🤝 Meet Joshua Budman. Originally from Toronto, he made the leap from a promising career in medicine to found and sell Tissue Analytics, a company that used machine learning (today we’d probably call it AI) to help do wound imaging with smartphones. Now he is building another startup in the AI-adjacent space. We chatted with Joshua about the differences in Canadian vs. American business culture, good and bad reasons to do a startup, why he thinks search traffic is going to zero, AGI skepticism, and more.
You left a career path in medicine to pursue entrepreneurship and it worked out well. For others considering taking the leap from a relatively safe career to entrepreneurship, what do you think is a good reason to do so, and what is a bad reason?
Joshua Budman: One good reason: You like to build and want to very clearly see the direct impact of what you build quickly and at scale. This is possible to do at larger companies/in different professions, but my argument is that it’s extremely rare and hard to accomplish if you’re not building your own company.
One bad reason: Money. This is kind of obvious, but the odds of massive success are extremely low. Even if a start-up is very successful, by the time an exit event takes place it typically isn’t even that lucrative for the founders, especially when you consider the time and work it takes to get to an exit.
Your first startup was in the health space, and using machine learning long before AI was the cool thing to do, so you know what you're talking about on this topic. What do you think are the most promising applications for AI in medicine?
JB: I think the market has basically “answered” this question for us, to an extent, with the wide array of AI health scribe companies that have emerged in the last few years (e.g. Abridge, Tali) and the incumbents that are investing heavily in these types of applications more than any other area (Nuance with DAX, Epic with their own Ambient experience).
The application of AI to more diagnostic tasks obviously has the potential to serve tremendous value - especially if they are framed as clinical decision support tools - but the regulatory burden on this type of tech makes it much harder to scale. I also think we’re still in the early innings of clinicians being able to make use of diagnostic aids; in contrast, clinicians have gotten pretty good at trying to make their workflows as efficient as possible as health software has improved.
Tl;dr - the use of AI applied to documentation efficiency/time savings-related tasks is, in my mind, the clear “killer app” for AI in healthcare right now. And I don’t feel any company has quite nailed it, yet.
You're Canadian but built a business in the U.S. What are the biggest differences you notice about doing business in the two countries?
JB: I haven’t tried to start a company in Canada, of course, but I have sold to Canadian institutions/companies and US companies (as well as companies in other parts of the world). It’s an objective statement that it is much more straightforward to garner private funding in the US vs Canada. Even within the US, companies with better physical access to VC’s in cities like SF or NYC tend to raise more funding relative to companies in secondary or tertiary markets. That’s not to say that funding is the end-all-be-all for becoming a successful company, but it’s getting harder and harder to differentiate as a bootstrapped start-up. That’s probably the main advantage of being an entrepreneur in the US versus trying to start a company in Canada right now.
On the sales/customer front, it was always much harder for us to sell to Canadian healthcare institutions than US healthcare institutions. Canadian institutions tend to be a lot more “consolidated” and the vetting process for new tech was typically very intense. In almost every case, an RFP was needed just to make a decision on a new vendor, even for tiny institutions. This is definitely not a bad thing - it’s obviously important for health systems to vet their new partners and tech - but we’d have had a much harder time as a business if we could only sell to Canadian companies. US companies/health systems were typically more willing to at least try products quickly. This led to faster customer acquisition, but probably less well thought out deployments as compared to Canadian institutions.
Thankfully, customer acquisition has gotten far more remote-friendly over the years, so borders don’t really define target markets as much anymore.
Your current business is pitched as a way for companies to show up in AI chatbot answers, so kind of like SEO for 2026. How does it differ from old-school SEO?
JB: There is a ton of overlap between pre-LLM SEO and SEO since the introduction of LLMs to search engines (AEO/GEO/LLMSEO/AI Search/etc). I’ve heard it framed as 80% legacy SEO, and 20% new tactics to get cited by large language models. I think this is a fairly crude but directionally accurate way to frame it. My two cents is that the math has objectively and significantly changed with the introduction of LLMs, so all of the traditional SEO “levers” should be weighted differently.
No one quite yet knows those new “weights” - they are guessing if they say they do - and those weights will continuously change as the core LLM developers (OpenAI/Perplexity/Google/Anthropic/etc) change their models and deploy new features. In my opinion, what makes a company like Noble possible/relevant now vs before is 1. Advances in automation driven by LLMs (agents) allow us to scale the process of garnering brand mentions (fka “link building”) better than ever and 2. LLMs are far less driven by spammy “link building” tactics (where companies try to get links on any and all sites), which creates the need for new frameworks of high-quality digital PR.
Should businesses be worried about their search traffic drying up? And what should they be doing now to prepare for more people using AI chatbots to search for information?
JB: I think website traffic is going to go to 0, yes, but I don’t think companies should be worried about that. My hypothesis is that the LLMs are going to start to own as much of the web as possible, such that they’ll act as the “front door” to every brand in the near future. We already see this trajectory starting with the LLMs facilitating checkout after a search for DTC products. I think the best way to prepare for this shift is to make sure your web presence - both owned and third party - is set up as well as possible to be consumed by web scraping tech and AI agents, so that the information surfaced by prompting an LLM is representative of what you want your prospects/visitors to see.
Do you think we're going to achieve something like AGI in our lifetime?
JB: I don’t think so, no. I once saw a talk by Andrew Ng (2018) and he hypothesized that we’re closer to overpopulating Mars than we are to achieving AGI. I agree with that assessment.
I think GPT-3.5 (ChatGPT) was definitely a “holy sh*t” moment for a lot of people both in the AI community and out of it, but I feel we’ve reached a performance asymptote at this point that can’t just be solved by applying Sam Altman-esque “Scaling Laws” and throwing a ton more compute at the problem. I think to truly achieve AGI, the LLM companies will need to somewhat rearchitect these models from simply being very good information aggregators and surfacers to actual thinking models. And I don’t think this is a very straightforward mindset shift. Right now, I truly feel the LLMs are still very shiny toys that perform well at a bounded set of non-deterministic automation tasks. I think companies like Yann LeCun’s new venture are starting to actually try to tackle the AGI problem, but I don’t think he’ll solve it in our lifetime.