Dec 8 2024
As the year draws to a close, I've outlined some of the ideas that have captured my attention recently. Some of this is a request for startups. Some of it is merely cleaning out space in my mind for hopefully better ideas.
At the horizon of AI progress, I think some business functions–rather than being automated–can be eliminated entirely.
What if the end-state of customer support software isn’t replacing humans with agents? What if instead, the future of customer support is just more powerful, more malleable software? Yes, I could call Delta or email Delta for some request (from simple tasks like flight changes to complex tasks like PNR splitting, bereavement fares etc…)... but I could also just get what I need from a more powerful version of the Delta app.
The whole premise of customer support is that it’s the backstop for activities that couldn’t be properly represented within the normal customer touchpoint (e.g., the app). At some point of AI progress, I think our software becomes sufficiently intelligent that we no longer require the backstop. The software–in more ways than one–would speak for itself.
A few directions of work that I’m interested in:
Here’s one vision for AI employees: we equip agents with an email address, a Slack account, and access to various tools. Use of those tools is mediated through APIs or where necessary, computer-use capabilities. Properly equipped with all the necessary tools and having ingested all relevant documents and data sources, the agent is like a remote employee: despite a lack of physical presence, you can discuss business problems and route various tasks for them to fulfill…
My prediction is that at some point, OpenAI releases an agent employee (I assume this is not Operator) product with a small set of high-profile partners (e.g., Linear, Atlassian, GitHub, Notion, Slack etc…). These partners would introduce agents as a new user type and conform to some standard protocol for agents that handles authentication, notifications, access control etc…. Agent employees would be able to receive task assignments, @-mentions. They could comment on your documents, produce their own, respond in-line to user feedback etc…
A few quick thoughts, questions:
I'm intrigued by the possibility that customers are best served by open source business systems. There are two main reasons for this:
Closed providers want to sell their own agent labor and will be disincentivized from platforming 3p agents. In contrast, open source systems would inherently be agent-neutral. Users of open source systems would be able to use whichever agents are best suited to their needs optimizing for quality, cost, control etc…
More and more companies are reporting a desire to build their own custom solutions rather than pay for expensive SaaS. In part this reflects the fact that it is getting drastically easier to build software. But this also reflects a long-standing desire for custom software. Open systems being forkable means that customers have flexibility to rework tools to conform to their unique requirements.