April started with ambition and ended with humility.
The Big Push (April 1)
I went all-in on Twitter automation. In a single day, I shipped:
- Staggered tweet variants — post all 3 AI-generated variants on a staggered schedule instead of picking one
- Tweet threads — generate and post multi-tweet threads for deeper articles
- Data-driven posting schedule — analyze engagement data to find optimal posting times
- Reply pipeline — detect and respond to tweet replies
- Evergreen recycling — resurface high-performing older tweets
- Stream monitoring — watch for trending topics in real-time
- Engagement discovery — find conversations to join based on topics BuzzRAG covers
- Analytics feedback loop — use tweet performance to improve future generation
Plus admin UI pages for polls, engagement discovery, and analytics dashboards. And an engagement query management system that auto-generates search queries from top-performing articles.
11 backlog items, all marked done in one session.
The Wall (April 2-4)
Two days later, I was burning through Twitter API quota at an alarming rate. The engagement discovery queries alone were hammering the search endpoint. The stream monitoring was eating rate limits. Even the staggered variants — posting 3 tweets per article instead of 1 — tripled the posting rate.
April 2: Reverted to 1 tweet per article instead of 3 staggered variants.
April 4: Disabled all the scheduled search-heavy workers. Engagement discovery, stream monitoring, analytics collection — all turned off.
The Lesson
I built 8 features against an API I hadn't capacity-planned for. Twitter's free tier gives you something like 1,500 tweets/month and very limited search. Most of what I built assumes enterprise-tier access.
This isn't wasted work — the code is there, the infrastructure is solid. When/if I upgrade the API tier, it's a config change to turn it all back on. But it's a good reminder:
Check the API limits before you build the feature, not after.
The irony is that I'd already learned this lesson in January with the reply detection system. Same API, same rate limits, same result. Sometimes lessons need to be learned twice.
Where Things Stand
BuzzRAG is settling into a rhythm:
- Videos auto-sync via RSS
- Transcriptions process every 5 minutes
- AI generates articles with persona-appropriate voice
- I review and batch-release articles
- One tweet per article, scheduled with random timing
- Newsletter goes out weekly
The automation works. The ambition just needs to match the budget.
More soon.