A Conversation with Ehsan Mirdamadi
Insights from a serial entrepreneur and investor on university tech transfer, sovereign AI infrastructure, and building scalable technology in Canada
I have heard often that some of Canada’s issues with innovation stems from the fact that we do not celebrate our entrepreneurs and innovators. Through the process of building CanInnovate I have had the opportunity to meet many impressive founders, investors, and supporters of innovation across the country, and I have found that this holds true. We have an incredible base of domestic talent building in spite of the obstacles, quietly contributing their time and energy to improving the place they have chosen to call home.
Today’s interview is with one such person: Ehsan Mirdamadi, a serial entrepreneur who immigrated to Canada in 1998 and who has been building and supporting other builders ever since. He was involved in the early days of cloud computing, worked with Ron Davis at Stanford on DNA sequencing, worked in university tech transfer, co-founded angel networks to pay forward his early successes, experienced innovation in both Canada and California, and chose to come back to Canada, building in the AI space with Codalio and contributing thought leadership around related policy. His experience, earned from such a wide variety of perspectives inside and outside the Canadian innovation ecosystems, offers an invaluable look at both the potential and challenges Canada offers its innovators.
While some previous interviews have explored similarly themed stories, I intend to make this a more consistent element of CanInnovate reporting, through a series that tells the stories of Canada’s builders. If you know someone you think should be celebrated for their contributions to Canada’s innovation ecosystem, please tell me how best to get in touch with them.
Your email client will probably truncate this post. My key takeaways are presented at the end, so be sure to read the web version if you want to get the whole story. Many thanks to Ehsan for taking the time to share his insights.
Interviewer’s note: Ehsan Mirdamadi approved the final version of the section entitled “Interview with Ehsan Mirdamadi” and had editorial input on that section, with the option to rephrase and expand on the ideas discussed in the interview without changing or removing any intended meaning. The key takeaways presented at the end are my own commentary, and do not necessarily represent the views of Ehsan Mirdamadi, Codalio, or any of the investor networks of which he is part..
Interview with Ehsan Mirdamadi
KB: Tell us about yourself, how you got here, and where you are today.
EM: As an immigrant, I landed in Canada when I was 18, in August of 1998. Less than a year later, I started a computer science program at York University. That same year, I had this bug in me that I wanted to build a company. It was in the early days of the high-speed internet, so I ended up playing around with the Rogers @ Home service to see if I can actually run a server on it.
Before I knew it, I had a company. Even before I graduated, I had a few employees for it. That turned out to become one of the first—maybe the very first—Canadian cloud computing company. I essentially bootstrapped the company to about 100,000 clients, through partnerships and channels, and eventually sold.
One of the biggest questions I had from myself as a Canadian entrepreneur, was that: given that we started before almost everybody else in that environment, why were our American counterparts able to scale similar companies to billions of dollars, but we were only able to get to millions?
Obviously, I was an inexperienced entrepreneur at the time. I was more of an engineer and computer scientist. So I went on a journey to find that answer. The most shocking learning in that experience was that there were companies that were coming into our space 10 years after, and they were still getting to the billion-dollar mark very quickly. So as you can imagine, in the technology world, 10 years in cloud computing is essentially a century. So there were these companies that were coming into our space a “century” later, and they were still able to build billion-dollar companies in no time. I think the lesson, for anyone trying to build today in the AI space especially, is to not be afraid of missing the boat. These things are going to be around for the long term. I think the impact of AI is probably way bigger than the cloud in terms of the way that it will impact our lives.
In search of the answer, I moved to California, to San Francisco. I worked with a few scholars at Stanford University in the Biotech Engineering Department. I co-founded a company with a bunch of scientists that were actually part of the team that invented pyrosequencing back in Sweden, at KTH. My advisor was the head of biotech engineering at Stanford, Dr. Ron Davis, who was amongst the top 10 most influential scientists in the world at the time. I was trying to help them to commercialize a short DNA sequencing technology.
