Key Concepts
Pattern: One of four categories that nearly every repeatable marketing task falls into: monitor, research, create, or report. Naming the pattern tells you how to build the agent and signals when a request is actually more than one job.
One-job rule: Each agent should do one job. When an agent monitors, drafts, and reports, it has three places to go wrong and no way to tell which one did. Scope to one job, get it running, then chain the next agent behind it.
Goal vs. job: A goal is an outcome (“keep us on top of the competition”) and usually hides several patterns. A job is a specific, checkable task (“check these five competitor pricing pages every morning and tell me in Slack when a number changes”).
Quick Reference
The four patterns
- Monitor: Watch something and tell me when it changes.
Fits: work where the answer is usually “nothing happened” but missing the one time it did is costly.
Example: a brand coverage monitor searches your brand and competitors for the last seven days and sorts results by sentiment. Negative coverage triggers an urgent Slack alert with a response brief. Routine coverage goes into a daily digest. - Research: Go find out X about Y and come back with a summary.
Fits: work where you know the question but not the answer, and the finding is the deliverable.
Example: a question miner takes a topic and returns top questions, themes, intent breakdown, unanswered questions, and recommended content angles. - Create: Turn this input into that output, in our voice.
Fits: work where the output shape is the same every time and only the input changes.
Example: a page optimizer reads a URL against your target prompts and returns a rewrite with self-contained sections, a summary near the top, an FAQ block from real buyer questions, and your brand language.
Note: this is the only pattern that publishes, so a person reviews it first. - Report: Pull these numbers on this cadence and format them this way.
Fits: anything someone rebuilds by hand every week and pastes into a deck.
Example: a weekly brand health report combines performance metrics and citation data on a schedule.
Splitting a request into patterns
- Competitor request example: watching pages for changes is a monitor, working out why a change matters and the counter is research, writing the updated battlecard is create, and summarizing everything that moved is a report.
Picking the first task to build
- Frequency: how often does it happen?
- Daily or continuous: high priority
- Weekly: worth considering
- Less than weekly: low priority at the start
- One-off or “whenever it makes sense”: deprioritized
- Impact: how notable is it? A task done once a quarter that everyone dreads and that blocks others can outrank an easy daily task.
- Frequency tells you how often the return arrives. Impact tells you how big it is.
Common failure to avoid
- Picking the biggest headache and handing the whole thing over at once. It may work on the first run, but when something comes back wrong you cannot tell which job produced it.