Prompting assistance

Why does prompt assistance matter?

It's a standard analogy now, comparing prompting to the old days of writing in the terminal and yet, it's still a major input type used in many apps. We also know that many users struggle to express their intent in words, and a weak prompt leads to mediocre AI outputs, thus further leading to "AI doesn't get me, it's useless!".

Prompt assistance bridges this gap.

Let's assist you in understanding this pattern better.

Key takeaways

1. One-click enhance

Many products let users type a short prompt and then offer a "Enhance this prompt" button.

Bolt uses this one click technique to create a prompt. On the downside, the feature is hidden inside the + icon making it less discoverable.  

2. Scoring prompts

It's like Grammarly but for your prompts.

Elicit is an amazing example of this.   Jenni AI has a unique implementation of this.  

3. Following up

Perplexity pro-actively asks the user questions to clarify what they meant and irons out the direction in which AI should explore.  

4. Use-case templates

This links back to the Suggested prompts AI pattern. Showing example prompts like “summarize this email,” “generate slide deck” right near the input field acts as training wheels.  

5. /plan mode vs doer

Claude code was the first that embraces how we humans have been using AI. Instead of always inputting into the system, "ask me questions before starting" there's a dedicated mode. Plan first, execute second.   v0 has a UI based approach to change the mode.

The creative tools version of this is "draft mode" - using less no. of credits to come up with a rough concept like how Midjourney does it.

Editor's notes

As a practitioner, I like to write notes — key takeaways and questions — to ask myself whenever I'm designing for prompt assistance in the future.

It’s a tangible gut-check for myself and for you to steal, if you see fit.

Future play

We would be lying if we said prompting is here to stay. It's definitely a good input when thinking of AI as a conversational partner, but not something in which you have to think about the syntax, or if you've mentioned enough things etc. The day AI is treated as an intern at the very least is when "traditional prompting" would be dead.

So the future play of prompt assistance? An intent assistance. While you're talking to AI, two models collaborate to help you solve your intent. Maybe one helps you plan, and the other is the doer. We're already seeing this play out in the agentic products, it'll only get better from here. Multiple agents working together to plan and execute your intent, not necessarily focusing on the words that you're typing.