How marketers can approach building an AI agent

Building an AI agent starts with a defined task, access to approved information and tools, and limits on what the software can do, according to HubSpot Marketing. For marketers considering the technology, the process also involves deciding whether an agent is appropriate for the workflow, then testing and improving it before relying on it in practice.

An AI agent uses a model to pursue a goal without continuous human input. It can select from permitted tools, assess the results and determine its next step until the job is complete or it reaches a set limit. That distinguishes the concept from a chatbot focused on responding to questions. HubSpot also identifies rules-based automation as a separate approach to solving workflow problems.

One marketing use case is campaign analysis. An agent assigned to review performance and suggest changes would need to examine channel metrics, compare them with the campaign objective and identify possible reasons for the results. It would then explain its recommendations. The example illustrates why the agent’s goal and the information it can access need to be established at the outset.

Interest in agents is growing, according to figures cited by HubSpot. A 2026 Menlo Ventures survey of 5,000 U.S. adults found that one in four AI users had used an agent. The survey also found that 24% of respondents reported regular use. HubSpot’s guidance extends beyond marketing to potential applications in sales and customer service.

Teams do not necessarily need to write code to start. HubSpot says its users have access to a no-code agent builder that lets marketers describe the automation they want to create. Whether using that tool or another approach, the planning questions remain: what the agent should accomplish, which data and tools it may use, and where its authority should stop. HubSpot also highlights testing, evaluation, guardrails and human handoff as parts of the process.

What this means for marketers

Start by identifying a specific workflow and deciding whether an agent is a suitable fit. Define the desired outcome and the approved resources before deployment, then test the agent and review its performance. Setting action limits and a path for human handoff can help teams establish how the agent should operate within the workflow.