Spotify Podcast Link: https://spotifycreators-web.app.link/e/ALVMgyYzyQb 

‍

Summary: 

In this episode, we kick off our AMA series with Framework’s Partner & CTO, Jim Texier, as he shares insights on how AI is transforming Vertical SaaS. Drawing from his experience as the former CTO of Lightspeed, Jim explores key trends, emerging opportunities, and the evolving role of AI in workflow automation, defensibility, and integrations. Founders and investors won’t want to miss this deep dive into the future of AI-driven SaaS!

‍

Transcript:

Mikhail

Hi everyone, and welcome to Framework's first AMA episode. I'm joined today by our partner and CTO, Jim Texier. In this series, we'll have various members of the Framework team answer questions submitted by founders, investors, and the broader ecosystem on some key topics and trends. And so today we're tackling vertical SaaS and specifically AI's role in reimagining workflows across all industries. And so with that, welcome Jim. For any new listeners, could you give a quick background about yourself and your early exposure to vertical SaaS?

Jim

For sure. Thank you for having me. A quick background—before joining Framework three years ago as a partner and CTO, I was an entrepreneur in the early-stage natural language processing domain, long before Transformers and LLM models. I then served as an executive, holding roles as Chief Product Officer or Chief Technology Officer at various companies across insurance, e-commerce, and SaaS software.

My last role was as Chief Technology and Product Officer at Lightspeed, where I was part of the team that took the company public. I exited in 2021 to join Framework.

Mikhail

Amazing. Thanks, Jim. So diving into some of our submitted questions, our first question comes from a founder curious about what industries you saw vertical SaaS or AI impact the most in 2024.

Jim

Yeah, that’s a good question. There are actually two answers to that.

Vertical SaaS has really started to displace some incumbent players and address new markets, probably within the last two to three years in a significant way. Various domains have been impacted—I’d say real estate, construction, healthcare, legal, and, of course, finance, which has been touched to some extent.

We’re also continuing to see change in the retail and hospitality sectors, but that’s more within the realm of generic vertical SaaS, which is still growing and thriving. When it comes to AI, however—which is more recent, at least AI as we envision it in 2025, particularly with LLMs—that’s a different play. One of the first domains to be significantly impacted has been those involving text, as text understanding has improved rapidly. Legal, for instance, has seen major disruption, with many startups emerging and even reshaping the professional services landscape.

Cybersecurity is also evolving at an incredible pace, both on the offensive and defensive sides—what you might call the "dark side" and the "light side" (or even the gray areas in between). This sector will continue to change dramatically with the rise of vertical AI.

Healthcare and life sciences are also seeing transformation, particularly in analyzing medical records, processing clinical trial transcripts, and assisting research efforts by providing a more comprehensive 360-degree view of patient data. Systems are improving rapidly, and we actually invested in a company called Evidently that is doing exactly that.

Beyond industries and domains, some jobs will undoubtedly be affected as well—but we’ll probably get to that later.

Mikhail

Yeah, and on that note, another founder was curious about which functional or operational roles are likely to see the most significant shakeup as companies shift spending toward AI tools rather than relying entirely on labor. In previous discussions, you’ve mentioned middle management and the evolving definition of what it means to be a world-class developer. Have your thoughts changed? Are there any new roles—perhaps in regulatory or compliance—that might undergo an interesting shift due to vertical SaaS?

Jim

Absolutely. Last time, I mentioned that I believe middle management will be impacted, which is a relatively new shift affecting white-collar jobs. Historically, technological advancements—whether in robotics, automation, or software—have primarily affected blue-collar jobs. However, with the advent of AI and its ability to express, understand, and engage in conversation, we can anticipate that middle management tasks will also be affected, necessitating a transformation in those roles.

There are two parts to this. From a technical perspective, there is a clear shift for developers, software engineers, and entrepreneurs in the kind of talent they need to hire. I mentioned this last time, but I’ll reiterate briefly: beyond being proficient in computer science and coding, developers now need to be "native prompters." We are seeing the emergence of AI-driven solutions designed to enhance developers’ ability to prompt, evaluate, and effectively use LLM models.

To make a quick aside, this shift is happening incredibly fast. A year ago, being a skilled prompter was a valuable job, requiring expertise in contextualization, breaking down prompts into smaller components, optimizing context windows, and improving evaluation techniques. Now, we are seeing new startups addressing this space. For example, an early-stage YC startup, Zenbase.ai (founded by Canadians), is working on tools that allow developers to program prompts rather than manually prompting. This represents a new layer of evolution.

When it comes to programming AI agents and managing multi-agent environments, open-source solutions like DSPy are emerging. Developers are now building tools specifically for other developers, which is forming a new vertical in itself.

