Next-Gen Automation: The Role of Agents

Here's my big prediction for 2025: it won't be the year of the AI agent. This isn't a new prediction. I made this claim last month when discussing how the typical technology hype curve applied to generative AI. I repeated it last week in an AI predictions article on my personal blog. Neither article mentioned the specific factors limiting corporate adoption of AI agents in 2025, hence why I am revisiting the topic today. In media terms, 2025 will be the year of the AI agent. There will be lots of hype around the technology during the year, as well as plenty of interesting demos and use cases. That won't be matched by production usage. There are multiple reasons for this, and AI vendors are a large part of the problem.

Vendors are doing an exceptionally poor job of articulating the benefits of AI agents over traditional automation solutions. Until vendors solve that problem, people just won't use them. After all, the underlying Large Language Models are now relatively well understood. People know the strengths and weaknesses of the technology, and are starting to figure out the best use cases for it. The issue is that vendors are still selling AI as a general purpose solution to any business challenge. The technology just isn't mature enough for that. Hallucination is still a problem, and AI reasoning models aren't yet a match for a skilled human. Capabilities and accuracy are improving over time. However, any process that uses AI agents needs to be able to handle incorrect outputs. Most automated workflows just aren't designed for that.

Rules vs AI

Fundamentally, agents are not a replacement for traditional rule-based automation approaches and shouldn't be sold as such. Agents provide new capabilities, which allow new categories of automation to become possible. If you're developing a process which takes a predictable input and transforms it into a predictable output, then you almost certainly don't want to use AI. However, there are plenty of workflows where existing rule-based automation just doesn't work. Scenarios such as extracting information from unstructured documents or generating a written output, work much more effectively when developed with the assistance of AI agents.

Indeed, the most effective automations will use a mix of agents and rule-based workflows. We're already seeing this in many Agentforce demos and implementations. Frequently, agents are used to trigger pre-existing workflows. Those workflows could update an opportunity in accordance with a pre-existing process, or they could add the record to an existing nurture stream. The AI is merely the trigger, reviewing the notes and activities associated with the record to decide when the account is ready for progression through the funnel. Sometimes, the agent will be given the option of selecting two different workflows - perhaps a nurture flow or a conversion flow - based on criteria listed in the agent prompt instructions.

Guardrails

Combining AI agents with rule based automation allows for the automation of additional business processes. Rule based approaches can't handle uncertainty or inconsistency. Introducing AI resolves that problem, allowing unstructured or dirty data to be used in workflow decisions. However, it is important that the agent is properly designed and its activities are fully scoped. Guardrails are essential. It's the only way to avoid errors or unexpected outcomes. AI vendors are sometimes guilty of selling agents as a method to avoid fully considering every scenario in a rule-based workflow. That's dangerous. Instructing agents on what they shouldn't be doing is even more important than training them on the intended process.

All this requires a very detailed prompt with some quite specific instructions. Writing a comprehensive prompt for an AI agent is not an easy task. It's not clear that business teams are ready for that undertaking at the moment. It's every bit as difficult as designing a basic rule-based workflow. However, it will become even more critical as the skills and abilities of AI agents develop. A lot of the most useful features, like screen automation and autonomous browsing, aren't widely available yet. However, these are the features that businesses are most interested in leveraging within agents. They allow executives to bypass many of the political barriers that impede automation initiatives. Until they're available, agents will remain a niche concern.

Marketo Releases New Email Editor

Stay tuned for a major change to the Marketo platform. Adobe are releasing a preview of the new Marketo email designer this month. This doesn't replace the existing guided email template editor yet, not least because custom template capabilities are still limited. However, it does unlock several new capabilities for Marketo emails created from templates.

Demand Agents: Do Customers Want AI?

AI has been pitched as a technology to increase efficiency and reduce costs. Has anyone considered whether their customers want the AI option? In a recent article on my personal blog, I examine the major use cases for agents in B2B and consider whether customers will accept the replacement of human workers with AI.

Marketing Operations Roadmap Matrix

A Marketing Operations Roadmap is an essential strategic tool that aligns marketing priorities with your overall business goals. Any roadmap begins with understanding where your marketing organisation is right now, and then defining where you want to get to. This free Marketing Operations Roadmap Matrix from CRMT Digital allows you to do exactly that.