AI App Development for Everyone | PromptLayer

The Rise of Prompt Engineering and PromptLayer

The recent surge in GenAI technologies has spurred the creation of numerous startups focused on aiding prompt engineering. This involves crafting precise instructions to guide AI chatbots toward delivering valuable outputs.

Consider platforms such as OpenAI’s ChatGPT and Google’s Gemini, which provide users with an open input field. The phrasing of your request, as well as the request itself, significantly impacts the results you receive.

PromptLayer's Early Entry

PromptLayer, a New York-based company, entered this market early, introducing a tool to assist application developers in managing prompts approximately two years ago.

Co-founder Jared Zoneraich explains that the founders were experimenting with AI chatbots and sought a method to organize their prompting efforts. (Zoneraich is pictured above left with co-founder Jonathan Pedoeem.)

Initially, they released a minimal viable product (MVP) on X on a somewhat impulsive basis. The tool, originally designed for their own use, quickly gained traction, prompting them to continue its development.

Evolution into a Comprehensive Platform

The platform has since matured into a comprehensive prompt management product, marketed to businesses to facilitate the development of AI-powered applications.

This suggests the founders accurately anticipated the increasing demand from businesses exploring how large language models (LLMs) can enhance productivity.

Despite the growing competition in the prompt support sector since PromptLayer’s initial MVP release, they have developed a fully-featured platform.

This platform offers a visual interface equipped with tools for managing and monitoring the process of maximizing the value derived from LLMs.

Seed Funding to Fuel Growth

The company has now secured $4.8 million in seed funding to further accelerate its progress.

The investment round was led by Ivan Bercovich (ScOp Venture Capital), with continued participation from Peter Boyce II (Stellation Capital), who also contributed to their pre-seed round.

Several angel investors and leaders in the AI field also participated, including: Michael Akilian, Joshua Browder, Byrne Hobart, Romain Huet, Josh Kamdjou, Logan Kilpatrick, Ben Lang, Alex Oppenheimer, Gokul Rajaram, Gabriel Stengel, and Luis Voloch.

Monitoring and Managing Prompts

According to Zoneraich, the fundamental component of PromptLayer’s offering is a “prompt registry.” He describes it as a system for prompt management, providing version control capabilities. Users can track changes between prompt versions and designate a production-ready version.

This registry serves as the central hub of the product, with additional features designed to enhance its utility. These include testing functionalities and logging mechanisms to record prompt usage.

The platform enables users to rigorously test and assess various prompts tailored to their specific application, such as an AI-powered coaching tool or a customer service chatbot. It facilitates performance comparisons across different Large Language Models (LLMs).

Furthermore, it provides valuable insights into the evolving landscape of application development, where effectively instructing advanced technology increasingly relies on carefully crafted language rather than traditional code.

Notably, PromptLayer distinguishes itself by specifically targeting users without extensive technical backgrounds.

Zoneraich explains that the company intentionally focused on building a prompt management solution for “domain experts” – professionals possessing specialized knowledge in fields like education, law, or healthcare – after observing that non-programmers were actively participating in the application development process.

“The development of AI in sectors like healthcare necessitates the involvement of medical professionals, legal AI requires legal expertise, and therapeutic AI demands the input of therapists,” states the startup’s mission, emphasizing that their tools facilitate collaboration between domain experts and engineers through a visual prompt CMS.

Zoneraich emphasizes that the platform empowers domain experts to take the lead in application development.

“While some training is required, the learning curve is manageable,” he notes. “It doesn’t necessitate coding skills, making it accessible to a broad range of individuals.”

A Distinct Strategy in the GenAI Landscape

With the proliferation of generative AI tools, largely spurred by OpenAI’s user-friendly interface, a focus on solutions for individuals without extensive technical expertise appears strategically sound. Zoneraich believes this approach differentiates his company within the current market.

“We are pursuing a markedly different path compared to most others,” he explained in a conversation with TechCrunch. “The idea of subject matter specialists taking the lead – this is a relatively uncommon practice. We arrived at this understanding through direct feedback from our clientele.”

