Gusto's head of technology argues that hiring a large number of specialists is not the right strategy for AI development.

That's "the wrong way to go," according to Edward Kim, cofounder of Gusto, and head of technology
Gusto's head of technology argues that hiring a large number of specialists is not the right strategy for AI development.

That's "the wrong way to go," according to Edward Kim, cofounder of Gusto, and head of technology, talking about the times when founders would be planning for an increasingly AI-centric future and hire a bunch of specially trained AI engineers instead of cutting their existing teams.

Instead, he argues that "non-technical team members actually can have a much deeper understanding than an average engineer of what situations the customer can get themselves into, what they're confused about," which puts them in a better position to lead the features that should be built into AI tools.

Kim — whose payroll startup generated more than $500 million in annual revenue in the fiscal year that ended in April 2023 — detailed in an interview with TechCrunch how Gusto approaches AI development, including non-technical members of its customer experience team as writing "recipes" that guide how its AI assistant, Gus, announced last month, interacts with customers.

People who aren't software engineers, but a little technically-minded, can build really powerful and game-changing AI applications," Kim said, pointing to CoPilot — a customer experience tool that Gusto rolled out to the Gusto CX team in June and is already seeing between 2,000 and 3,000 interactions per day.

We can, in fact upskill quite a number of our folks here at Gusto to really build on AI applications, said Kim.

This interview has been edited to make it longer and clearer.

Is Gus the first big AI product that you've released to your customers?

So that's Gus, the big AI functionality we've launched to our customers, and in many ways ties together a lot of the point functionality that we've built, because what you start to see happen in apps is they get littered with AI buttons that are like, "Press this button to do something with AI," whereas ours was, "Press this button so we can generate a job description for you.".

But Gus lets you take all that away, and when we feel like Gus can do something that's of value to you, Gus can pop up in an unobtrusive way and say, "Hey, I can help you write a job description?" That's a much cleaner way to interface with AI.

There are some companies which claim to have been doing AI for a million years but only just recently received attention, and then there are those like them which only learned of their opportunity in the last couple years. Does Gusto fall into one camp or the other?

The huge difference for me is, when you talk about software programming, to most people it is not accessible. You need to learn how to code, go to school for a few years. Machine learning was even much more inaccessible. First of all, because you had to be a very special type of software engineer and had this data science skill set and knew how to create artificial neural networks and things like that.

The main thing that changed recently is that the interface for creating ML and AI applications has become much more accessible for just about anybody. In the past, we would have to learn the language of computers and go to school to learn that language, whereas now computers are learning to understand humans more. And that doesn't seem like a big deal, but if you think about it, just makes building software applications so much more accessible.

That is exactly what we have seen at Gusto: People who are not software engineers but who have a little technical aptitude can build truly powerful and game-changing AI applications. We're actually using a lot of our support team to extend the capabilities of Gus and they don't know how to program at all. It's just that the interface that they use now allows them to do the same thing that software engineers have always done, without needing to learn how to code. If you want, I could talk through one example of each of those.

That'd be great.

There's this one individual who's been at the company for about five years. His name is Eric Rodriguez, and literally just moved into the customer support team [and then] transferred to our IT team. While on that team, he started getting pretty interested in AI, and his boss comes up to me and is like, "Hey, he built this thing.". I want to show it to you. Well, this was my first time ever meeting him in person but he showed me what he had built and what it was more or less-a CoPilot tool for our [customer experience] team. Ask it a question and it will just give you the answer in natural language, like ChatGPT might, except it has access to our internal knowledge base of how to do things in our app.

At this point, we present this to our support team, and they just loved it. It completely changed their workflows and how efficient they are. So basically, anytime they get a support ticket, instead of going through this knowledge base that we've built, they actually ask this CoPilot tool, and the CoPilot tool actually answers the question for them. There is still a human between the CoPilot and the customer, but very often they just get the response through the CoPilot tool and then copy paste it to the customer. They will verify whether it's correct which most of the time it is.

So we transferred [Eric] to the software engineering team right away. In fact, he reports directly to me, believe it or not, and he is one of our best engineers now. Because he was one of the early adopters of just playing around with AI and now he's on the forefront of building AI applications at Gusto.

Not everyone is as technically inclined as Eric, but we found a way at Gusto to leverage the domain knowledge expertise of non-technical people within the company, especially in our customer support team, to help build more powerful AI applications-and particularly, to enable Gus to do more and more things.

Any time the customer support team gets a support ticket-that is, one of our customers reaches out to us because they want our support team's help on something-and if it comes up repeatedly, we actually have the customer support team write a recipe for Gus, meaning that they can actually teach Gus without any technical ability. They can coach Gus in walking that customer through that problem and sometimes even acting on it.

We built an internal interface, which is an internal-facing tool where you can write instructions in natural language to Gus on how to handle a case like that. And there's actually no-code way for our support team to be able to tell Gus to call a certain API to accomplish a task.

There is a lot of conversation out there right now like, "We are going to eliminate all these jobs in one area and we're hiring these AI specialists that we're paying millions of dollars because they have this unique skill set." And I just think that's the wrong way to go about doing it. That is because the people who are going to be able to progress your AI applications are actually those that have the domain expertise of that area, though they do not have the technical expertise. As we could upskill a lot of our people here at Gusto in helping them build AI applications.

The scary AI scenario is this top-down thing where executives are saying, "We need to use AI" and it's disconnected from the reality of how people work. It feels like this is a bit more bottoms up where you've built tools to allow teams tell you what AI could do for them.

Exactly. In fact, people closer to the customer-non-technical people-they speak to them every day-they basically know better, much more so than an average engineer, what situations a customer would get himself into, what confuses them. So, they are in a better position than engineers or AI scientists to write instructions to Gus about how he can solve that problem.

The truth is that the best AI engineers are really the domain experts who have learned how to write good prompts.

As you contemplate how this will play out over the next several years, do you feel like the headcount across different teams is going to be pretty similar, or do you see that changing over time as AI is deployed across the company?

I think the role does evolve a little bit. I think you'll see a lot of our CX folks not directly answering questions but actually writing recipes and doing things like prompt tuning to improve the AI. Everyone's going to just move up the abstraction layer, and then obviously it will bring more efficiencies to the company and also better customer experience, because they'll get their questions answered immediately.

And that really unlocks Gusto to do more things for our customers. There's such a huge roadmap of things that we want to be doing, but we can't because we're constrained in resources.

Blog
|
2024-10-21 18:26:17