On August 19, 2025, Christina Blake — Director, Product at the Philadelphia-based digital marketing agency Seer Interactive — published a post on her firm's insights blog with a deliberately blunt headline: AI Blogs Are Mostly Trash (Unless You Build One Like This). The piece doubles as a working document. Blake describes the exact six-agent system she used to write the post she is publishing, and she is precise about what an "agent" is in her setup and what it is not.
The interesting claim in her account isn't autonomy. It's the opposite. The system she describes is a sequence of narrowly scoped prompts, each governed by a markdown file, with a human checkpoint after every meaningful step. The mobile element is small, specific, and quoted in her own words.
"To help, I built six agents — but don't let that scare you! In this example, 'agents' are just markdown files with personas and clear instructions. They give the LLM specific jobs to do; they're not fully autonomous systems."
The six agents, as Blake names them:
get_keyword_suggestions tool through Zapier MCPShe built it, by her account, in under an hour. The construction was conversational: she talked her process through with Claude, watched it propose a workflow, adjusted, re-prompted, and ended up with one markdown file per agent plus a master orchestration file. The orchestration file — reproduced in her post — is plain enough to be transcribed:
# Content Creation Workflow
## Overview This workflow transforms raw ideas into publication-ready [CONTENT_TYPE] using specialized modular agents.
## How to Start Share your idea in any format: rough thoughts, audio notes, problem to solve, etc.
The mobile component in Blake's setup is a single sentence describing the first stage of the pipeline.
"I usually start with stream-of-consciousness thoughts (often talking directly into Claude's mobile app). This agent takes my roundabout way of thinking and adds some structure."
She does not claim, anywhere in the post, to be dictating while walking, driving, or doing anything other than thinking out loud into her phone. The mobile element is a capture mechanism for ideas, not a full publishing surface. From there, the chain runs through her review steps on whatever device she happens to be on.
Blake's argument for specialization is short and operational. A single all-purpose prompt, in her telling, produces content that is "technically correct but feels generic and misses half the nuance you wanted." She compares the alternative to staffing — different departments, each with one clear job.
She gives a working example. While drafting a post on AI tools, her Devil's Advocate agent flagged a paragraph in a way she quotes back:
"An engineer could handle large data processing by adding compute through cloud functions — but you're writing for non-technical marketers who don't know how to do that. Are you being clear about the limitations for your actual audience vs the tool?"
That kind of mid-draft critique, she argues, "catches problems before publication instead of after."
The workflow is not autopilot. Blake lists four checkpoint passes — after the outline, after research, after Devil's Advocate, after the editor's voice pass — at which she has to approve before the chain proceeds. She is explicit about the prompt discipline this requires.
"Be explicit during review steps that there must be approval. Don't let your agents assume approval and jump ahead — you want control over the process at each step or you might need to start at the beginning."
The final agent in the chain copies content into a Google Doc from a Seer template and hands it off to the marketing team. The pipeline does not publish to a CMS. Publishing remains human.
Blake's framing throughout the post is that the value is not in the tools.
"Bad AI content is just bad content. When you ask an LLM to write something without bringing your own insights or perspective, you get generic regurgitation."
The post repeats the point with an italic insistence near the end: a reader who walks away wanting to buy Claude when they already have ChatGPT Plus has "entirely missed the point." The system she describes is a workflow design exercise that happens to run on an LLM. She names the orchestration logic and the prompt discipline as the artifacts that matter, and she ends the post with the markdown file that bootstraps the whole thing made available to download.
Blake's post was retrieved from Seer Interactive's insights blog, where it carries her author byline and headshot. Her role and authorship were cross-checked against her team page on the Seer Interactive site. All direct quotes — including the orchestration markdown block, the Devil's Advocate flag, and the description of mobile dictation — are reproduced verbatim from her published text. The stat citations she uses (Semrush, Orbit Media) are her sources, not this article's; they appear here only because she quoted them in the workflow context.
Join our monthly marketing magazine to receive the latest news and updates from our team of professional marketers and copywriters.
(Don't worry, your information will not be shared.)