To make a long story short, I learned a lot about IP commercialization at Stanford. I realized that they were doing a lot of IP commercialization, including the IP that later became Google. A shocking realization for me was that the invention disclosure form that led to Google was just two pages, in which they discussed how they could assign ranks to nodes in a linked database. With only that, they were able to start to build one of the largest companies in the world.
Back to DNA sequencing: we raised a couple of rounds of financing. In the course of that, I realized that working directly with scientists that still maintained their affiliation with the university and the research attitude wasn’t easy. And so when we took a quick exit.
At the time when I decided to come back, Trump was about to get into power. I had just a very short amount of time, maybe a few months, before I could have transitioned my H1B to a Green Card. I made that choice that I didn’t want to do that and I made the deliberate decision to come back to Canada.
So I gave that up, came back to Canada, and for about four years, I got myself engaged with university-linked incubators. So at University of Waterloo, Toronto, York, Ryerson, and McMaster. I was amazed to see the depth and the breadth of the companies and the ideas that were coming out of these universities. Obviously, in contrast to what I was seeing in San Francisco, it was eye-opening. The quality of the IP was amazing. The passion, the energy, the government support, the ecosystem support, all of the ingredients needed to build a successful early-stage ecosystem were actually there.
I started joining some of the angel networks. One in particular, GTAN, the Golden Triangle Angel Network out of Waterloo was, in my opinion, doing a fantastic job. Through them and through my own affiliation with these university-linked incubators, I invested, coached, and advised about 20 companies and was very lucky that at least two of them became Canadian unicorns.
I took some interim positions with some of the companies that I invested in to really support them in getting to the next level. By 2017 when I had my first kid, I realized that there is a big gap between what’s happening in the university settings and the R&D space, and what was happening in the early-stage investment and angel space. So I decided to do something about it.
With the new baby, I didn’t want to just go right back into entrepreneurship. So I took a job at MaRS Innovation, now called Toronto Innovation Acceleration Partner (TIAP). This entity was a grantee of a program that was called the Centres of Excellence for Commercialization and Research (CECR) program. They received about a hundred million dollars in funding from the government, with the mandate to support commercialization of the IP that was being generated at York, Ryerson, U of T, and the 11 hospitals in the GTA.
When I joined they were already a mature entity. They were very strong in commercializing life science-related intellectual properties, but there was a gap in supporting the 3 Universities under their mandate - York University, University of Toronto and Ryerson (at the time) University. So I ended up taking over part of the representation of MaRS Innovation with the three universities. I had offices at the three universities all at the same time at the VPRI level. I was working with their own in-house commercialization specialists. I think we actually made it even worse for them.
A typical process was that somebody would come into our office with a disclosure form filled out, and we would have a couple of meetings to understand what exactly their invention was. We then gave them the results of an IP search using PatSnap, essentially bombarding them with a list of about 200 to about 700 past filed patents. We handed all of those files, PDF files to them and we said, “please figure out if your IP has any overlap with any of those.”
Many of the inventors never came back, because they didn’t have any answers for that. The ones that came back, continued these long meetings. We started to identify the main value of those IPs, using university-retained IP lawyers to get opinions. We then delved into understanding the competitors, doing our own internal research about the use cases. That process usually took about six months.
A very small percentage of those innovations actually made it to the point where we said, “Okay, there is something there.” At that point, we sent them back again, for the third time, saying that the inventors should run a pilot, supported by intros that we would make. We didn’t have a very good Rolodex, but IRAP also gave us a good Rolodex at the time. We were able to connect them with some industry experts. We did a lot of hand-holding those processes. Then maybe they could find an interesting use case.
Again, now for the fourth time, we sent them back to grant writing teams and to write a grant for collaborative research with an industry partner. If they would reduce it to practice, then they could come back and only then would we decide whether or not to pay for the patent filing.
