There’s more to real-world AI work than asking questions.
Jason Michael Perry’s Builders Conference workshop, “Real World Applied AI Workflows for Right Now,” highlighted the importance of context when working with generative AI tools such as Claude and ChatGPT. He set up attendees to go beyond one-off chatbot queries to create repeatable systems that can carry out multi-step work.
“A prompt does one thing. A workflow does the whole job.”
Jason Michael Perry, PerryLabs
Perry, founder of Baltimore-based AI business service company PerryLabs, introduced the RTCCO prompt framework (Role, Task, Context, Constraints and Output) as a way to make AI prompts more reliable, accurate and auditable.
A vague prompt, Perry said, leaves too much for the model to guess. A stronger one tells the AI who to act as, what to do, what background information to use and what form the answer should take.
He demonstrated it with a career-prep prompt: asking AI to act as a hiring manager at a midsize tech company interviewing candidates for a junior cybersecurity analyst role, then giving it the candidate’s background as context.
The bigger shift, Perry said, comes when teams move beyond typed questions and start feeding AI unstructured information such as emails, PDFs, resumes, images and business data to answer more complex questions or generate more useful outputs.
From there, Perry moved into AI workflows: systems that combine triggers, context, actions and outputs to complete multi-step tasks automatically.
“A prompt does one thing,” Perry said. “A workflow does the whole job.”
He also previewed agentic AI tools such as OpenClaw, which can use skills, plugins and memory to complete tasks with less step-by-step direction. This kind of work is often called automation, he said, but that framing isn’t entirely accurate.
“Automation is not the part that’s difficult,” Perry said. “The part that’s difficult is the things that we can’t automate, and the parts that we can’t automate are what AI is very good at.”