From Prompts to Agents: Understanding the Next Phase of AI in MarComm

The ability of AI to complete complex MarComm tasks continues to get better and more sophisticated. Since I began using AI heavily about two years ago, I have found that my own use of the tools has moved from basic prompts to more robust tasks that involve reviewing data and making judgment calls. 

As an early adopter, I see the change that is happening across AI and want to help others better understand where the technology is heading. 

One of the most important places to start is with basic definitions when using AI to achieve a particular outcome. This post has basic definitions of AI functions (from basic to sophisticated) and potential use cases for higher education marketers at each stage.   

Many of the definitions build on one another, so it’s important to understand what each term means. Here’s how the AI terms work together: 

Prompt → Task → Skill → Automation → Agent

Because these approaches can build on one another, one example will carry throughout the post, showing how the same work evolves from a prompt to a task, skill, automation and, ultimately, an agent.

The goal of this post is to help show where the technology is heading and provide examples of use cases. 

Key Definitions 

Prompt – This is asking the AI to do something, and it responds to the specific request. In a prompt, the user initiates the request each time and determines the amount of detail to provide. A single prompt is usually sufficient for requests that don’t happen every day, so consistency isn’t that important.  

I’ve used prompts to help with writing emails to teammates, giving feedback, or brainstorming solutions to a problem. 

Consistent example: Review this webpage and tell me if anything feels off-brand. 

Task – This is when AI completes a defined piece of work. In this situation, the user is still initiating the work but is much more likely to reference other materials to get more specific and consistent results.  

I have given AI a task to review media coverage and categorize by topic or sentiment. Additionally, I’ve asked AI to analyze themes that emerge in student surveys about campus events. In both situations, the AI is referencing a provided document.  

Consistent example:  Review the university brand guides, then review this webpage. Point out instances where the page is not aligned with the university site. 

Skill – The challenge is that prompts and tasks are not just one-time requests. Instead, they are often requests that happen regularly, and there is value in ensuring similar results are produced over time. When that is the case, it can be helpful to convert the prompt or task into a skill. A skill is a reusable set of instructions that tells AI how to complete a specific prompt or task. Often, once you’ve used a prompt or task and gotten the results that you want, you can work with your AI of choice to save that into a skill, allowing you to use the instructions again with different inputs. 

Some of the skills I use regularly include developing a press release draft based on source information, institutional messaging points, and quotes. Another skill I use helps me prepare for interviews. I have built a skill that allows me to provide basic information about the interview, and the skill helps me brainstorm anticipated questions from the reporter. 

Consistent example: Turning the website review into a brand-review skill can create consistency. By providing the university’s standards for naming, tone, accessibility, calls to action and visual identity, the skill can now be used for any website to produce consistent results.  

Automation – The next step in sophistication is when the AI can partner to complete the task without having to be specifically asked. Instead, a timeline or a trigger starts the request, which moves AI from being in a “pull” environment to “pushing” the information to the user. The task itself usually does not change when automation is added, but the trigger allows AI to automatically provide the information to the user. 

I have used automation to review enrollment reports each week. This allows me to prepare for a meeting without having to manually ask AI each week to assist. When the report is released each week, the AI reviews the prior week’s report and gives me key areas of success and key areas of concern. Another way to use automation is to have AI access the team meeting calendar and send a report with details about meetings each day.  

Consistent example: Each Monday morning at 7 a.m. automatically run the brand-review skill on webpages published or updated during the previous week and send me a summary of any issues that need attention.

Agent – An agent is the most sophisticated use of AI, and it is a relatively recent development. In this situation, the AI agent is given a goal, access to relevant tools or data, and clear guardrails. Then, it decides what steps to take or decisions to make to move the work forward. If the AI is unsure what to do or something happens outside its defined scope, it is trained to escalate the situation to an individual. 

I’m just beginning to explore using AI agents, but one example I’m thinking about includes monitoring campaign performance, investigating significant changes in metrics, recommending an action, or escalating to a human if the situation is outside the AI’s authority. In this example, the AI agent can provide insight about campaign performance, which is something I don’t always have time to dig into as much as I would like. 

Consistent example: Monitor the university website for brand issues, determine which ones require action, route routine corrections to the appropriate owner, and escalate higher-risk issues to MarComm for a more thorough review. 

Take it Slow 

More sophisticated doesn’t always mean better. However, it’s key to understand the levels of sophistication to decide what is best for your organization or institution. 

Some requests are better as a prompt, especially if it’s a one-off task. The best way to address what level of AI is needed is to look at the approach to the work. Processes that are similar and happen frequently are great places to explore creating skills, automations, and agents. Additionally, understanding what decisions the institution is comfortable allow AI to make can help guide the MarComm team in how to best use AI. Understanding the differences is a good first step toward making smarter decisions about where, and how, AI fits into MarComm work.