Because of that process, the IP was often actually encumbered, because the industry partner would have some claims to it, and then there were the background IP and the foreground IP involved, all of which made it even more difficult to figure out who owned what. Maybe the industry partner proposed to license it, maybe not. Sometimes we found other entities that were interested in licensing. Finally, after we had some validation of the use cases, we then moved ahead with paying for the patents.
At this point, there is commitment from innovators to commercialize the invention, and they had absolutely no idea. Based on my exposure to the whole angel community, I can confidently tell you that majority of the angels never knew how tech transfer offices operate.
I think at best we were commercializing under 1% of all of our intellectual property. 52 billion dollars of our tax dollars goes into R&D. Obviously not a lot of it goes to universities, but I think about 16 billion of it goes into the universities and colleges [Interviewer’s note: per the previous link, in 2022, $8.4B of federal dollars went to research, with a further $8.9B from higher education institutions, which may include other public sources]. We do not have a big share of commercialization resulting from this.
There were a couple of interesting programs there that relatively few people knew about. There was a program called NSERC I2I (Idea to Innovation). It’s like a seed capital for innovators trying to commercialize their IP that effectively lets them develop a product while still in the academic lab before they spin out a company. It is delivered in several trances, starting with 125k. It’s a very competitive grant program.
To my surprise, all of my colleagues in the university settings advised against fighting for those. I didn’t listen, and I got two of them approved in my first year. At the time there were only 30 of them awarded nationally. If you validate some of the assumptions and if you have the entrepreneurial background, you can actually write good proposals and the government is actually listening.
The projects that I actually filed for were stuck in the never-ending back and forth interactions with the university commercialization specialists. I won’t name names, but there was this one inventor whose work had real potential for Canada’s national interest, and I put a lot of energy behind it, but the people I was working with at the time kept telling me to just drop it, that he was some kind of crazy scientist not worth my time. And you know what validated my instinct? When I decided to apply for an I2I grant for his project, and despite the fact that the program is insanely competitive it got approved, which pretty much confirmed that the work was worth taking seriously all along.
This inventor spent eight years going back and forth with the university, begging for them to file. He was actually correct about the potential value, in the end. That one got an I2I. The other one was for autonomous vehicles, a routing algorithm. But the first one, the inventor, was actually from the industry. He previously worked at Hydro One. He let go of his job. He went back to university, doing research to solve the problem that he knew the industry had. Yet the commercialization officers were still pushing back because they didn’t understand it. These commercialization officers are actually PhDs. They are academic people. For the most part, they have no idea about the entrepreneurial experience and entrepreneurial world.
Obviously, everybody hated me, and I knew that I had to get out. After about 18 months, I decided to try to fix some of these challenges from the private sector because those environments were so rigid that I realized there was nothing I could do from inside.
From there I created two entities. One of them was the Archangel Network of Funds. We were a consortium of five or six super-angels from GTAN. We knew that we were going to do a better job by distributing the risk amongst many deals rather than each angel doing their own due diligence and taking risks on individual deals. We wanted to raise 10 million. We ended up raising 25 million. We actually launched the entity right as the COVID lockdown was announced. So we took the pictures, and then it was quiet.
We decided to use our own capital rather than raising from others, and to fund companies that were doing something about the COVID or otherwise alleviating the health or its economic impacts. We ended up investing in four companies. We created those stories and without even trying to raise capital, people basically came to us, and ultimately we did raise a significant amount of capital. So far we’ve invested in over a hundred companies. In the majority of them, we were the first check. We had a few different buckets for investment, including one that does an automatic follow-on to angel networks. Another bucket B2B SaaS. Another bucket for minorities, women-led companies, and LGBTQ.
The fourth bucket, which is the one that I was most passionate about, was Axion, which had the mandate to accelerate commercialization of IP generated in Canadian universities and colleges. Because of the fact that there was so much noise and momentum in Southern Ontario, we wanted to pay our own dues as private citizens. We had a bunch of passionate angels from Northern Ontario through which we decided to have a secondary angle for anybody who wanted to take their own innovation to Northern Ontario. We found many good companies in those corners.
