ViktorPetrov

We are doing demos with engineers and knowledge workers and the difference is huge.

It's common to say you don't need engineers anymore since 70% of the code is written by AI, but you need engineering thinking to produce outcomes that are good.

Engineers structure their outputs, put guardrails and use deterministic approach when they create a workflow. They structure prompts like they structure work/software.

Non-engineers expect our agents to deliver magic without putting the effort.

We are trying to explain that if they want the automation to work, they need to put more effort into imagining what the agent will conduct and how to provide as detailed as possible instructions to achieve that.

I think AI Agents are there, but the rest of the people don't know how to utilize it properly.

We've spent the last 3 months doing demo of our agent to two groups of people: software engineers and knowledge workers.

The problem we are facing is not that our agent does not perform, but rather how knowledge workers do their job.

Engineers almost always get the agent to finish the job on the first or second try, because they use structure and detailed instruction in their prompting.

When we demo knowledge workers on the other hand we see them prompting 1-2 sentences and expect the agent to fill in the context, to understand their intent, position they are in. I mean the agent delivers, but I'm not sure the expectations are met.

So the problem we are facing right now is how do we teach knowledge workers to structure their prompts like an engineer would define a task before they go ahead and start work.

I think the whole agents industry is facing this issue and our question is how do we teach them to put effort into defining their tasks?

I see, so you are hesitant to provide financial information to our service and we need to: - make sure we keep it safe - provide a better value proposition to users ?

the onboarding needs rework