Now, moving to the second point—which I didn’t mention last time but should have—certain jobs will definitely be impacted, and we are already seeing that shift. The most obvious examples are customer support and customer success. AI-powered chat interactions are improving at an incredible rate, making these roles increasingly susceptible to automation. Quality assurance (QA) is another area being transformed, as AI is enabling smarter and more automated ways to develop and test applications. This process was already somewhat automated, but AI is now changing the game entirely.

Looking further ahead, any repetitive human-in-the-loop task is at risk of being automated, particularly with the rise of AI agents as opposed to just AI applications. Agent-based AI can already learn from human inputs and continuously improve, which could open the door for AI-native applications that disrupt not only vertical SaaS but also incumbent SaaS players and even professional services firms.

To give a concrete example, I recently came across a startup making a bold move to reinvent accounting software—essentially competing with solutions like QuickBooks but using an AI-first approach to handle the entire process end-to-end, with nearly zero manual intervention. This is a fascinating shift. While we don’t know how quickly this transformation will happen, it is inevitable.

Mikhail

So then, I guess on a similar note, our third question comes from a founder asking, how can startups create real defensibility around AI products within their vertical, especially, like you mentioned, when new models and new approaches are constantly emerging? And so, do you have any insights on how companies can build that defensibility today, but also how that might change in the future—say, in a year or two?

Jim

Yeah, again, good question, obviously, and a tough one. So it's going to be a mix of opinion and gut feeling here. I don't have a crystal ball that tells me what to do, but maybe two things.

First of all, I'm going to take a step back. If you look at the new AI and LLM, the evolution and the rapid change, but at the same time, the insane investment being made in foundational models, it's just crazy, right? Billions of dollars are being invested to make them better and better. We have an incredibly fast evolution of large language models. You now see chain-of-thought reasoning models like o1, o3, and o4.

I believe that even at the foundational level, between infrastructure and application, I'll come back to application later—as a startup, unless you're a startup that can raise a billion dollars, but let's put that aside—most probably, there will be consolidation there. I don't think we can have a sustainable ecosystem where multiple companies are investing multi-billion dollars into foundational models. Eventually, they are getting more or less on par with each other.

Why I'm saying that is, as a startup or as a CEO or entrepreneur of an AI-first startup, you might want to be sure that you're betting on the right horse. Or, from a technology perspective, we often use gateways or proxies or middleware, and you might want to really investigate multiple models and multiple solutions—Anthropic, OpenAI, and the like—because there's no guarantee that all of these will survive. That might happen more quickly than we think.

And if your company is highly dependent on one provider, one fundamental model, you might go down with it, or you might have a hard time. So it's a tough one because there is a trade-off between speed and flexibility. Either you play the fast game, and you're lucky, and you're on the right horse, and your company will grow fast enough to give you the resources or be acquired and make a nice exit, or you want to make sure that you keep your eyes open on multiple options. It's not because you rely on OpenAI that you're going to win. Maybe you need Anthropic for your use case, or maybe Anthropic will be acquired or disappear. Who knows? But I am predicting, or at least guessing, that something will happen in that space that will disrupt and cascade through all the layers above—like infrastructure, professional providers, and application layers. So be aware of that.

That's one.

The second part, obviously, going back to the original question, is that you need to hire people who understand and are native with this kind of model. It's a tough nut to crack, but yet again, we are now seeing solutions for developers growing and helping with that.

But the third part is still the same. If we’re talking about moat and defensibility, the time frame for maintaining that defensibility is getting shorter and shorter. So, in my opinion, it’s always about being faster than the others.

And the moat will eventually be built over time because you have a significant market share, because you address core issues and not just the nice layer cake, you know? Having really strong expertise in the domain or vertical you address is essential.

And that's also an evolution. If you look at the difference between AI applications and AI agents, basically, an AI agent is a tool that acts using AI. So, it's a smart combination of goals, learning, understanding the goal, and executing a workflow. It’s not that different from software, but the user experience will change. The interaction will be between agents, and eventually, some humans will be in the loop. So, you have to think about your architecture and application in a very different way.

Long story short, domain expertise matters in vertical SaaS and vertical AI. You need to rely on the proper infrastructure and architecture to get the flexibility you need because nobody knows who will win the long game of the underlying model. And I'm predicting—I might be wrong—that there will be consolidation, and some companies will disappear despite the billions injected into them.

That’s my guess. We'll see in a couple of years.

And last but not least, let's not forget user experience. I think everything else is a waste of time. If you're lucky, you might have some potential patents that you can file and whatnot. But in the history of software, specifically in recent history, getting faster to the goal is what really matters.

That is your best moat - credibility, reliability, and solving core issues. I think that's always a good recipe for success.

Mikhail

Amazing. Yeah, that definitely makes sense.  

