Haven’t heard? Artificial Intelligence (AI) has eaten the world – or at least the majority of capital available to the world. Q4 2024 was another record quarter for AI investment with 60% of all VC deals focused on the much hyped technology. Of course, the number of companies on the receiving end of 2024’s $180bn VC boon tells a slightly different story; a story in which this chapter’s climax also took place this week with Open Ai’s release of ‘Operator’.
Meet OpenAI’s ‘Operator’. Operator is what’s known as an ‘agent’ in the AI space, and is what is widely agreed to be the first milestone of real value to be produced from the Ai hype-cycle.
Until now, AI Large Language Models (LLMs), and the chat bot interfaces that help train them, have been fun tools to play with; cool oddities to show my 7 year old via Chat GPT, Grok, and image generators like DeepAI. Largely they can be helpful as assistants to parse through large amounts of information and return succinct analysis – but in terms of real business value, most feel that the first mile-marker on the road of true disruption begins with Agents.
Agents are a system or program that is capable of autonomously performing tasks on behalf of a user. Hypothetically they can book your flight, order your meal, or perform any series of tasks based on your needs/preferences. In short, they will change your life and mine. How that exactly happens is yet to be fully seen, but it is unfolding right in front of our eyes.
One variable that’s undeniable in this evolution is scale. More data, more processing capability, more power – resources are the fuel that ultimately feeds AI models. That’s why you see so much capital being deployed into relatively few organizations. The predominant bet is that similar to what has happened with web and search infrastructure, AI LLMs and Agents will ultimately be offered by a very small handful of very big providers as a ‘backbone’ of sorts. Open AI has been a very visible front runner in this race, however the web is a great example (thanks Yahoo, Cisco, etc.) that the first to scale isn’t always the ultimate winner.
Beyond infrastructure scale, data is the other key input in achieving the lofty promises of AI. Early users of Operator are already exposing the natural data gaps that come with training models. Not only does AI need to know ‘how’ to do something, it needs to know ‘who’ it’s serving in order to complete the task effectively. Anyone who has trained an exec assistant in the real world, knows this pain. You can read up on how this manifests here as one user attempted to order groceries with the budding tool.
Context is king, and that fundamental maxim fuels our investment strategy and fundamental outlook on AI. I think of horizontal AI, or AI that serves common needs of fundamental disciplines as the table stakes of this new world. Soon every sales professional will have their own personal super agent who researches prospects, creates a custom strategy to engage them and executes on that game plan to secure a meeting. A lot of folks like 1Mind, whose CEO Amanda Kahlow we had on the podcast this last season, believe AI will go even further, conducting discovery and providing demos of software. I’m not sure buyers will be comfortable speaking with a bot for considered purchases, but as with all things AI… eventually the singularity is coming.
Extending agents into specialty tasks is extremely difficult to do at scale, and therefore this is the area of opportunity we feel will be filled by vertically-focused SaaS providers. Purpose built agents powered by access to bespoke data and context necessary to service needs beyond ubiquitous tasks like buying groceries, instead tackling aspects like leasing an apartment (portco: NurtureBoss) or booking orthodontic appointments for multiple children (portco: Peerlogic). These are complex activities that have deeper risk than simply having your pizza show up to the wrong state, and the only way to ensure accuracy of the models that power these activities is a lot of data, curated over time, and structured in intentional ways.
AI models can do a lot of things, but only when they’ve been taught to do so, and therefore the holder of experience has the unique ability to teach. We continue to look for founders and startups who are capturing unique and valuable datasets, who deeply understand the behaviors of a specialized market. We also see enormous value in those who can make existing data more accessible, and therefore more valuable, to the agents of the future. Search as we know it (google search) was shot this week. It will take some time for it to bleed out, but make no mistake it’s a dead man walking. Therefore all of the information we’ve all worked so hard to serve up on the web to buyers now has a new consumer – an autonomous agent that will seek out and judge our relevancy before any human ever lays eyes on it (if they ever do). That’s just a hint of the incredible shifts in store for us all. The agent of change is here – and in this new world, we serve them, or get left behind.
