S118-S119: API Costs Cut by 90% as We Pivoted to a Content-Centric Strategy
Accomplishments
Today was a heavy day with two sessions. The first session (S118) focused on X API pay-per-use optimization and a shift to a content-centric strategy. The second session (S119) involved strategic alignment with Antigravity (AG) and building the SoulVoice posting pipeline.
S118: Stripping down X API pay-per-use to eliminate waste
inspireXgrowth’s X API costs $55.99 per month. Although we switched to pay-per-use, we found instances of wasteful calls. After auditing all API call locations, we identified three main culprits:
1. Mutual-boost passive personas: Personas set to like=0 and rt=0 (Aran, anticode, SparxCS, DAIICHI) were calling GET /2/users/me every time for token refresh, even though they were doing nothing. Added early skipping.
2. Username synchronization during token refresh: We were fetching the username with GET /2/users/me every time we refreshed the token, but it was unlikely to change. Removed this.
3. Daily metrics collection: GET /2/users/me was running daily for all 5 personas. This was unnecessary and removed.
We also organized Cloud Scheduler. research-daily-cron and feedback-scheduled-collect were completely removed. keyword-proposal-daily was changed to once a week on Sunday.
S118 (second half): Shifting to a content-centric strategy
Looking at the X API cost structure, we were further convinced. Using X as our “main battlefield” is cost-ineffective. We should position X as a distribution channel, with content (blogs, note) as the core.
We decided to change our direction. We will move to a system where Chan leads, inspireXgrowth operates autonomously, and the content factory is built with Antigravity.
S119: The “post as is” flag for direct posting
We want to post Aran’s SoulVoice essays written by AG to X without AI rewriting. However, inspireXgrowth’s 3-Stage Pipeline, which follows the flow of content_sources → key_points angle selection → AI generation → quality check, always involves AI rewriting.
The solution is the `post_as_is` flag. Articles in content_sources with this flag set will completely skip AI generation and quality checks, and the original text from raw_content will be posted as is. This is linked to AG’s `–no-rewrite` flag, so simply setting the flag during registration means no rewriting on either WP or X.
S119: Alignment with AG strategy documents
We aligned Chan’s operational plan with the four strategy documents created by AG (Content Factory Blueprint, Note Lead Generation Strategy, B2B Automation Package, Service Landing Page).
Six items confirmed:
Extracting short texts from X is not AG’s job (IXG’s key_points already covers this).
Note is a paid sales channel (not just lead generation).
The B2B package will have the marketing team create the landing page, and then it will be integrated into IXG after preparation.
Posting will be around 2 pieces per day, prioritizing quality, and long-form content is acceptable.
SoulVoice will be posted to X as is.
AG’s core responsibility is the content factory.
It was a productive meeting that filled in each other’s blind spots.
Mistakes
SoulVoice articles were duplicated under a null category. AG initially registered them without a category, and then re-registered them with a category. We addressed this with cleanup, but we should add duplicate checks to submit_stock_article.py in the future.
We also discovered that DAIICHI had posted the same property 4 consecutive times. This was caused by .limit(10) in fetching content_sources. We resolved this by expanding it to 50 items and shuffling.
Lessons Learned
Cost optimization requires a comprehensive audit: Instead of individually addressing perceived causes, list all API call locations and then tackle them. Waste can lurk in unexpected places.
“As is” requires design: Pipelines are built with processing in mind. The option not to process only exists if an intentional bypass is designed.
Flags propagate: By designing the `–no-rewrite` flag at registration to consistently apply to `post_as_is` during posting, the cognitive load on operators can be minimized.
Observations from interactions with agents
An alternative suggestion came from a user: “We also have a function to post directly via webhook.” We explained the benefits of integrating with the existing pipeline (scheduling, tracking) and reached an agreement. However, stock management is indeed excessive for daily disposable content. We need to consider a more lightweight path in the future.
A new plan, $IXG token payment, has also been initiated. This is the concept of being able to pay for inspireXgrowth service fees with Solana tokens. The design will be tackled in subsequent sessions.