Is Prompt Engineering A Profession?

With close to 39% of businesses seeing a dominant role of AI in revenue and operations in the next 5 years, developing a comprehensive AI strategy has become an imperative. The changing landscape of AI has made old plans obsolete and with Gen-AI offering new possibilities, a fresh approach to strategy is needed.

During the course of one of my strategy sessions, the discussion turned to the role of prompt engineering as a job function in an organization. This was important not just from a immediate resource planning needs but also from a strategic impact to the operating model of the organization given the role it plays shaping several job functions in future. Adding impetus to the current trend, prompt engineering skills are becoming an in-demand function and command a premium in the industry.

However, as with most new things and evolving technology landscape, there seems to be an over-rotation of the idea. Anybody looking to invest in a prompt engineering career should go in with eyes wide open.

What is prompt engineering ?

Prompt engineering, as the name suggests, is a fairly new discipline, which focuses on crafting the best text instructions, or prompts, for an AI engine to learn or produce effective results. Prompt engineers, the people who develop the prompts, aka "AI Whisperers", play an important role in plugging the gaps in AI engines. With the right combination of nouns, verbs and sentence structures, these engineers enhance the quality of the AI engine either by improving the quality of the results or identifying problems. ?

While the actual function and role are still evolving and vary vastly depending on the industry and role, the common thread of a prompt engineer function continues to be a fine tuning of the AI engine's behavior to specific human expectations via instructions.

In addition to being an expert in the subject area of the domain, to be a good prompt engineer following skills are a must:

• Ability to provide detailed instructions - AI engines thrive on instructions. It is a no-brainer to assume that somebody good at providing detailed instruction to achieve a task can be good at prompts.
• ?Language - Again, an ability to articulate instructions in clear, concise sentences with variations in parts of speech is critical.
• Problem Solving - Establishing correlation, finding causation, connecting disparate ideas, understanding results of AI output, identifying hallucinations and their causes, are all invaluable to train a AI engine with good prompts.
• Prompt Techniques - Mostly techniques such as "chain-of-thought" and "Tree of thought" have been tools of choice for prompt design. The former being prompts that show detailed instructions and the latter combines multiple prompts to deduce a common approach. arXiv published a great paper on the topic of prompt patterns.

Is it really the job of the future?

One of the most fascinating things about prompt engineers is that they need not be actual engineers or even be software or AI practitioners. As long as they carry domain expertise and a good grasp of English language(for an English AI engine), these skills can come from anywhere. If "English is the new programming language", as is often quoted, then the bigger question is what is the role of traditional software engineers, and is the future going to be prompt engineers who know how to ask the right questions? How should enterprises plan to staff these roles, how to grow and retain talent? Is prompt engineering a profession that needs to be cultivated and invested in the future.

The answer is a qualified 'no'.

While there is no denying the good prompt engineers can help build a robust Gen-AI engine in the short run and we should certainly invest in it, overall the job and the role are more of a temporary blip than a sustained trend. There are few reasons for it.

• Maturity of Gen-AI engines: As powerful as ChatGPT and others look like, we are still in very early years of the evolution of the technology. The goal is not to have a Gen-AI engine respond to questions framed as complicated strings, but instead should be easy conversation. As the AI engines learn and become more efficient, so will their response to prompts. This would mean, there won't be a need for extensive training through prompts.
• Domain Expertise: Sensitive areas such as defense, health, space etc. cannot have traditional prompt engineers to train AI for critical life decisions. They will need domain experts to double as prompt engineers for the foreseeable future. The roles are very specific and constrained to the specific scenarios using AI that doesn't at all look like the AI-chatbots we are used to.
• Lack of Fungibility of Prompts: The nature of AI engines precludes usage of prompts across them. In other words, development of prompts become too specific to the engine they are used for. With limited ability to reuse prompts, the domains of usage remain small and thus cannot sustain a full profession
• AI Solving its Own Problems: We are not too far off to ask the AI engine to generate its own prompts. Or even better, ask another engine to generate prompts.

To summarize, relying on humans to develop good prompts is a necessity today, especially to plug fairly sizeable gaps. However, these are known issues and need to be addressed. Once done, there isn't a whole 'prompting' needed. For enterprises and individuals, waiting for the landscape to settle is a more prudent approach.

Comments and thoughts welcome.

Epilogue: There is one way the function can survive as a skill if not a dedicated profession. That is the age-old domain of problem solving. The future of software development and other areas will be significantly change with an estimated 15% of new apps will be AI generated by 2027. There will be a need for skills at a higher level function who can understand, parse and solve problems using AI.