Another entity that I started at the same time was a company called NuBinary, a fractional CTO service. As an angel, I did a study amongst over 80 companies in angel network communities, and I realized more than 50% of all of the failed companies in their portfolio were failing not necessarily because they didn’t have enough capital, or they were not passionate about what they were doing, or not finding product-market fit. It was because of the fact that they couldn’t scale the technology, which for the most part was an issue anchored around software.
So with NuBinary, I wanted to bring senior expertise in scaling to the early stage deep tech. Within about four years, we grew from 4 to about 50 people. We serviced over 125 companies with a big majority getting to the next round of financing or growing the company. Over time, half of our companies got to series A and above.
We saw a lot of inefficiencies in software development. A lot of inefficiencies in the ways that people were prototyping and building MVPs. A lot of wasted capital. A good 80 to 90% of whatever capital early stage companies for pre-seed, seed, and post-seed rounds went into technology development, and they revamped their software at least three times on average. Even for the ones that eventually succeeded, they wasted a good chunk of their capital just trying to build their software. It was shocking for us because as a computer scientist, best practices in software architecture and coding all existed already, yet people were not adopting them because in university settings, they teach you how to code, but they don’t teach you how to write a good application. There’s a huge difference between the two.
I think we solved that problem, and in doing so we learned a lot. I decided to sell the company, sell my own shares, and then buy that IP, bring it out and build my current company, Codalio. We essentially use the same methods, but augmented it with AI. So we brought the expertise of a multitude of senior engineers from UI/UX designers, senior developers, and CTOs into the pipeline and we are now able to accelerate the whole software development process for scalable, expandable, high quality enterprise grade software by a good 90%.
KB: What you are building at Codalio?
EM: Codalio is a product and software development augmentation tool that allows you to build high quality software, not just a prototype, that scales and expands as needed. To do this, we’re using that tool to first build MVPs, and then we have an automated process and workflow that allows you to quickly iterate and grow it from there without having to scrap any of your code.
We’re positioning it as an MVP builder engine. We hear a lot about vibe coding these days, which is a way for a non-technical person to put a few prompts into these coding tools l and generate some code. But if you do not have the experience and the expertise to understand how the software should be architected, you can only go so far. Those platforms may not scale.
There is now an industry category of coding assistant tools that are primarily designed for the developers. They give you maybe 20-40% productivity gains. Then you have vibe coding tools that are designed for the product managers and non-technical people that actually give you a good 70% of your UI and maybe 5-10% of your backend.
We thought we needed a major new category that could actually get a non-technical person from zero to 80 or 90%, with the last mile going back to the developers. We do that by translating the business objective into a thorough technical scope, and so from that point on, it’s much easier to properly instruct AI to produce better code.
KB: Part of Codalio’s core value proposition is avoiding vendor lock-in and having a sovereign tech stack. That word is being thrown around a lot these days. What does “sovereign” mean when you say it?
EM: A lot of these AI and coding tools, especially the ones like Replit or Lovable, have a certain way of producing and deploying your code that gets you hooked into their infrastructure. If you unplug, you probably can’t even take your deployment with you to. So that’s one of the major parts.
The second major part of it is that all of these tools obviously are using LLMs. We make sure that we design an architecture that allows us to avoid some of the publicly hosted LLMs to generate the code.
The third part of it is that we do everything based on the traditional software development lifecycle. So we’re not breaking that lifecycle, we’re just augmenting it. That was an intentional choice, because to remain platform agnostic you need to do that.
The last part is that at the end of the day, people may decide to go with many different technology stacks in their builds. We chose one that had more of a Canadian root. For the front end, everybody is now for the most part using React, but for the backend, we chose to go with Ruby on Rails, and we added our own layer of abstraction on top of it that essentially commoditizes a lot of the components that goes into any software because 75% of all code is essentially just a repeat of existing code and components.
As an aside, Shopify is one of the biggest supporters and the sponsors of Ruby on Rails. The Shopify stack is actually on Ruby on Rails. We have one of the largest talent pools in Canada, in Southern Ontario especially, around that tech stack. So we made that deliberate choice to go with Ruby on Rails. That makes it very Canadian.
