Only a couple of years ago, writing AI prompts was a potential career path. Since ChatGPT’s public release in 2022, AI has come a long way. From basic text generation we now have Claude Mythos capable of ‘autonomous hacking’. The Anthropic CEO recently sounded a cautious note about the imminent ‘self-development’ capability of AI. In short, things have moved so rapidly that it is no longer a highly-valuable skill to write prompts for basic LLMs. Keying something into ChatGPT, Gemini or Co-Pilot is now a business-as-usual skill as part of life and employment. So, let’s discuss why writing AI prompts is no longer a clear career path.
Writing AI prompts
There are various guides and advice covering best practice when writing AI prompts. We would suggest these 3 tips to help your interactions:
- Provide detail about what you want the AI to do and what you DO NOT want the AI to do within the specific context you are working on;
- Specify how you want the output to look including any formatting criteria that you can save yourself time tidying up later;
- Prioritise iteration rather than recreating what you did, which can have multiple benefits from fine-tuning to lower compute costs.
Ultimately, this is now a ‘foundational skill’ in fields from software development to financial analysis. The better that you are at this, the better and more reliable your AI output is. For businesses, you can immediately see why this is advantageous. In essence, we have passed the ‘trial and error’ phase of the AI transformation. So, let’s look at some other challenges for careers.
AI prompts and career paths
Managing multiple AI agents to deliver predictable workflows with reliable outputs is quickly becoming the core of some people’s jobs. People who effectively manage this ‘AI pipeline’ of autonomous work in expert domains will command a salary premium. Ultimately, one person is potentially doing the work of a team with multiple co-ordinated AI agents. We should also recognise that many LLMs are simple, text-generating models, providing limited benefit in the workplace. Only those employers that have licenced or developed more specialised models may yield such productivity benefits.
Within organisations, people are creating the modern equivalent of end-user developments using AI prompts and in-house agents. Where once this was creating a database system or internal web portal, people are now creating AI templates and optimised local, multi-modal models. The best way to prepare yourself is to develop your data analytics and AI knowledge. Of course, you can also utilise consumer AI models and AI functionality built into software applications. These also help with basic practice at writing prompts and familiarity with what models can and cannot do. So, do you need to become an AI expert?
No longer valuable unless you are an AI expert?
In some professions, AI remains peripheral. In other professions, it is being held back due to concerns over data privacy, the need for absolutely accuracy (i.e. hallucinations) or lack of transparency. Where professions involve your hands and interaction with other human beings, there is generally less impact from the AI boom. Where professions are highly-specialised, sensitive or economically critical, the take up is tightly-controlled until firm guardrails, guidelines, security and transparency are in place.
Critically, some organisations have expressed nervousness about their employees accessing simple, public LLMs. This is due to concerns over data leaks, data privacy and security. Sensitive industries and Government are worried about the sovereignty of AI models, both in terms of who you rely on to host/run them and who can access or use the data for other purposes. You may also recall our article suggesting that 90% of AI developments do not generate ROI. In summary, in many fields you will remain valuable for quite some time.
Accountability with AI-generated content
Yes, there have been a few court cases regarding the use of AI-generated content. This has (so far) largely revolved around the largest accounting and management consulting brands. In general, the relative inexperience of AI operators and a lack of understanding of how the AI agent works ‘under the hood’ is leading to litigation. Sure, you may have figured out how to generate a slide deck using a tweak to your prompt – but is it accurate? Are there glaring errors? Has it disclosed some sensitive information from another client? Is it recommending something that is inappropriate for the recipient? Can you explain how it was generated? Is it very similar to another presentation for a similar organisation?
As one senior leader recently reminded us, directors remain liable for breaches of company law. Therefore, unleashing AI models on a workforce that is ill-prepared to develop them, operate, co-ordinate and optimise them is a mistake. Similarly, in many contexts, it is hard to rely on something that you cannot explain where it came from. Sure, your report took all of 30 seconds to generate but where did the data come from? Is the data accurate? Are you permitted to publish this content or use it for this purpose? Did we take a health and safety, regulatory or legal decision based on a hallucination? It would be an interesting defence.
Making sure your organisation is ready for AI
Getting the basics right is fundamental to AI adoption and reaping potential benefits. In many cases, organisations lack the skills, data or infrastructure for rapid AI adoption. They may also unwittingly expose themselves to risk, ‘AI slop’ and uncontrolled compute costs. AI prompt writing is now a foundational skill. Building your skills, co-ordinating, optimising and operating multiple AI agents will make you more valuable in specific domains.
If you would like to speak with experienced consultants, simply reach out to our team.
Alternatively, please check out some of our broad expertise and where we tend to work.
Finally, check out a related AI article about pursuing cost savings before an investment.