He continues, noting a potential bias in Silicon Valley: “Building for non-technical users is often perceived as less appealing than catering to technical audiences.”

“We don’t anticipate needing to demonstrate the validity of this method,” he asserts. “The market itself will provide the necessary validation.” He further argues that successful implementation within specialized fields necessitates the involvement of domain experts in the process of prompt engineering.

“The availability of engineers is limited, even if we were to prioritize staffing with them,” he points out.

When discussing competitors, Zoneraich mentions platforms like Zapier, alongside “LLM ops” companies such as Braintrust and LangChain. However, he believes there’s an overemphasis on developing tools specifically for technical users.

“Our conviction is that, for the majority of organizations aiming to harness the capabilities of Large Language Models, the relevant expertise for application development will reside with non-technical personnel,” he states.

He also suggests that the abilities required for effective prompt engineering don’t automatically align with traditional programming skills.

“The art of prompt engineering isn’t entirely dependent on engineering proficiency. While there’s some overlap, it fundamentally requires an experimental mindset – a willingness to test various approaches and observe the resulting output,” he elaborates. “Some individuals attempt meticulous planning and research before crafting prompts, but in our experience, those individuals are less successful.”

“In fact,” he added, “a deeper understanding of the underlying LLM may actually be detrimental to effective prompt engineering.”

Demand for LLM Tools

Zoneraich expresses a strong belief in the substantial market demand that will emerge for tools designed to maximize the potential of Large Language Models (LLMs). He doesn't anticipate that the relatively new field of prompt engineering will prove to be a fleeting trend, quickly overtaken by further advancements within the rapidly evolving Generative AI landscape.

His reasoning centers on the idea that even a hypothetical Artificial General Intelligence (AGI) would still require some form of input to function effectively. This suggests that human involvement, in the form of tooling and support, will remain crucial for guiding these machines for the foreseeable future.

The Core Challenge

“The central difficulty lies not in the technology itself, but in determining its application,” he explains, highlighting PromptLayer’s commitment to developing long-term solutions. “The real challenge is defining the specific task to be addressed.”

He elaborates, stating that if numerous viable solutions exist for a given problem, the role of the prompt engineer becomes paramount. Their task is to identify the precise problem to solve and establish the appropriate context for doing so.

Abstraction Layers

“The LLM essentially serves as the instrument for transforming a problem definition into a solution, but this merely shifts the level of abstraction,” Zoneraich continues. “We progressed from machine code to contemporary programming languages, and then from those languages to prompts. It’s conceivable that we’ll move from direct prompts to inputs that refine those prompts.”

However, he emphasizes that a fundamental requirement will always remain: “Ultimately, some form of input is always necessary – there’s an irreducible element to the process.”

Future Development

The recently secured seed funding will be allocated to expanding the team, which currently consists of eight members. A primary focus will be on recruiting in-house engineering talent to guarantee the quality and dependability of the service provided to customers, according to Zoneraich.

Furthermore, they intend to broaden the platform’s capabilities to accommodate a wider range of applications and increase user engagement. Significant effort will also be directed towards community building, fostering the growth of this emerging field of prompt engineering.

Community and Growth

“The precise definition of a ‘prompt engineer’ remains undefined, and we see it as our responsibility to cultivate a community around this discipline,” he states. “We aim to be pioneers in prompt engineering, demonstrating best practices to others.”

While PromptLayer doesn’t currently disclose the number of paying customers, Zoneraich reveals that over 10,000 individuals and organizations have utilized their website, encompassing both free and paid users. (ParentLab and Speak, backed by OpenAI, are cited as examples of paying customers.)

The company has experienced a 13-fold increase in revenue this year, attributing this growth to word-of-mouth referrals as teams recognize the need for domain expertise, not solely engineering skills, in AI development.

A Scientific Approach

“Resolving prompting issues often requires simply modifying the prompt and observing the resulting changes – and we offer a suite of tools to facilitate this process at scale,” Zoneraich concludes. “At its core, it’s a matter of applying the scientific method.”