I guess on a similar note, but kind of switching gears from defensibility to scaling, one of our founders asked, what's the most important aspect of scaling a vertical SaaS product after obtaining the first customer versus passing that first million or two million?

Jim

So actually, there is not a single answer, obviously, because we see this every day, and that's not new. But specifically, because you're addressing a vertical SaaS product in a specific industry or domain, what I would say might sound like a very generic answer, but it really depends.  

What I mean by that is that some industries will be more effective using channel-based access, while others will be more effective with direct, account-based marketing. In some cases, you need a very traditional approach—where you know exactly who you're targeting, have AEs who can engage directly, and a team that can manage the follow-up work. On the other hand, for SMBs, you might rely on more traditional growth strategies—virality, keyword targeting, and push marketing.  

So there’s not a single answer, obviously. Long story short, you have to deeply understand your customers.  

Let me give you an example. If your vertical is in financial institutions, clinical studies, or optimizing transportation for airlines or aerospace, your marketing and sales cycle will be very different compared to targeting real estate, hospitality, or e-commerce.  

Again, I think there are a lot of great books and documentation on how to scale sales. One of the classics, "Founding Sales" by Peter Kazanjy, is a good reference. But ultimately, you need to find the right channels and understand your sales life cycle—that’s how you scale.  

Interestingly enough, we now have tools to help with this. There has been an explosion of AI-based solutions designed to improve sales, SDR performance, and marketing. These tools analyze revenue operations, or RevOps, to increase the efficiency and productivity of teams and sales channels.  

We actually recently invested in a company called Vasco, which is doing exactly that. Shoutout to my friend Guillaume Jacquet, a repeat entrepreneur. I had the pleasure of working with him at Lightspeed after we acquired his company, and now he’s running a startup that focuses on AI-powered RevOps—helping companies optimize their sales funnel based on the most effective journey, channel, and conversion rates.  

So, at the end of the day, you have to experiment. There is no universal playbook that works for all companies, and you have to understand the nuances of your vertical.

Mikhail

That's definitely, definitely helpful.  

Our next question moves a little bit more back to the tech side of things in terms of the emergence of vertical AI tools, as opposed to the traditional end-to-end SaaS platforms. So, with vertical AI focusing on automating or augmenting a specific task or solving a wedge problem within that broader workflow, how important do integrations and APIs become for both existing as well as new AI tools and agents?

Jim

Yeah, so I would say APIs have always been essential. It took some time for people and software companies to fully understand that. I would say maybe 20 years ago, but certainly 10 years ago, tech people really started to embrace the concept of API-first. We moved away from the classic client-server model or multi-layer architectures—presentation layers, all these technical terms—to an API-first approach. Now, we’re trying to say AI-first.

But if you think about it from a technical perspective, it’s all about the interface that allows you to program or interact with your application. That’s what an API stands for, by the way. But in the context of vertical AI and virtual AI agents—applications made up of agents, specifically multi-agent configurations—it’s even more critical. These systems rely on agents talking to each other, making decisions as a group, or using a central agent to handle arbitrage, replace human tasks, and execute workflows. In that setup, interfaces become even more essential.  

So I think being AI-first means you're API-first by default. However, I don’t want to discount the value of user experience. At the end of the day, when it comes to the application layer, UX is king. And I’m not just talking about UI—the graphical interface—but UX, the overall experience.  

Nowadays, the experience could be 100 percent conversational, which is new and needs to be deeply understood. You can imagine that some jobs could be replaced, like PMOs or project management roles, where most of the work involves collecting information, ensuring projects stay on time, on budget, and on scope, and synchronizing people across teams. That entire process could be done much more efficiently and in a reliable, repeatable manner using a combination of AI agents and the right user experience.  

You could even have a virtual PMO inside Slack, functioning just like a human, managing coordination with an empathic approach—even though it’s an algorithm. And yet, it works.  

So, obviously, integrations and APIs are key. They always have been—it’s not new—but they are even more essential now as we shift toward more automation, more agents, and applications talking to each other, rather than just humans interacting through an interface.

Mikhail

Yeah, and it seems like AI is sort of the next layer to come down, similar to how cloud kickstarted a bunch of SaaS platforms, and how the fintech and payment layers, which we focus on heavily at Framework, were and are still being added on.  

So, have you seen this new AI layer in vertical SaaS really starting to accelerate upsells and expand ACVs for startups now that it can handle a lot of the service or even voice aspects of workflows, as you mentioned?

Jim

Yeah, so there are two parts, maybe two parts and a comment.  

Interestingly enough, if we talk about big companies, large corporations, or successful enterprises, we all know that these companies usually have a hard time innovating. Change management is tough. They typically acquire to go faster because innovation is hard. I'm saying that because, somehow, I’ve already seen startups—software startups—having a hard time shifting their paradigm.  