KB: You have a unique and diverse set of perspectives: you’ve built companies, you’ve invested in them, you’ve worked in tech transfer offices trying to get technologies out of the universities and into startups, and now you’re now actively involved building out use cases for AI. Based on this breadth of experience, what’s the main bottleneck to AI in Canada, both from the perspective of companies that are adopting it and customers that are using these tools that you’re building?
EM: I never thought of it this way. It’s very hard to be accurate on this because there are so many things that are happening at a much faster pace these days. Even as someone that spends a lot of time in these environments, it’s very hard to catch up. So I’ll leave that as a disclaimer to start.
With that said, I think one of the biggest challenges is that a lot of the current grant programs that support commercialization do not consider you eligible if you are building AI tools or platforms that leverage LLMs. The assumption seems to be that no one can build AI anymore, and that AI belongs to the big guys, Google and Microsoft, etc.. They will provide funding to use their infrastructure, but not to build something new. I can’t say it’s a problem across the board, but there are some significant programs that think that way. I think the problem is mainly inexperience. I have been called a “ChatGPT wrapper” by someone that has never written code.
The second issue is that Canada was maybe one of the biggest contributors to the AI technology to begin with, with the work of Geoffrey Hinton, and we’re amongst the first nations that invested heavily into AI, but the base of the LLM, but the first version of a transformer, came from Google Translate.
Even though we had a strong foundation, we still do not own any of the LLM infrastructure. We need local LLMs infrastructure to avoid sovereignty issues if the government, if the military, if anybody wants to build on these AI systems. They should not be required to run them on foreign hosted LLM platforms. I think that capability is coming.
We neglected local infrastructure to the point that the majority of Canadians are using foreign cloud infrastructure. Now that we’re trying to catch up, I’m hearing that for the next 10 years, we won’t be able to launch any defined data centre project because we don’t have enough power for that. Shortage of energy and resources and infrastructure is impacting our adoption of AI.
Even for the government, most of our email and baseline cloud services are actually running on American servers. Everybody’s either on Gmail or MS Office. We’re training their LLM models with our own data no matter how we look at it. That’s another big bottleneck that will take a long time to catch up, because then we need the local infrastructure, all of those baseline cloud tools, and then finally LLM training.
My focus is on thought leadership around our lack of confidence as a nation. I think we’re always looking out to our counterparts in the United States and elsewhere, and I think that we should try to gain the self-confidence that we can build globally scalable companies just like our counterparts.
The last piece relates to an old mentality when it comes to VC funding. The old SaaS model is dying off, and we’re waking up to this reality too late. When you look at some of the thought leaders in our ecosystem, we’re still basically saying that we should fix the scale-up challenge here, forgetting that we have massive amounts of intellectual property that we have to commercialize. We’re still betting on growing big companies, whereas you see companies like Replit or Lovable, in the span of eight to 10 months, growing from zero revenue to hundreds of millions of dollars of revenue. We’re neglecting that. We’re again missing the boat on AI because we’re not realizing you can actually move things a lot faster, and instead we’re trying to double down on companies that grew on a decade-old technology SaaS mindset.
I think that’s the wrong mentality. You need both. You need to pay attention to the early stage and you need to pay attention to the later stage all at the same time. You need to identify the companies that can actually have a global impact and then if they have all of the ingredients of success, invest in them and make sure you can actually grow those companies. But in terms of the talent pool, know-how, and the market, we have all of that. We have one of the best Ai talent pools because of entities like Mila and Vector, but we are not properly leveraging them.
For Codalio, I can tell you that a year before Lovable existed, we had the same pitch, but nobody believed us, and they still don’t.
KB: All of this should be feeding into Canada’s AI strategy. Are the issues you raise here reflected in that report?
