Note: This essay only represents my personal views.
For the past 8 weeks, I’ve been working as a researcher/engineer at a new startup, Fault Line AI. It’s been a blast; so much so, I thought it fitting to write up a brief investment-style memo on the company.
What is an investment memo? In a past life, I briefly worked at an Australian venture capital firm called AirTree. Before deciding to invest in a startup, the team would research the industry, market, and team’s background, and collate this information in a memo. The memo would be opinionated, and argue why this company is worth throwing a couple $M at.
Although IMs are typically written for other investors or funders of the firm, I think this line of thinking can also be useful for early employees when evaluating whether or not to join a startup. Given that most of them fail, you need to have a strong conviction as to why it’s worth spending the next few years of your life on one opportunity.
Here are 3 questions worth asking when you’re considering joining an early stage startup, and why I’m excited about Fault Line.
—
Why does Fault Line exist?
Every company exists to make money. Some companies also mould a particular future in service of values, through economic incentives.
Fault Line is an example of the latter. They exist to architect incentives for safer clinical AI, by building safety testing software for AI companies.
Why is this important, and why now? With every new breakthrough in technology or science, there is more opportunity for entrepreneurs, and unforeseen risks for regulators and institutions.
In 1937, an antibiotic elixir, sulfanilamide, killed 100 people across 15 states. It was this disaster which led to the political impetus to expand FDA powers - marking the beginning of pre-market drug regulation. Some say that it was this early regulation which prevented a worse tragedy from ravaging the US 20 years after, where thalidomide caused thousands of infant limb deformities across the globe. It was then in the aftermath in 1962, the foundations of the modern clinical trial apparatus were created.
AI - specifically the Large Language Models that power chatbots - has unleashed pandemonium for regulators. The promise of infinite medical superintelligence in your pocket comes with its risks. Last year in particular, countless case reports of AI-assisted suicides, anecdotes of AI-induced psychosis, and concerns around AI hallucinations took the media by storm.
Although the true prevalence and nature of these concerns is up for debate and will reveal themselves in time, regulation has a role to play now. Do you wait for the thalidomide harms to occur, or try to enable some form of prevention?
Frontier AI companies recognise this issue, and many have in-house safety teams. Many clinical AI companies are well-intentioned but have no clear guidance on what the standards are, and instead, are left to regulate themselves. Is this the right incentive structure? We are in the wild west of clinical LLMs.
Many prominent figures (including Demis Hassabis, Anthropic co-founders) have called out the need of third party auditors for these systems - companies or institutions that have in-house specialty, and divorced from specific labs.
Fault Line (FL) is an example of a company which answers this call.
What makes the FL team uniquely suited to tackle this problem?
With pre-seed/seed companies, there often is no revenue and little product to evaluate. Instead, investors gauge the mettle and narrative of the founders - after all, practically all of them fail. So, you need to believe this team is different: What do they do that is world-class? What unique insights have they earned about the industry? What makes them able to work as a team?
The FL trio (Max, Ben, George) hit every one of the typical signals:
Ex-founders who’ve worked together. George and Max have raised seed rounds for previous ventures, and understand early-stage startups. The trio previously were all at the same company, which is a great sign of co-founder fit. Second, or third-time founders are always easier to invest in, since if they’ve gone once through the fire and survived, they’ll be able to go again.
Industry-specific insights. They faced similar problems during their work in clinical AI - shipping out the first regulated AI chatbot in the world, with Ben spearheading the mechanics behind healthcare procurements across the UK/US.
Regulatory and scientific credibility. Building a company like this requires deep credibility and expertise to execute to differentiate against all the emerging startups in the space. Max’s research has been published in journals like Nature, Frontiers, and has the scientific pedigree from his computational psychiatry PhD. It helps to have a team of PhDs in a technically challenging space!
Of course, a strong team fit only gives you higher conviction that they can execute well, not certainty. Businesses fail for plenty of reasons outside of what we can control. Hence, early teams need both a convincing narrative and the subsequent revenue-based numbers for future rounds.
Can this become a venture-scale business?
Not all problems need venture-shaped solutions, and not all businesses are suitable for venture.
Venture is useful for a young company with big ambitions; it’s the classic David vs Goliath problem. If you’re tackling a big problem like building/inventing the new infrastructure for regulating clinical AI (and beyond), investment is necessary for a) investing in R&D for core technology, b) demonstrating seriousness and credibility to institutions, and c) having resources to move faster against competitors.
However, VC needs to believe your company can achieve significant returns. They run on the doctrine of the power law; the idea that you take many small bets on companies, expect most of them to fail, and with one or two being able to return the fund many times over. The starting points of business models do not matter - things often do not go to plan in startups, and the ability to quickly pivot to a new idea is what gives them an advantage over incumbents.
Because this is so unpredictable, you often rely on general heuristics on the market size: Is the long-term, best case scenario of this company able to achieve venture-scale returns?
For FL, you need to believe that there are enough clinical AI companies willing to pay enough. FL’s market size grows along with the growth of clinical AI companies - the bet here is that current clinical AI tangibly improves patient outcomes and encourages adoption, whilst future agentic clinical superintelligence opens even larger markets.
Then, you need to believe that FL can overcome the specific classes of risk relevant for a particular startup/problem:
Product-Market Risk: Does FL provide significant value for clinical AI companies? Clinical AI companies have a myriad of competing interests at any point in time: how to improve their product efficacy, integrate into different hospital systems/softwares, and expand into new domains. FL needs to be easy to use for engineers/clinicians, whilst also improving companies’ credibility to their users (patients, clinicians, and hospital systems).
Regulatory Risk: Will regulation create a durable role for third-party clinical AI evaluation? The only thing that differentiates one regulatory evaluation product from another is its ability to truly predict outcomes (or best proxies) of clinical harm. Everything else is marketing dressed up in regulatory clothing. FL needs to prove to regulators and the wider community that it is serious in its science and clinical rigour, so that legislation and frameworks evolve in ways which support FL’s approach.
If these hypotheses hold true, then you can believe that FL succeeds. Given the structure of the team and their early traction, I’d place good odds on them.
—
Startups are both incredibly exciting and incredibly challenging. They are an opportunity to bring something new into the world, have autonomy in shaping a project, and work with like-minded people who share your values.
There’s plenty more pragmatic economic considerations than the ones listed here (equity packages, market sizing, competitor analyses, etc) but this essay details what I think is the most important: Narrative.
Jobs are a mix of pragmatics and purpose. If a startup fulfills your sense of purpose, and (after thinking through the investment economics) fits the pragmatics… then what are you waiting for?
—
Thanks to Alice Park, Matthew Shu, Shayan Lahijanian, Jackie Vullinghs for giving feedback and reading drafts.