The GenAI boom of the last few years has unleashed a wave of startups promising to support the process of prompt engineering — i.e., coming up with instructions to precisely steer an AI chatbot to serve useful output. So think tools like OpenAI’s ChatGPT and Google’s Gemini, which present the user with a blank field for their query — and where what you ask and also how you ask it can lead to very different results.
New York-based PromptLayer got into this space early, launching a tool to help app developers manage the prompting process around two years ago. Its founders had been playing around with AI chatbots themselves and wanted something to keep track of their own prompting, recounts co-founder Jared Zoneraich (a TechCrunch Disrupt hackathon alum, pictured above left with co-founder Jonathan Pedoeem).
On a bit of a whim, they put their MVP out there (on X) and the tool they’d built for themselves made a splash, so they kept building. It’s since evolved into a prompt management product they’re selling to third-party businesses to support their development of AI apps, suggesting the founders hit the right timing to cater to growing interest from businesses in how large language models (LLMs) might help boost their productivity.
While the prompt support space has heated up considerably in the years since PromptLayer released an MVP they’ve gone on to build out a fully featured prompt management platform — offering a visual interface packed with tools for managing and monitoring the process of trying to extract the best value from LLMs.
Now they’ve closed a $4.8 million seed raise to continue building momentum for their approach. The round is led by Ivan Bercovich (ScOp Venture Partners), with Peter Boyce II (Stellation Capital) participating again (he also funded their pre-seed), along with several angels and AI founders and operators — including Michael Akilian, Joshua Browder, Byrne Hobart, Romain Huet, Josh Kamdjou, Logan Kilpatrick, Ben Lang, Alex Oppenheimer, Gokul Rajaram, Gabriel Stengel and Luis Voloch.
Keeping tabs on prompts
Zoneraich says the core of PromptLayer’s product is a “prompt registry”. “It’s CMS, it’s version control for prompts,” he explains. “You have a prompt, you create a new version, you could see why versions are different, and then you could choose which version’s your production version … That’s like the center of our product — and everything kind of expands out from that and tries to make that more useful.”
“For example, tests on top of that, or logs on top of that of when you use which prompt, or A/B tests between the prompts, and kind of like deeper insights into which version is the best one.”
The platform is designed to support customers to test and evaluate different prompts for their particular app use case — say an AI coaching app or a chatbot for customer support — letting them test how different versions of prompts perform across a range of LLMs; and, more generally, get a better handle on this brave new world of app development where the language required to command cutting-edge tech is, mostly, just words (rather than code).
Unusually for a dev tools maker, PromptLayer is deliberately focused on non-technical users.
Zoneraich says they made a conscious choice to build a prompt management business geared towards what he refers to as “domain experts” — i.e., professionals with key expertise in their field, whether that’s education, legal, healthcare, and so on — after they found early users were bringing non-coders to the app development party.
“We believe you can’t build healthcare AI without doctors, legal AI without lawyers, or therapy AI without therapists,” the startup writes in a mission statement that says its software tooling “allows domain experts and engineers to collaborate using our visual prompt CMS.”
Zoneraich goes further — saying the platform puts domain experts in the “driver’s seat” of app development.
“This becomes something they need to sort of be trained in — but it’s not a huge leap,” he suggests. “It’s not like that they have to learn to code. So it’s something that the average person could pick up.”
Taking a different tack
Given how GenAI has cracked open the AI toolbox — thanks, in large part, to OpenAI’s decision to embed generative AI in an accessible, natural language interface — the choice to focus on tools for non-technical users makes logical sense. Yet, Zoneraich reckons it sets them apart from the majority of players in the space.
“We’re taking a very different approach than everyone else,” he suggests during a call with TechCrunch. “This whole concept of domain experts leading the charge — basically nobody’s doing that. I think we learned it from our customers. But in Silicon Valley, it’s kind of a little bit less sexy to build for the non-technical, rather than the technical.”
“I don’t think we need to convince anyone this is the right way to do it. I think the market will do that talking for us,” he goes on, arguing: “I don’t think you can win in a lot of these domains without employing domain experts [to do prompt engineering].
“There just aren’t enough engineers anyway, even if we wanted to staff everything with engineers.”
On the technically focused AI tools side, Zoneraich name-checks the likes of Zapier — as well as pointing to what he refers to as “LLM ops” companies like Braintrust and LangChain — when we discuss the competitive landscape. “But I think everybody’s over fitting on that [tools for technical users],” he contends, saying it’s his conviction that for most companies seeking to leverage the power of LLMs, the relevant domain experts for the app they’ll want to build will not be technical staff.
He also reckons the skills required to be a good prompt engineer are not necessarily the same as those that make for a good programmer.
“The skill of prompt engineering is not 100% correlated with engineers. There’s a subsection, but it’s really, like, a tinker type of skill [that’s required] … ‘I’m just gonna try this random thing, and then I’m gonna see what the output is,’” he tells us. “Some people try to really plan it out and do research on what the prompt should be. And in my opinion, those people are not good at [prompt engineering] because there’s not really a science.”
“I think the less you try to understand the LLM, the better you are,” he added.
Building demand
Zoneraich is bullish on how large the market need will be for tools to get the best out of LLMs. Nor is he concerned that this newly emerged field of prompt engineering will end up a blip in the history of work that’s quickly steamrollered by fresh developments in the fast-paced GenAI market.
Even an AGI — were such a generally intelligent AI to be brought into being — would need something to work with, he argues, implying that humans are still going to need some form of tooling/support to nudge the machines for the foreseeable future.
“The hard part is, what do I do with it? The hard part is, what task do I give it to solve?” he says, underscoring PromptLayer’s confidence that it’s building tooling for the long term. “The hard part is defining what to do.”
“If you believe there’s no one perfect solution to a lot of these problems, there are infinite ways to solve it, and that’s the job of the prompt engineer — to choose what problem am I even solving? What is the context to solve the problem?” he goes on.
“The LLM is kind of just the tool to go from problem definition to solution, but you’re just moving the abstraction layer … We moved it away from machine code to modern programming languages. Then we moved it from modern programming languages to prompts. And maybe we’re going to move it from raw prompts to be like input to the prompts,” he said.
“But at the end of the day, you still need some input. There’s that irreducible part of it.”
The seed round will be used to expand the team (currently eight-strong), with a focus on adding in-house engineering talent to ensure quality and reliability of service for customers, per Zoneraich. He says they also want to expand the platform to serve more use-cases and grow usage — as well as putting effort and energy into community building to help nurture this unfolding field of prompt engineering.
“The jury’s not out on what a prompt engineer looks like or what a prompt engineer is. And I think it’s our job to kind of build a community around this — be like a pioneer of this prompt engineering field and show people how to do it. So that’s a big focus.”
PromptLayer isn’t disclosing how many paying customers it has for its tools as yet — but Zoneraich says they have over 10,000 free and paid customers who have gone through their website. (ParentLab and OpenAI-backed Speak are two paying customers he names.)
The startup has also seen 13x revenue growth this year — and it claims this rapid revenue growth is purely through word-of-mouth “as teams discover they need domain experts, not just engineers, to build AI”.
“All it takes to fix these prompting problems is you go, you change the prompt, and you see how the results are — and we have a lot of tooling to help you do that at scale. But that’s the core thing: just scientific method,” adds Zoneraich.