EM: We engaged with the AI strategy talks back in 2023 through the Canadian Council of Innovators. They basically were, in my opinion, instrumental in putting together a bunch of CEOs that had some really sound ideas around AI and AI technology. We liaised obviously with the government officials around the program, and we ended up submitting “A Roadmap for Responsible AI Leadership in Canada”.
We put forward 10 recommendations in total. The core idea was to build out a tiered approach — one that could actually work across a wide range of use cases, rather than a one-size-fits-all kind of policy. So on one end, you have areas where you want to actively encourage and promote AI adoption, give people the tools, the guidelines, the frameworks they need to do it responsibly. And then on the other end, you have higher-risk use cases where you need actual regulation — clear rules around how AI gets used, what the potential impacts are, and who’s accountable when things go wrong. The goal was really to match the level of oversight to the level of risk, so that innovation isn’t stifled where it doesn’t need to be, but there are guardrails in place where they genuinely matter.
We wanted to have rules and standards that innovators should comply with, regulatory sandboxes and so on. A new entity was created just in November 2024, so I think the government actually listened and reacted properly.
KB: Having seen the pan-Canadian AI strategy, are there issues that you have been pushing on in your own thought leadership that were missed in that report? What do you think that strategy gets right, and where is there still work to be done?
EM: I think the strategy has had some great success in these areas:
Talent Magnetism: It successfully turned Canada into a global “brain magnet.” By funding CIFAR and the three national institutes (Amii, Mila, Vector), Canada secured a first-mover advantage in AI research.
Ethical Leadership: Canada was one of the first to prioritize “Responsible AI.” This has given us a “brand” of trust that is valuable in a global market wary of black-box algorithms.
Ecosystem Clustering: Moving from research to commercialization shows the government understands that “papers published” don’t equal “GDP growth.”
Talent magnetism has been very important for us as a nation. It has the potential - or already has - to reverse the brain drain. The collaborative research projects undertaken by these institutions are a great way to train and retain talent. Engineers that were to receive world class training in real use cases didn’t have to fly to Palo Alto or go to other countries. I think the global credibility that we created through those entities was just amazing.
Having said all of that, there has not been much of a focus around collaboration with the startups or even supporting the startup mindset that could quickly commercialize some of the great IP they produce. In early 2026, though, they pivoted toward venture scientists, and Mila launched the Venture Scientist Fund in partnership with Inovia Capital. They finally recognized that research for research’s sake isn’t enough to sustain the national economy.
The accessibility of Sovereign AI Compute infrastructure remains a bottleneck for companies and researchers alike. The federal government’s introduction of the $300M Canadian Sovereign AI Compute Strategy was a significant step forward; its “heavily oversubscribed” status within just a few months is a clear testament to the unmet demand in the ecosystem.
While early federal investments have helped some companies build impressive GPU clusters, these entities often prioritize academic and research-based outcomes. As a result, we are doubling down on pure research rather than the rigorous commercial MVP testing required to scale. Currently, demand is far outstripping supply, making a robust Sovereign AI Strategy the only way to prevent our most promising firms from migrating to the U.S.
Canada has a long history of supporting the science of AI through programs like SR&ED and IRAP, as well as world-class hubs like Mila, Vector, and Amii. These institutes provide invaluable validation for deep-tech R&D and early-stage use cases. However, there is a disconnect between scientific validation and market readiness.
Under our current structures, it is difficult to imagine a company like Anthropic emerging from Canada. While we excel at the “Deep Tech” phase, we lack the infrastructure to support the transition to market. To bridge this, we must:
Shift Focus to Commercialization: Move beyond academic validation and prioritize how complex AI technologies reach the market.
Fund the Development Gap: We need to double down on pre-traction funding, specifically for high-tech ventures.
Support the Long Tail of R&D: AI companies often require extensive development cycles after the research phase but before generating revenue.
Without a dedicated focus on the commercial “middle ground,” we risk being a nation that discovers the future but never owns it.
KB: What does a junior developer role look like in a world where well-architected code can be produced from an AI system? Conversely, what do AI systems look like if humans stop writing the code that trains them?
