The second installment of the AI Edge workshop series unwrapped one of the tech’s biggest buzzwords: automated workflows.
This AI Edge Session, hosted by Brandon Z. Hoff, chief AI officer of Responsible Use of Digital Intelligence (RUDI AI), showed marketers how artificial intelligence can move beyond chatbot interactions and complete multistep tasks with limited human direction.
During a live simulation, Hoff used Anthropic’s Claude Cowork environment to build a marketing operations system. The system researched a company, analyzed its website, developed brand guidelines and generated marketing materials.
According to the event organizers, the broader goal was to change how marketing professionals approach AI. Instead of asking a chatbot to complete one task at a time, Hoff encouraged participants to build workflows that allow AI to manage connected tasks.
What Are Automated AI Workflows?
Hoff said decision-makers are placing greater emphasis on agentic workflows, which allow AI systems to complete sequences of tasks with less human intervention.
A workflow gives the AI a defined process to follow. It can include instructions, tools, files, quality checks and specific outputs.
He explained the approach differs from a traditional chatbot interaction. Unlike a chatbot that responds to an individual request, an automated workflow can work through several steps to reach a defined outcome or objective.
For marketers, a workflow could include researching a target audience, reviewing a company’s website, identifying competitors and creating campaign materials.
Building The Right Foundation
Hoff emphasized that the key to effective prompts and workflows is the quality of the information provided. The goal is to give the system structured files and information that processes with less friction.
Common file formats can contain formatting information that creates unnecessary noise for an AI model. Hoff recommended using cleaner, machine-readable formats such as Markdown, JSON, CSV and HTML when building workflows.
The Framework
He introduced participants to a five-part instruction model called the CRAFT Framework, explaining that CRAFT stands for context, role, actions, format and testing.
Under this framework, users begin each prompt by providing AI with context and background information about an organization, project or objective. Once the context is provided, users then assign it a role that tells the AI who it should think and act as. For marketers, this can be a specific role, such as a marketing analyst or executive persona.
After setting the persona, users outline actions by listing precise, numbered to-do steps that instruct the system. The following step is for users to specify the final deliverable format. Hoff encouraged participants to define the exact format, rather than only describing the specs.
The final step is testing. Users establish quality controls and boundaries before the workflow begins.
Proving Access
Hoff explained that setting up successful Claude workflows requires sharing direct access to workplace tools and local computer systems. Connections expand what the AI is able to do. It also allows an AI workflow to function more like a digital worker operating within defined boundaries.
Based on what is shared, capabilities can expand to include sorting information from emails, organizing files or conducting deep research.
The system can also connect with tools such as Gmail, Slack, Chrome and Google Drive. With browser access, the system can also visit websites and collect information from public sources.
A Live Marketing Simulation
Hoff put the framework into practice during the workshop. For the first demonstration, Hoff simulated attending AfroTech for a target organization.
Participants received a Marketing Operations Starter Kit, which included brand guidelines, campaign strategy and content planning objectives. Attendees were instructed to download the open-source materials and load them into a new Claude Cowork project.
Hoff then submitted a master onboarding prompt based on the CRAFT Framework.
The prompt instructed Claude to review the project’s instructions and assume the role of a marketing operations executor. The system reviewed the available files and skills before asking for two pieces of information: a target website and a business objective.
Claude then used the Claude in Chrome extension to inspect the AfroTech website.
The system identified information about the organization and its audiences. It also analyzed the site’s visual system.
The demonstration included extracting CSS information and identifying color codes from the website. The system used that information to establish elements of a brand identity system.
The workflow then generated marketing copy, messaging pillars and other materials based on its research. It also created prompts for image-generation tools.
From Chatbots To AI Systems
Hoff also encouraged participants to centralize these workflows in shared environments such as Google Drive. A shared system can give teams access to the same AI processes and reduce redundant work.
The exercise demonstrated one of the key advantages of automated workflows: replication. Once a workflow is built, users can change the inputs without rebuilding the entire system.
For marketers, that shift could change how routine work gets done. Instead of using AI to complete isolated tasks, teams can build systems that manage entire processes from research to execution.
To learn more about upcoming events, check out the AI Edge newsletter, or secure your spot at the next workshop now.
