This article was written by Benjamin McCulloch, Content and Conversation Designer at VUX World.
It seems there isn’t agreement within the Conversational AI industry about who should be responsible for prompting LLMs. Should it be programmers or conversation designers who design the prompts?
Generally speaking, I’ve seen conversation designers and programmers who have adopted prompting with relish. I’ve also seen some who are cautious about it; they think the hype may be too overblown, or they feel that they don’t know where to start, or they even fear that LLMs may steal their job and replace them.
Regardless of our enthusiasm for adopting the technology, shouldn’t we conversation designers first ask whether we should adopt it?
The prompt has many interpretations
I was surprised when Philip Hunter stated that ‘Prompt engineering is coding. Coding is not designing. Designing well is a goal worthy of single focus.’ in response to one of my previous articles. Philip is one of the few people who has got decades of experience in this industry. He went on to say that ‘Designers should understand and guide the translation [from design to code]’ rather than writing the prompts. This gave me much to think about. I respect Philip’s opinion. Before he said that, I’d jumped into LLM prompting with both feet, and considered it to be the next step in the evolution of conversation design. I thought we’d all be doing it soon.
During the first Agony Agent Roundtable Wally Brill gave a pragmatic view on how conversation designers should view LLMs. He said, “What I tell conversation designers is you’ve got a wealth of knowledge, you’ve got a wealth of experience, so leverage that understanding. There are new tools that you can use, but they’re just new tools. They’re just new things to put in your toolkit.”
You could say that’s happening already. When Greg Bennett spoke about his work at Salesforce, he said that the conversation designers in his team were each assigned a company vertical and tasked with writing prompts for them, such as creating unique sales emails. “So every time a user interacts with a large language model that’s been integrated in Salesforce, my team is responsible for the prompt that lives behind the scenes… we [design within the] larger context that considers ethical and social scope, conversational style, linguistic variation, as well as any kind of relevant data that may be includable from your CRM.”
What if prompting does get assigned to programmers in Conversational AI teams? There’s a danger there too. It could undermine the conversation designer’s work, and some teams may even consider working without a designer entirely. As Rebecca Evanhoe said, “I have a little anecdote. I gave a talk about LLMs. A nice guy came up to me, and he was a developer and said, ‘Oh, is conversation design even a thing anymore?’ And I said, ‘Yeah,’ and I said some stuff [about conversation design]. And the guy said, ‘Well, I can do that. I can have an LLM do all this stuff.’ So then I was like, ‘But is your bot good?’ And he was like, ‘I don’t know.’ It’s easy to make a mediocre bot, but it’s still a high effort thing to make a really good bot.”
We’ve not had the final word
So who should be responsible for LLM prompting? That affects where we hedge our bets career-wise. It’s our jobs that are in flux. There’s a fear among some conversation designers that they’ll get left behind if they don’t know prompting.
On the one hand, the prompt you write will be an input to a machine. You write your input, test the result, debug and iterate. That sounds like software development. You’re giving software instructions. With LLMs there’s a lot of refinement required to get the desired output from the machine.
However, because those instructions are based on language, and the effectiveness of your use of language will impact the results you get, conversation designers are more likely to be able to design more effective prompts.
In reality, the responsibility will be split, depending on the use case. However, it’s one of the many areas where results will improve if there’s collaboration between designers and developers.
Speaking of which, UNPARSED!!! We’ll be in London this June to talk about conversation design, programming and the collaboration between the two. You can guarantee there will be rooms full of people talking about the evolution of Conversational AI skills from both viewpoints. For the foreseeable future, designers and programmers are still in demand, and will still need to collaborate. Come along and get involved in the discussion.