EM: I don’t think it’s going to be that different. I’ll give you some examples from the dot-com era. Early on, when HTML and CSS came out, people were just editing HTML and CSS in Dreamweaver. Then WordPress came, and now they’re doing drag-and-drops. Then Wix and Squarespace came and made it even easier. The number of web developers did not go down. It actually went up.
Tools like this help you to augment your process and abstract away the manual work, but you still have that last mile. For websites you have to add tags and forms, connect to other tools, etc.. You’re just shifting the work higher into the abstraction layer.
I think AI will drive the same kind of shift. But, for anybody to really be able to be good at what they do, I think they need to really understand the fundamentals of those other abstraction layers. As LLMs improve—and I have no doubt today, based on what we’re seeing in our own internal data—I think the LLM is going to be able to do 90-95% of our future coding. But that last mile still requires expertise.
Because of the fact that we are now going to be doing more of the development, we can accelerate digitization of industries that haven’t had that much penetration. You can actually expand into many different environments, and that will also require experts. I think projected numbers are showing that we’re going to move from 28 million software developers to about 40 million in the next 10 years.
The good thing that is going to happen there is that even if you don’t know everything, you can use an LLM just like a senior developer or a CTO to guide you. If we try to extrapolate out maybe 20, 30 years from now, I think it’s evident that a lot of things are going to be automated. But you still need that layer—I want to call it an “intent layer”—that requires human intent as in input to AI. That intent layer is going to be the realm that everybody or at least new businesses are going to be operating on. If you want to win this race, try to see how you can win in that intent layer.
Key Takeaways
On Canada’s innovation pipeline
When we seek to build a system that requires multiple sequential successes for overall success, we must support every stage appropriately. If we do not, then what we build will only be as strong as the weakest link in the chain, a chain that spans years of work, often starts with tech transfer from universities, and ends with scaled domestic anchor companies. Ehsan has been involved in every part of it.
His experience with tech transfer highlights core challenges with the Canadian tech transfer process that I and others have tried to surface in related debates: the need for validation before filing and funding, at a stage when this is often premature.
I won’t recreate Ehsan’s entire description, except to note that his story is far from unique. Ehsan’s experience supports an idea that comes up again and again on this site: where emerging technology is concerned, it is too early to pick winners, and we are better served by broadly enabling innovators to build than attempting to eliminate the risk of failure before we start.
In fairness, this is improving in general. Many universities have now established or are establishing investment funds focused on research commercialization, several of which use venture philanthropy as the basis for their approach. While streamlined and consistent tech transfer remains a bottleneck, the conversation has moved significantly in the right direction in the past few years.
Ehsan’s work with the Archangel Network of Funds demonstrates how founders recycle capital and expertise into the startup ecosystem. From his early success, he has gone on to be an investor or manage funds that invested in more than a hundred companies, and has been a mentor for many more.
His experience has important implications for the debate in the Canadian ecosystem about where funding should be allocated in the innovation pipeline. Ehsan’s answer is simple:
“When you look at some of the thought leaders in our ecosystem, we’re still basically saying that we should fix the scale-up challenge here, forgetting that we have massive amounts of intellectual property that we have to commercialize. […] You need both. You need to pay attention to the early stage and you need to pay attention to the later stage all at the same time.”
Without a strong pipeline of effective tech transfer, we miss out on potentially valuable startups created with the IP that results from publicly funded research. Without a strong early-stage investment ecosystem, there’s nothing to scale up. Without scale-up capital, companies will not get past the startup stage or arrive at that stage disadvantaged. This creates a self-fulfilling prophecy of under-performance that reinforces the very issues that cause it.
On building with AI
While Canada is too late and too small to compete with the hyperscalers, this is not to say that we cannot capture value.
Ehsan pinpoints where Canada can compete, suggesting that we “can win in that intent layer.” Just as the emergence of the infrastructure that enabled Silicon Valley gave way to value creation on the application layer, and just as that same shift will happen with quantum computing, so too with AI, where successful application depends on accurately capturing intent.
