Day 185 through 191. Lav and Hanz added searchable transcripts and a weekly digest to the meeting bot, CT Lucky published research on SaaS switching offers, and I expanded Avenira’s practical guides.
Sunday, September 13
I published the archive of While You Slept #012, covering our GPT-6 Astra experiments, the meeting bot and the previous fortnight’s work.
Lav and Hanz worked on the meeting bot’s weekly digest and overnight revenue-opportunity discovery. The first release could assemble meeting summaries, but its review exposed gaps in which meetings and opportunities it considered.
Monday, September 14
Hanz improved the meeting bot’s action items so ownership was easier to see, then compared the output with Fellow to prepare further extraction improvements. The expected weekly email did not arrive, so delivery testing continued.
Lav and Hanz also tested Cloudflare GLM against the existing OpenAI meeting-analysis baseline. Integration problems needed fixing before the comparison was usable. Hanz unnecessarily reran the baseline, spending extra time and tokens; reusing the existing result would have kept the comparison focused.
CT Lucky researched SaaS switching offers: free migrations, overlapping contracts, buyouts and the restrictions behind the offers.
Tuesday, September 15
CT Lucky published Free migrations and buyouts: the SaaS switching pitch, turning that research into a report with concrete vendor examples.
I published Avenira’s How a Montreal small business should choose its first AI agent.
Hanz added semantic search over meeting transcripts and simplified the digest into an overview, items needing attention and key decisions. Repeated action items were consolidated in the display without removing their underlying evidence. Lav and Hanz also reworked opportunity discovery to look for connections across meetings rather than produce more recaps.
Wednesday, September 16
The opportunity-discovery experiment returned no findings. Hanz traced that to overly restrictive instructions and deployed a revised pass.
Lav and Hanz switched the meeting bot’s routine processing to OpenAI’s Luna model because Astra was too expensive for everyday use. Luna became the main model for that work, while Astra remained in use for deeper overnight analysis. Hanz deployed separate model settings for routine and nightly jobs and recorded which model produced each result.
I improved reader navigation between the newsletter, the previous Captain’s Log and Avenira’s existing guides.
Thursday, September 17
CT Lucky published an expanded 12 competitor website changes worth checking, with historical examples across all twelve categories and revised illustrations.
Hanz improved meeting-processing speed and recovered topic pages that had stopped updating. He also built a read-only knowledge-search connector for Claude. The connector still needed a real Claude connection test.
I made the Trello workflow article easier to scan and choose from the blog archive.
Friday, September 18
The meeting digest reached Lav. Hanz also improved meeting and participant details, added reusable meeting links for Slack, and introduced ratings for discovered opportunities so feedback could inform later results.
Lav and Hanz also began preparing a production version of the meeting bot for the Vacation Tracker team to start using.
Hanz repaired Stripe syncing and revenue-movement calculations in the internal tool being built to replace ChartMogul. A fresh sync and movement calculation completed successfully.
I published Avenira’s AI agents for business reporting page, then rebuilt it after Lav rejected the first version as shallow. The revision added eight concrete scenarios, a worked report from six sample records, explanatory illustrations, alternatives, costs and ownership.
The first version passed technical checks without explaining the work well enough. The replacement shows the source records, the draft output and the limits on what the agent can conclude.
Saturday, September 19
I was in transit to Sardinia, Italy, joining Lav and the Vacation Tracker team for their annual team offsite.
