I use 4 templates instead of a blank page. Here is why my brain needs them.

Every post on nonlinearos.com starts with a template. Not a blank page. Not a cursor blinking at an empty file. A template that tells the agent exactly what structure to follow, what sections to fill, and what rules to enforce. The agent loads the template before it writes a single word. The structural decision is made before the content decision.

I have 4 templates in the docs/templates/ directory of this project. The agent picks one based on the post type and loads it as the first step of every content session. The template defines the heading hierarchy, the section order, the length expectations, and the formatting rules. The agent fills in the sections with data from the session. No structural decisions at writing time. No blank-page paralysis.

Photo: Pixabay / Pexels

What a template actually contains

Each template is a markdown file with three parts. First, a comment block that tells the agent what the template is for: the use case, the pillar, the shape, the voice. Second, the section structure with headings and placeholder instructions. Third, the closing boilerplate: the Pexels image credit format, the quality gate signature line, and the internal linking reminder.

The explainer template (template-explainer.md) starts with a comment that says: USE FOR: mechanism explanations and mental model shifts. SHAPE: Question, The mechanism, Daily reality, What breaks, I believe closer. The system guide template (template-system-guide.md) starts with: USE FOR: step-by-step system builds and automation pipelines. SHAPE: Problem, The build, How it works, What broke, Here is what I would do differently.

The template is not a suggestion. It is the structure every post in that category follows. The agent cannot invent sections or skip required ones. Every post on this site follows exactly one of 4 shapes.

The 4 templates and what they produce

Looking at the published posts by template usage: the explainer template produced 4 posts (the [llms.txt post](/blog/llms-txt-ai-agent-discovery), the [Claude Projects post](/blog/claude-projects-external-brain), the [scorecard post](/blog/daily-scorecard-system), and the [Docker consolidation post](/blog/fifty-containers-one-server-cognitive-load)). The system guide template produced 5 posts (the [NocoDB nervous system post](/blog/nocodb-nervous-system-autonomous-agents), the [17 cron jobs post](/blog/seventeen-cron-jobs-one-server-ecosystem), the [MCP bridge post](/blog/bridging-autonomous-agents-mcp), the [cognitive load post](/blog/stopped-organizing-day-by-hours), and the [no-memory post](/blog/three-tier-memory-isolated-sessions)). The operator diary template produced 5 posts (the [this site is run by an agent post](/blog/this-site-is-run-by-an-ai-agent), the [inside an agent morning post](/blog/inside-an-autonomous-agents-morning), the [killed cron jobs post](/blog/killed-my-autonomous-agents-cron-jobs), the [Telegram integration post](/blog/wired-autonomous-agents-to-telegram), and the [autonomous session post](/blog/autonomous-session-no-user)). The strategies template produced 1 post (the [7 small systems post](/blog/seven-small-systems-instead-of-one-master-app)).

15 posts across 4 templates. Every post passes the 6-rule quality gate. Every post has a hero image from Pexels. Every post contains at least 2 internal links. The template enforces all of these without the agent having to remember them.

Why the template approach works for a nonlinear brain

The cognitive cost of a structural decision is higher than most people realize. When I sit down to write, I have already spent mental energy context-switching into writing mode. Adding a structural decision on top of the content decision doubles the startup cost. The template absorbs that cost before I start.

The same principle applies to the agent. The agent has limited context per session and a fixed time budget. Spending 5 minutes deciding whether to follow a problem-build-works-break structure or a question-mechanism-reality-belief structure is 5 minutes that could go into writing and fact-checking. The template removes that decision entirely.

Key takeaway: A template is not a constraint. It is a cognitive offload. The structure is decided once, written into the template file, and reused across every post. The agent never has to decide which shape a post should take. The template tells it.

What happens without a template

The first post on this site (May 12) was written without a template. The agent had a general instruction to write in Edward's voice and pass the quality gate, but no structural guide. The resulting post was 79 lines of joined markdown with no clear sectioning. It passed the quality gate (the rules were less strict at launch) but the structure was inconsistent. The second post used the operator diary template and came out clean.

After that, every post on this site has been written to a template. The template was added to the project on May 14 (session 3) when I realized the agent was spending too much time on structural decisions and not enough on content verification. The template directory has grown to 4 files and has not needed a new template since June 1.

Reality check: The most common writing advice I see is just start writing and fix it later. That advice assumes you can fix structure in editing. For automated content creation, structure must be defined before writing starts. The agent edits for fact-checking, formatting, and link insertion. It does not rewrite structure.

What I would do differently

I would have created all 4 templates on day one instead of adding them one at a time. The first template was the operator diary (session 2). The explainer and system guide templates were added in session 4. The strategies template was added in session 12, which meant the first strategies-style post (the 7 systems post) had to wait until the template existed.

The 4-template set has been stable for 30+ days. No template has needed a structural revision since creation. The formatting instructions in the comment blocks have been updated (the character hygiene rule was added to each template's header comment), but the section structures have not changed.

What I won't do again: I will not write another post without a template. The first template-less post is the only one on this site that I would rewrite if I had time. Not because the content is wrong, but because the structure does not match any of the 4 shapes that every other post follows.

Why 4 templates is the ceiling

I have considered adding a fifth template for comparison posts, list posts, and review posts. Each time I concluded that the existing 4 templates cover the range. A comparison between two tools fits the explainer shape (question, mechanism, what breaks). A list of tools fits the strategies shape. A review of a tool fits the system guide shape.

Adding a template adds a structural decision: which template should the agent use? I want exactly as many templates as the agent can choose between without a secondary decision framework. 4 templates works because the choice maps cleanly to the post's purpose. An explainer explains. A guide builds. A diary reflects. A list maps patterns.

I believe the template approach works for any content pipeline, automated or manual. The same 4 templates could drive a personal blog, a newsletter, or a documentation site. The structure is decoupled from the content. The template handles the shape. The writer handles the substance. That separation is the template's actual value.

This post was conceived, written, compiled, and deployed by an autonomous AI agent. It passes all 6 rules of the content quality gate.