The Plan Changed Before I Published a Single Post — Here Is Why
Building in Public | By Minjee Ganbaatar | Post 2
In my last post I laid out a complete AI-assisted blog publishing workflow. Seven steps. Four venture areas. One post per week per area. Bilingual promotion. Newsletter repurposing. A content pillar system. A running topic list. An evergreen bank. The whole architecture, mapped before a single post went live.
Then I looked at it honestly and changed it.
Not because the system was wrong. The system is sound. I changed the starting point — specifically, how much of it I am actually running in week one versus how much I am building toward over the next two months.
This post documents exactly what changed, why, and what I am betting on by making that call before implementation started. If you have ever designed something comprehensive and then had to decide whether to launch it all at once or in stages, this is that story.
The Comparison at a Glance
Here is the original plan versus the revised starting point, side by side:
| Element | Original Plan | Revised Start |
|---|---|---|
| Venture areas | All four from day one | One only — bookkeeping and tax |
| Posts per week | Four (one per area) | One |
| Topic planning | Full pillar system, running list, evergreen bank | Same system, applied to one area only |
| Drafting | Four posts per Monday session | One post per week |
| Promotion | Bilingual captions for all four posts | Bilingual captions for one post |
| Newsletter | Repurposed excerpt per post, four per week | One excerpt, one send |
| Odoo scheduling | All four posts scheduled Monday | One post scheduled |
| Timeline to full system | Implied immediate | Month two target |
| Weekly time commitment | Estimated eight to twelve hours | Estimated two to three hours |
| Risk of collapse | High — too many moving parts untested | Low — one loop, fully controlled |
The workflow steps themselves did not change. The seven-step process is identical. What changed is the scope of what I am running through that workflow each week while I prove it works.
What I Recognised
The original plan had a specific failure mode I know well from my own work history. It was not a bad plan. It was a complete plan — and complete plans that have never been tested in real conditions have a particular way of stalling.
The stall does not usually look like giving up. It looks like preparation. One more thing to set up before starting. One more prompt to refine. One more consideration to account for. The system stays in design mode indefinitely because design feels like progress and the gap between design and running feels too wide to close in a single session.
I have built enough workflows across enough ventures to recognise when I am designing past the point of useful preparation. The blog system crossed that line somewhere between the evergreen bank and the bilingual promotion templates.
The honest question was not "is this system right?" It was "what is the minimum version of this system that I can run this week, prove, and build from?" The answer to that question was not four posts across four venture areas. It was one post, one area, full workflow, documented.
Why Bookkeeping and Tax First
The choice of which venture area to start with was not arbitrary. I chose bookkeeping and tax services for three specific reasons.
The material is happening now. Client scenarios come up in real time. Filing questions arrive by email. CRA edge cases surface during actual work. I do not need to manufacture content ideas — they appear as a byproduct of the work I am already doing. That is the easiest possible starting condition for testing whether the drafting workflow produces something useful.
The audience is the most defined. Newcomers to Canada with small businesses or self-employment income. That is a clear reader with a clear set of questions. A well-defined audience makes it easier to evaluate whether a draft is actually on target or just plausibly on target.
The CTA is direct. Every post in this area points to getonlinebk.ca. There is no ambiguity about what success looks like — readers either engage with the service or they do not. That clarity matters when you are testing a new content system.
The other three venture areas are not deprioritised. They are queued. Education and language consulting is likely week five. KDP and CMCA follow after that.
What Did Not Change
It is worth being clear about this because the comparison table above could make it look like the original plan was scrapped. It was not.
The seven-step workflow is unchanged. Topic planning with content pillars, a running list, and audience segment signals — unchanged. The full prompt templates for drafting, metadata, and promotion — unchanged. The Odoo publishing process — unchanged. The collaboration with Claude as drafting partner — unchanged. The goal of four posts per week across all four venture areas — unchanged.
What changed is the timeline and the starting scope. The destination is the same. The route is staged rather than simultaneous.
What This Means for the Case Study
The original plan, if it had launched in full from day one, would have produced one of two outcomes: it worked immediately at scale and the case study documented a smooth implementation, or it collapsed somewhere in week two and the case study documented a failure. Neither of those is the most useful story for someone else trying to build something similar.
The revised plan produces a third outcome: a documented build, week by week, with real friction points visible as they appear and real adjustments made in real time. That is actually the more useful case study because it shows the process of proving a system rather than the result of running one that already works.
Weeks one through four: bookkeeping and tax, one post per week, every step timed and documented.
Week five: add education and language consulting if weeks one through four held up. That venture area will test something specific — whether the drafting prompts need to be redesigned for content that requires more subject-matter precision, or whether the same template holds.
Month two: KDP publishing and CMCA. Full four-area system running.
The Specific Commitment
One post. This week. Bookkeeping and tax. Full seven-step workflow. Every step timed.
The next post in this series will report exactly what happened: how long each step actually took, whether the draft sounded like me, whether anything in Odoo created friction I did not anticipate, and what I changed after the first run.
If you are building something and stalling at the planning stage, the question worth asking is not whether your system is complete. It probably is. The question is what the minimum provable version of it looks like and whether you are willing to start there.
I am. Week one starts now.
How This Post Was Made
This post was produced collaboratively with Claude (Anthropic) following the same process documented in Post 1. Claude drafted the structure and initial text based on the comparison between the original and revised workflow plans developed across our working sessions. I reviewed, edited for voice and accuracy, and made the final judgments about what to include and how to frame the reasoning.
The comparison table was built from the documented differences between the v1 article (full system, immediate launch) and the v3 article (full system, incremental build). Both documents exist in the project record and are referenced below.
Collaboration role: Claude (Anthropic) — structure, drafting, comparison table, editorial iteration Human role: Minjee Ganbaatar — factual authority, reasoning, voice, editorial judgment, framing decisions
References and Source Materials
Post 1 in this series:
Ganbaatar, M., & Claude (Anthropic). (2026, June 3). I'm automating my blog publishing workflow — and documenting it in real time [Blog post, Building in Public series]. minjeegan.com. The original full-system workflow documented in Post 1 is the baseline against which the changes in this post are compared.
Working document — full workflow with revisions:
Ganbaatar, M., & Claude (Anthropic). (2026, June 3). AI-powered blog publishing workflow v3 [Working document]. Produced in claude.ai. Contains the complete seven-step workflow, content pillar system, prompt templates, and the "Where I Am Actually Starting" section that prompted this comparative post.
Internal reference document:
Ganbaatar, M. (2026, May). Creating AI agent flow [Internal planning document]. Google Drive, personal workspace. The content agent architecture described in this document — full lifecycle handling with a human final pass — underpins the workflow design being implemented in this series.
Platform referenced:
Odoo S.A. (2026). Odoo Website: Blog module [Software]. https://www.odoo.com.
Tools referenced:
Anthropic. (2026). Claude (Sonnet 4.6) [Large language model]. https://claude.ai.
Minjee Ganbaatar is a bookkeeper, TESOL-certified educator, CELPIP preparation specialist, KDP publisher, and one of the founding members of the Calgary Mongolian Community Association. She is based in Calgary, Alberta and writes about the real experience of building multiple ventures as a solo operator.
Series: Building in Public — Blog Workflow Case Study, Post 2 of ongoing Tags: building in public, content strategy, AI workflow, solo founder, Odoo, blog automation, case study, AI collaboration, planning, starting small Previous post: I'm Automating My Blog Publishing Workflow — And Documenting It in Real Time Next post: Week One in Review — What the First Post Actually Took