AI is all anyone can talk about this year, but it can be tough to determine what’s merely headline fodder vs. what’s actually happening on the ground. In our latest survey of fellow software CEOs, we dug into the details of what’s happening inside companies when the AI rubber hits the road.
Not surprisingly, a majority of CEOs in our network reported that AI is having a significant impact on the direction of their product, and they are committing both talent and dollars to building AI capabilities. The two main themes that emerged from our CEOs were the sheer dominance of incumbent platforms and how they’re already starting to see the second order effects of this tidal wave shift in tech.
Power of the Incumbents:
The massive flow of investment dollars into new AI-native platforms - whether those are cloud providers or one of the cohort of foundational models and LLMs - indicates investor optimism that new entrants in these segments can take substantial market share on the AI playing field. Maybe that will be the case one day, but it’s not showing up in our data today.
Instead, we found that despite the ample capital flow and tremendous ink spilled on AI up-and-comers, we are still seeing incumbents dominate the scene. For example, almost 80% of CEOs in our survey that are using or plan to use pre-trained models are going with OpenAI as their vendor of choice. We can consider OpenAI an incumbent in its own right, but it is of course also backed by OG-incumbent Microsoft.
For CEOs training custom models, over half are relying on AWS as their cloud platform. GCP and Azure were both pretty far behind, but the three tech giants together dominate cloud provider market share. Any AI-native cloud platforms such as Lambda and Mosaic were barely represented in our sample.
We saw a similar trend around coding co-pilots. As you might guess, adoption of a coding co-pilot was wildly popular, with over 70% of engineering teams already using one and another 11% in the process of evaluating. Among those who have adopted a co-pilot already, the winner in a landslide at over 86% is GitHub (hello again, Microsoft).
Largely this dominance sounds well-deserved: over 70% of the teams using GitHub Co-Pilot arrived there as the result of a bottom-up decision based on engineers’ love for the product or the conclusion that it simply works the best. Another 16%, however, ended up with GitHub by default - either because it was viewed as the incumbent or the team was somehow already using it. We expected GitHub to be popular but were actually surprised that none of the CEOs in our survey reported using Replit or other co-pilot tools.
We do believe there will be room for a few key players in each of the large AI market segments (LLMs, cloud providers, co-pilots, etc.). But what’s clear to us for now is that the incumbents will be hard to dislodge, and those upstarts that want a real shot at becoming one of the oligopoly winners need to have either a very well stocked war chest or strong technological product edge (if we’re sticking to Peter Thiel’s rule of thumb, a proprietary technology that’s at least 10x better than the closest substitute) – or both.
Second Order Effects: What Does AI Adoption Mean for Headcount?
One of the key questions on everyone’s mind - from software engineers to bank tellers to journalists - is what the adoption of AI tools will mean for their job security. We will refrain from speculating on what might happen across the broader economy, but we are already seeing how these tools (specifically coding co-pilots) are beginning to affect hiring plans within our part of the universe.
In good news for current employees, most CEOs (80% of our survey sample) don’t report making changes to their engineering headcount based on efficiencies gained from using a co-pilot. We think this could signal that they’re capitalizing on the increased productivity to move faster, rather than reducing headcount and saving costs to keep output steady with where they were before AI tools.
However, a solid handful (16%) shared that use of a coding co-pilot has reduced their future hiring needs. Adoption of co-pilots is already quite high, but we believe we’re just at the beginning of the journey exploring what they can do and how they might increase productivity and add value. With time, we expect to see more companies leveraging these efficiencies through lower headcount and for that 16% to rise.
What’s Next?
More broadly, we anticipate that the landscape will continue to shift and more of these second order effects will materialize as adoption of all AI tools (not just coding co-pilots) continues to increase. We’re seeing this in the data already. For example, even some of the CEOs in our survey who reported they are not currently planning to integrate LLMs into their product have not fully ruled it out for the future. 30% of the admittedly small number of non-adopters shared that it was due to a straightforward lack of bandwidth.
Others who had concerns about relevance or value-add commented that integrating LLMs might become relevant down the road, potentially with more work to validate product fit or show higher ROI. No matter which way you slice it, we see pent-up demand that will only drive adoption higher from here.
Additionally, more value-add use cases for AI will become relevant (indeed, necessary) with greater adoption. Even though incumbents are currently dominating LLM and cloud offerings, we see opportunities for startups to step in and address those demands where the incumbents are not well suited or don’t already have a built-in advantage.
For example, we expect co-pilots to emerge for other functional areas besides engineering but in new forms that fit the tasks at hand. One potential incarnation could take the shape of tools that simplify and reduce workloads for various types of operations functions. By “operations” we mean tasks that currently require human intervention to connect discrete software endpoints. These tasks span various functions within a business (HR, sales, marketing, legal, etc.), and new tools or platforms could replace operationally-intensive functions with AI-enabled connective tissue.
Another area is related to quality control after implementation of AI tools. Yes, AI enables us to do things faster and perhaps better on average, but it is by no means perfect. As adoption of AI tools grows - and in tandem, the average experience level of the marginal new user declines - it will become more important to have checks in place to ensure AI mistakes don’t derail the progress or undermine the hard work of the engineering teams using them.
To this point, Microsoft’s Scott Guthrie estimated during a session at the Morgan Stanley TMT conference in March that AI-generated and unmodified code represents north of 40% of the code uploaded on GitHub. While he was referring to that as a win for productivity (and we agree), we also see it as an opportunity for value-add tools to ensure quality and avoid potential pitfalls.
Finally, we envision opportunities for AI-enabled applications where incumbents do not have a data advantage. This could apply where a new entrant is leveraging data that was previously considered useless, where customer data is hidden in walled gardens, or where there is sensitivity to public use of specific data sets (for example in healthcare or financial applications).
From the Flex portfolio, we already have numerous examples of how companies are putting AI to work in these value-add use cases. To name only a few, e-commerce helpdesk platform Gorgias is using AI tools to automate tickets and workflows, enabling faster customer responses. Regal.io has introduced AI-powered SMS conversations with customers for quicker lead generation. And AngelList is tapping AI tools for a portfolio analyzer that enables data extraction from documents in client email.
By now there’s no doubt AI tools are a crucial factor for CEOs and for us as investors. Although in our view AI represents more of a sustaining technology rather than a platform shift, we see clear opportunities for new, independent players and applications. Whether we’re tracking the progress of incumbents or determining where we think startups can gain share, we’ll continue to stay on the pulse and share insights from our CEO network along the way.