Jeremy KayeProduct EngineerGet in touch
← Writing
Essay · AI · Agentic Engineering · Jun 2026

Automating my blog post workflow with an agentic loop skill.

I built Blog-Writer: an agentic loop skill that interviews me, writes a draft, converts it to TypeScript, and publishes with animations — all from a single chat.

Jeremy Kaye
Jeremy KayeJUN 13, 20264 MIN READ
blog-writer — main.ts
1// assembling output...
2
3 
idleblog-writer
Automating the pipeline

I'm a solo builder constantly working on shipping the next thing, and the backlog of "I should write about that" never gets cleared. It's not that it isn't valuable, I just never find the time to stop and write.

So like any proper builder, I automated it. You're reading its output right now.

◉ ◉ ◉

The Problem

My existing workflow to write a blog post was a bit tedious. When I wanted to write about a project or feature that I'd been working on, I'd point Claude to my project documentation and codebase, and ask it to help me come up with a draft. After that, I'd spend time revising it to make sure it expressed what I wanted to share and fit my voice.

Draft in hand, I would then use Cursor to convert the draft into a blog post page on my website, using a pre-existing page template. Lastly, I would use Claude once again to help me generate thumbnails, animations, and hero images for the blog post, and then publish the complete post.

While the workflow above is incredible compared to manually doing everything, there were still quite a few manual steps, and I just never had the activation energy to shift into blog writing mode.

◉ ◉ ◉

The Solution

To automate this process even further, I designed Blog-Writer, an agentic loop orchestration skill that manages three specialized sub-agents:

Blog-Draft conducts a structured interview between the AI and the user to learn about the blog post topic, audience, direction, and voice. Using that data, it then writes a full markdown draft for the user to review, revise, and approve.

Blog-Convert takes an approved draft and converts it into a TypeScript object, wires it into the site registry, and handles all the boilerplate: slug, kicker, reading time, body array, etc.

Blog-Animate creates animated React components and a thumbnail, and then adds them to the post. The skill works from a palette system, a set of defined animation concepts, and the draft itself, so there's visual coherence across posts without requiring individual design decisions.

3
Sub-agents
draft, convert, animate
2
Quality gates
humanizer + type check
1
Human pause
you approve the draft
~20min
To published
from topic to live
Blog publishing pipeline
One prompt → published post
Chatyour prompt
blog-draftagent
You approvehuman gate
blog-convertagent
blog-animateagent
Publishedlive on site
Chatrunning…
The full Blog-Writer pipeline — from a single prompt to a published post.

With the Blog-Writer orchestration loop in place, I can now simply fire up the skill, revise the draft it creates, and then the skill designs, creates, and publishes the post autonomously.

◉ ◉ ◉

Why an Agentic Loop and Not One Big Prompt?

One big prompt to write copy, generate TypeScript, and design React animations would produce mediocre everything, and optimize for none of them.

An Agentic Loop uses an orchestrator agent to manage specialized sub-agents that are each built to only do one thing. That specialization allows each sub-agent to focus on what it does best, and not get hung up on other parts of the larger task.

Each agent does one thing. That's why each one does it well.

In addition, the orchestrator agent works as a manager of the whole process, and checks the work of each sub-agent as the task progresses. This ensures that errors are caught and fixed within the agentic loop, and not manually by the user whenever they occur.

◉ ◉ ◉

The Actual Value

Automation for its own sake is interesting (at least to me), but that wasn't the point here. With the Blog-Writer skill, documenting my work while I am building it is now cheap enough to actually do.

Everything I am working on can now become a post. The interview step forces specifics out of me while I'm still inside the work, not reconstructing it from memory three weeks later when everything has blurred into "yeah I built a thing."

Writing was never the bottleneck, but going from builder mode to documentarian mode and back felt expensive enough that I kept deferring it. Now I describe what I built, and the system documents it.

The backlog of "I should write about that" has a drain now.

END
Jeremy Kaye
Written by Jeremy Kaye
PRODUCT ENGINEER · LOS ANGELES, CA
Subscribe
◉ Up next
Essay · AI · Agentic Engineering

Going from a single AI chat to a shipped product

How I built an agentic workflow that writes sprint plans, builds features, and tests them on its own, starting from one design conversation.