AI, Thinking Beyond Incremental Improvements

Following an interview with Clarence Bethea where he prompted me to share thoughts on where we see the most potential for AI to transform not just workflows but technology and infrastructure at its core, we’ve been thinking about where the most opportunities lie for founders and what’s fundamentally redefining our expectations of what’s possible.

At the core of that discussion is the real value proposition of AI; not marginal improvements to our existing workflows but rather solutions that create and operate AI-reasoned workflows that are different from scenarios where humans are typing and clicking away at software on their screens. In other words, a 10-20% productivity improvement is not as relevant as a wholesale replacement of prior software workflows as well as associated staff time on that software application.

The real magic happens when AI enables entirely new possibilities – where AI software initially augments and then eventually replaces software and laborious work altogether. Having an AI review a meeting you couldn’t attend and engaging in a natural conversation about what transpired is one thing. But deploying 10,000+ software programmers or IT experts with a single line of text to build and operate complex systems is quite another. 

To that end, the stakes for AI startups as well as startups looking to work better and smarter with AI are about to get that much higher. If you can harness AI to do the work of 100+ engineers on-demand in three weeks instead of 20 engineers in three years, that’s the kind of transformation that could really change the world. And it doesn’t need to be perfect. But it would be transformative. Much like with the tipping point of robotics, this doesn’t mean jobs are taken but rather many roles and skill sets will transform into entirely new ones. 

One frame of thinking is that the human touch will remain crucial at both ends of the AI spectrum. The initial 1% –  the creative spark and direction – will always need human input. The final implementation and oversight will also require human judgment, curation, and sound strategy. It is the previously inefficient middle 98% where AI can deliver radical transformation. 

What does this all mean for investors? Well, the velocity of innovation will require us to evolve faster than the startup community we serve. For example, our learned patterns of historic metrics for the number of people building a tech startup has remained pretty consistent for the past two decades. Now, a small group of AI-native innovators can build, deliver, and operate the equivalent of 200-2,000 engineers within two years.

Do we value it with previous frameworks for investment dollars, valuation, and time/scale to the next fundraise? Of course not. For a startup seeking to replace the output of 10,000 engineers a year at upward of $100 thousand a year in salary per engineer, it would mean approximately $1 billion a year of replaced cost and at a 50% discount, $500 million a year of revenue. 

We will see companies race to unprecedented revenue, margins, and scale. And even at 10,000 engineers, the only real inhibitor to further scale (to 100,000 engineers or even 1 million engineers) is infrastructure in computing, networking, and storage – fundamentals that we know how to build and recursively, most of that infrastructure can be designed, orchestrated, and operated with armies of AI agents. 

Perhaps the only inhibitor to this unprecedented scale is the pace with which we can build data centers and connect (hopefully green) sufficient power to them. Specifically, one way early stage venture capital may evolve is trying to understand the sheer pace of change in this exciting field. Not a week passes without significant new advances in scale, training, and new models and capabilities. How we understand this new way of computing integrated with human and perhaps superhuman reasoning is by being ‘in it’. 

If this is not like any prior tech wave – and all signs so far point to that – ‘rational design’ of investing strategy and execution likely won’t work. We must learn here by ‘doing’ and engaging as much as we can while suspending our patterns developed in the last few tech waves. There is also a component that doesn’t change — that of the founder mindset that takes on an amazing array of challenges and the founder skillset that zeroes in on innovation. If we’re not relevant to this new class of founders and startups, we shouldn’t be in this business.

Rohit Sharma is a Partner at True. As a former founder and CTO of ONI Systems, he helped pioneer optical switching technologies and led the company through its IPO and acquisition. Now, he’s deeply passionate about GenAI and its transformative potential. Once you’re backed by True, Rohit’s one of the many experts available to you.