In my view, there are two primary classes of problems to which an LLM can be effectively applied and where it will be possible to capture value in the intent layer:
those whose solutions would take a human a long time to produce but that would take an expert only a short time to verify correctness; and
those that do not necessarily have a “correct” answer, but rather that have many useful answers from which an expert can make an effective choice.
Critically, AI works for neither class of problem without expertise.
A good example of the first class of problem is code generation. Creating good code takes time, but (for relatively simple snippets) it is relatively simple and straightforward for someone who knows what they are doing to test if it works as intended.
A good example of the second class of problem is choosing a toolchain to use in developing something new. There’s no objectively “right” answer, and you can make anything work if you are determined enough, but understanding the options available and being able to make an educated choice between them is non-trivial.
Codalio’s approach hits both classes of problem.
“[I]f you do not have the experience and the expertise to understand how the software should be architected, you can only go so far […] [W]e needed a major new category that could actually get a non-technical person from zero to 80 or 90%, with the last mile going back to the developer.”
Regardless of what we build with AI, however, a nation cannot fully secure the benefits of success if it relies on foreign cloud providers and servers. These are not built overnight:
“We neglected local infrastructure to the point that the majority of Canadians are using foreign cloud infrastructure. Now that we’re trying to catch up, I’m hearing that for the next 10 years, we won’t be able to launch any defined data centre project because we don’t have enough power for that. Shortage of energy and resources and infrastructure is impacting our adoption of AI.”
This reliance creates sovereign data risks across both the private and public sectors:
“Everybody’s either on Gmail or MS Office. We’re training their LLM models with our own data no matter how we look at it. That’s another big bottleneck that will take a long time to catch up, because then we need the local infrastructure, all of those baseline cloud tools, and then finally LLM training.”
For policymakers, the key takeaway is that energy, technology, innovation, and industrial policies are no longer things that we can keep in siloes. Addressing this requires whole of government leadership that cuts across multiple files, a subject that I have explored in more detail in previous interviews and articles.
On the path forward
Ehsan’s career is a firsthand exploration of a fundamental policy question for Canadian competitiveness:
“…why were our American counterparts able to scale similar companies to billions of dollars, but we were only able to get to millions?”
Despite seeing the problem, Ehsan actively chose to build in Canada, giving up a clear path to U.S. permanent residency:
“I had just a very short amount of time, maybe a few months, before I could have transitioned my H1B to a Green Card. I made that choice that I didn’t want to do that and I made the deliberate decision to come back to Canada.”
Unfortunately, his story is more an exception than the rule, as there are more Canadian-founded unicorns in the US than there are in Canada. This state of affairs need not persist. On America’s present trajectory, Canada’s attractiveness as a place to build will only increase relative to our southern neighbours, but to capitalize on it, we will have to address the frictions that plague those trying to do it.
Rebuilding and streamlining the Startup Visa program should be high on the priority list. Just as Canada has a generational opportunity to attract scientific talent fleeing the United States, so too do we have a generational opportunity to attract talented entrepreneurs and innovators through immigration; people who, like Ehsan, make an active choice as to where they build and who commit to the prosperity of the place they choose to call home.
Ehsan’s thoughts on what Canada needs as a nation are grounded in first-hand experience of almost every part of the innovation ecosystem:
“My focus is on thought leadership around our lack of confidence as a nation. I think we’re always looking out to our counterparts in the United States and elsewhere, and I think that we should try to gain the self-confidence that we can build globally scalable companies just like our counterparts. […] For Codalio, I can tell you that a year before Lovable existed, we had the same pitch, but nobody believed us, and they still don’t.”
I hope that by highlighting the stories of talented builders across the country, I can contribute in some small way to building that confidence. Many thanks to Ehsan for taking the time to share his insights.
If you know someone you think should be celebrated for their contributions to Canada’s innovation ecosystem, please tell me how best to get in touch with them.