Even though we’ve been profitable for years (and as a result, we don’t have to orient our entire business around “what’s most likely to get funding in the next 6 months”), a question that comes up from candidates is something along the lines of “what about AI?” (or my favorite from a recent interview: “AI… question mark?”).
It’s easy to dismiss, but I think it’s a fair question – in technology, you may only see 1-2 major paradigm shifts in your entire career, and it’s important to be cognizant of them. “What about the internet?” and “what about mobile?” would have been good questions to ask at various points in the recent past, even if the view ahead is too blurry to formulate a more specific question.
In the case of Clipboard, there are roughly three ways that I think about AI (or really, LLMs and related tooling) today:
How does it affect our core business? What threats does it create for us?
How can we leverage LLMs in our product to deliver a better experience for customers?
How should LLMs affect the way we work? What does it mean to be an AI-enabled business?
I’ll tackle each individually.
How do LLMs affect our core business?
At the risk of sounding like Paul Krugman, the answer, thankfully, is not much. Our business is about connecting professionals to workplaces that need their specialized skills either on a short-term, flexible basis or more permanently. The types of professionals we work with are in-person, hands-on roles like Nurses, Teachers, and Dental Hygienists. At some point in the future, autonomous robots may be able to take over these tasks, but we’re not there yet, and that future doesn’t appear imminent. These tasks are also ones in which it will likely be difficult to ever remove the person from – they are caregiving tasks for which a human is (probably) essential.
So our core business, the thing that customers pay us to deliver, is unlikely to disappear in the near-term, even with significant continued advancements in the field (which are not guaranteed).
How can LLMs make our product better?
Today, our business has a massive amount of human labor behind the scenes to turn someone that signs up for the Clipboard Health app into a qualified professional for a shift at a specific workplace. We’ve started working on using LLMs to turn much of that human labor into automation, for example, by using LLMs to validate TB test results with a higher accuracy rate (and much faster) than humans. That means a smoother onboarding experience for our customers and a bigger pool of professionals that can work shifts.
Those applications of LLMs are a bit “easy”, and they also have a bigger impact on our margins than on the customer’s experience. While we care about margins, we care much more about the value we’re delivering to customers.
I’m currently more excited about LLMs as an interface for customers. There are two that we’re experimenting with right now. One is an SMS scheduling interface for job applicants for our Full-time Hiring customers so that they can do interactive scheduling without downloading yet another mobile app. We shipped this experimentally and discovered that we had unknowingly also shipped multi-lingual support when one of our users scheduled her interview in Spanish. The other is a natural language interface for setting the schedule for healthcare facilities – back when Clipboard Health was much smaller, our customers just told us what they needed over the phone and we handled it. We now have the opportunity to do something similar, and free up schedulers at healthcare facilities to do more important work than fiddle with our user interface.
How do LLMs affect how we work?
The last piece of the puzzle is how AI and LLMs affect Clipboard as technology consumers. The short answer is that we don’t yet know what the future looks like for AI-enabled companies, but we’re experimenting to try to figure it out.
We have an annual stipend for engineers to spend on any tools that make them productive, and the vast majority of our engineers spend that on some form of LLM tooling (e.g. Claude Code, ChatGPT, Cursor, etc). We’ve avoided large enterprise contracts – the space is evolving too quickly to be locked into one tool, so we want to let our engineers experiment and tell us what works best.
We’re using Devin.ai internally to enable small changes without an engineer. A few engineers have discovered ways to use it to spin up multiple parallel threads to make small changes (usually upgrades) with significantly less overhead. We’re now working with a few non-engineers (mostly Product Managers) to experiment with different tasks using Devin + an engineer as a reviewer to see what guidelines and guardrails we should put in place. So far, we’ve found that simple changes (i.e. ones that take more time to create a PR for than make the code change) and finding answers to questions are the best use cases, but we’re trying to see if we can add more complexity. One use case that I’m particularly excited about is “make field X available in table Y in Snowflake,” which sounds deceptively simple, making it perfect for AI – it requires very little in the way of judgment, but it’s a task today that requires a data engineer. By opening up this small piece to non-engineers, our data stack becomes more self-serve: data consumers can, on their own, start building new analysis without ever getting blocked on data engineering.
What’s next?
We’re fortunate to have a business that doesn’t face immediate threat due to LLMs and related technology, but we can’t just put our heads in the sand. Our profitability means we can control our own destiny, so we don’t have to force AI into our product for the sake of it. Instead, we can take the long view of where we’re going, and expand our imagination as to what’s possible.
We don’t want to ban nor mandate AI – we want everyone at Clipboard to be thinking about how AI can change the way we work and the way our customers work for the better. For people joining now, that means the chance to shape exactly how and where AI fits into our business and work.
The only thing I’m really sure of is the shelf-life of this post: 6 months would be a stretch!