“The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x. I need to recalibrate.”
Kent Beck, prodigious programmer 1
Placing her cup of morning coffee on her desk, Sarah logged in to her desktop and fired up her browser, navigating to the front page of the Financial Times. As she skimmed across the page, an article caught her attention – the EU parliament had just approved Markets in Crypto Assets (MiCA), a uniform legal framework for crypto asset markets in the EU, and had taken significant steps towards finalising DAC8, legislation covering the reporting and exchange of information for tax purposes.
As a specialist FinTech and Digital Asset venture capital investor she intuitively understood these were significant developments that presented a sizeable opportunity for start-ups and investors alike. Sarah immediately sent her colleague Leo a message asking him to summarise the key elements of the proposed regulation and its impact on all relevant stakeholders, identify the size of the revenue opportunity and to provide a screen of EU based start-ups operating at the intersection of data, compliance, RegTech and digital assets. She was enthused and eager to speak to the best teams operating in this space as soon as possible.
This may sound like a mundane or quotidian description of a day in the life of a VC but consider now that Leo is not a VC analyst but rather an autonomous AI agent set up by Sarah via alphakit to manage a team of other goal-driven autonomous AI agents:
- an agent directed to use Upword to summarise and extract the key components of the EU legislation,
- one to use aomni to automate research related to the impact on the relevant stakeholders and the potential revenue opportunity,
- another to plug-in to databases like LinkedIn or StartupRadar to screen and identify similar start-ups based on a set of learned criteria,
- and to use Backengine to build the API connectivity (without writing any code) to directly input the start-ups into the CRM combined with one page auto-generated summaries for each…and so on. 2
The tools highlighted above relate to the investment sourcing and evaluation workflows, but this is broadly applicable across the VC workflow stack and similar tools can be stitched together for other functions within investment management teams, whether it be portfolio reporting, monitoring, investor relations or fund admin/accounting. For example, legal diligence and execution are already being disrupted with the likes of Dori, which allows for the instant generation of closing documentation based on a term sheet and promises simplification of the process by transforming complex legal language into easy-to-understand vocabulary.
If your curiosity was sufficiently piqued to explore the embedded links above, it should be evident that this is not just a fantastical thought experiment. This is our new reality, as the recent advances in AI automate workflows in previously unimagined ways, ushering in a new industrial revolution (a topic that was also briefly discussed in one of our prior GP letters). Having absorbed the impact of this technology on the venture capital profession, it does not take much to stretch one’s imagination to grasp the potential for disruption and opportunity that these tools offer across other professions.
Several questions arise – what does AI-led venture capital (or any other white-collar profession) look like? Which skills and capabilities have historically been valuable? Which of these can now be fully automated and are therefore redundant? Which can be augmented and enhanced? For which does it free up time? For which does value persist, with opportunity for substantial leverage?
We believe the future of VC is very likely to be leaner, more automated, and highly strategic. Or as our Managing Partner, Pascal Bouvier, has previously observed, it becomes more about the why, what, where and when vs the how? Value is created by asking the most insightful and probing questions or framing questions in such a way as to elicit the most appropriate and comprehensive responses. In a world where Large Language Models (LLMs) have proliferated, leveraging AI becomes less of a one-way passage and more of an interactive journey dependent on the quality of user inputs (note the increase in career ChatGPT prompters!). Further, resource requirements will likely decrease for certain investment team workflows (and this will of course impact total required investment team FTEs). However, this should in theory also free up resources that can be invested into portfolio services/ value creation initiatives and personnel requirements associated with them, efforts previously somewhat limited to the large AUM gathering platforms. There will likely be a reallocation of spend on resources with the potential for smaller funds to compete with the larger funds on value-add initiatives. If workflows across the VC stack can be automated end-to-end, the commoditisation of venture capital is likely to accelerate. Managers with a strong heritage and brand (à la the Sequoias of the world) or deep domain expertise (by sector/ business model/ geo) are most likely to win (we would humbly submit that MiddleGame Ventures, with its FinTech and Digital Asset sector specialism, is well placed to benefit from this paradigm shift).
A natural extension of this discussion is to ponder what this implies for the future of start-up creation and the nature of early stage investing itself. What does it take for founders to launch a venture today vs a few years ago? What are the capital needs of start-ups as they scale? How does this impact venture capital funds and venture investing?
There could be scope for dramatic gains in cost efficiency with some heralding the recent advances in AI as another AWS/ cloud moment for start-ups by removing the capital friction to launch a viable product. AWS and APIs drastically reduced the amount of capital needed to get a start-up from ideation to technology to product and some say all the way up to product/ market fit. However, scaling to unicorn status and beyond has remained very capital intensive. It could be the case that a quintessential pre-seed or seed stage start-up no longer consists of a handful of engineers in a garage or basement but rather an engineer and a team of autonomous AI agents. Start-up creation could now be cheaper and more efficient with less required investment to get to an MVP and establish product/ market/ sales fit.
In contrast to the advent of cloud computing, which led to lower initial costs for start-up creation but not the scaling aspect of growth, it may be the case – although it is too early to support this with any meaningful data – that the rise of LLMs and autonomous AI agents considerably reduce the cost of scaling post-start-up phase as well. Sales teams, sales processes, pipeline management, marketing, client onboarding and client management will become much more productive and automated (as with the Sarah example above).
This raises some important questions about how much capital a typical seed stage company may need. What will the size of an average financing round be and how many rounds will a typical start-up need? What does this in turn mean for both early stage and growth stage fund managers with respect to their own fundraising and deployment strategy? We believe that smaller, specialist funds are the future and the likely winners in this new environment.
While the industry is still collectively working through the answers to these questions, it is becoming apparent that start-ups and venture investing will both change.
And we need to recalibrate.
1 Kent Beck is a prominent figure in the software development community, known for his contributions to the field of software engineering and his role in the development of various software development methodologies. He is recognized as the creator of Extreme Programming (XP), one of the most well-known agile software development methodologies. He has also made significant contributions to the practice of software development through his work on agile methodologies, test-driven development (TDD), and design patterns.
2 This is an oversimplification, but the concept still holds. In theory a single autonomous agent can be used to work towards an overarching goal by breaking it down into a series of sub-tasks and could achieve the same outcome here. As a nascent technology, there continue to be occasional challenges with lapses in logical reasoning, recursive loop architecture leading to higher compute costs and lack of memory recall but given the rapid pace of iteration in this space, it is expected that these will be overcome in the short to medium term.
