have two or three, at start of the week I'll have two or three problems I identify that I ⁓ to solve by the end of the week. And so that's really where I focus my attention and decide.

day a long time when you're building with AI. You can achieve a lot in a day or two. And obviously we the that's a weekly activity. It really depends on sort of what's what's coming up, but I've ⁓ used to help, because when you're on your own, I guess it's very easy to let time slip.

And I've used the very people I've talked to and been part of through Alpha and I've often used their to create discipline for myself along the process. has, since it has definitely... been more challenging is I've had a lot of distractions. yeah, think where we've got to is a pretty good result in I get to be here on stage and I've done something right.

speaker-1: Fantastic. unfortunately, that is all the time that we've got for today. I'd like to do ⁓ last thing if we can, ⁓ we don't get to do this with people like you very Well, first off, we'll be around for the rest of the day if you want to talk to Faith or me and ask us any other questions about the manifesto. You have all these other people you can talk to as well.

But if you'll indulge us, can we do a quick selfie with the crowd behind ⁓ yes. speaker-0: Yeah. speaker-1: And if you want to make noise for this and be excited, you very much. Thank so much.