So basically, what I'm trying to say is, yes, AI could accelerate and change things dramatically, but interestingly enough, this is a message for my fellow entrepreneurs who have scaled up and reached a certain stage of maturity, but are not yet billion-dollar companies.  

Don't sleep on this. Watch out. You're going to be disrupted very quickly if you don’t react and embrace the new paradigm. It seems easy to say, but it’s not always that simple. You might have internal challenges, not necessarily linked to friction or resistance to change, depending on your culture, but more so because of your architecture. So think about that now, because it's coming, and you might be disrupted.  

Now, back to the question of acceleration and increasing value. I think the most obvious change we’ve seen, including within our existing portfolio and from other companies sharing similar experiences, is cost reduction. We talked about this, right? If you automate or semi-automate most of your customer success, customer support, or QA functions, and improve the efficiency and productivity of your developers, you're essentially doing more with less.  

Once you start doing more with less, you can grow faster, add new features, and expand the scope of your solutions and value proposition within an account. So it’s more of a consequence of being more efficient and having more resources that we see as an immediate effect of AI.  

To give you a specific example, B2B enterprise software typically requires a lot of configuration. Each large customer is unique, with integrations, data connections, and the need to understand local language, jargon, and company-specific vocabulary—even within the same domain. If you look at the extreme end of the spectrum, take a solution like an ERP such as SAP. Today, we have hundreds of thousands of consultants making their living simply because they understand SAP and can translate customer needs into configured solutions.  

Going back to SaaS in the B2B enterprise space, companies still spend a lot of time on implementation, configuration, and fine-tuning their solutions for customers. This is the part that is now changing dramatically. Most of the configuration, which used to take months and required consultants—whether third-party or in-house professional services—is being displaced by AI and automation.  

Just to give an example, I was recently speaking with the co-founder and chief technology officer of Maxa, one of our companies. They have a very large and private customer, and what used to take months to configure complex solutions has now moved from a couple of months to six weeks, and recently from six weeks to just two days.  

How did they do that? They used transcripts of conversations with their customers, fed them into a fine-tuned AI model, and the AI model was able to generate the necessary configuration and parts of the required code. So suddenly, 95 percent of the job—understanding the customer’s needs, environment, and translating that into code—has been replaced by a properly fine-tuned AI. It’s amazing.  

And going back to what I said before, I mentioned customer success and QA, but I think professional services will be dramatically impacted as well. Once large software companies in system automation can significantly reduce the implementation cycle of their solutions for customers, all of these professional service roles will be at risk.  

In my opinion, that’s either a risk or an opportunity—depending on which side of the software you’re on.

Mikhail

Yeah, that's super important to consider.  

And so, as we sort of come up on time here, Jim, a lot of investors had reached out to ask this question to you. As we kick off 2025, what is the most important thing to consider when trying to determine whether it’s too early or too late to invest in Vertical SaaS?

Jim

It's a one-million-dollar question, right?  

I don't know the answer, but I'm going to share an anecdote to wrap up on that. Yesterday, I was with a group of investors, and one of the participants was sharing how he’s now trying to use the new LLM AI to make better investment decisions.  

Let me tell you the story. I won’t give names, and I’m not saying it works, but it’s an interesting one.  

So, the big, well-known investors—Andreessen, Peter Thiel, and others—have been quite vocal about how they invest. They’ve written extensively, spoken publicly, done podcasts, and given interviews, all expressing their frameworks of thought. This investor I was speaking to, based in the States, has a pet project where he has trained AI agents to act like Peter Thiel.  

It’s very funny, but it’s actually part of their investment committee now. They feed information into these AI agents and ask, "What do you think, Virtual Agent 1? What do you think, Virtual Agent 2?" The idea is that these agents attempt to behave like the top investors in the world, following their known frameworks. It’s an interesting use of AI. I don’t know if it’s going to work, but I found it fascinating.  

But again, this just shows that once knowledge is understood, you can train an agent to help you make better decisions.  

So back to the question—when is it too early, and when is it too late? Obviously, do your homework. Look at the competition, incumbents, and market size. The earlier you are, the more weight you put on the founders, obviously. And the world is full of great stories of founders pivoting.  

I don’t think there’s a single unicorn today where the end product is exactly what was envisioned at the first investment stage. If you believe in the ability of the founders to pivot and execute, that’s still a strong bet.  

Everything else is just a combination of homework, hard work, and luck. That’s my take, at least.

Mikhail 

Well, that's all we can fit in today's episode, Jim. Thanks so much for sharing your expertise and taking the time. And a big thank you to everyone who submitted their questions.  

Make sure to stay tuned for more episodes featuring an incredible lineup of guests coming up. Visit our website at framework.vc, and thank you for listening.

‍

Jim

Thank you! 

‍