A government relations partner made a daily news brief run itself, built for

Project Description: - Achievements: AI Mastery's coursework shaped this project in two concrete ways, on top of the year-old methodology it started from. First, working through the material on MCP reliability, I migrated the brief's Airtable connection from a ChatGPT-based agent to Claude, eliminating a daily manual re-authorization step that had been silently defeating the entire point of automating it. Second, using the ICP and client-archetype methodology taught in the program, I had Claude review and revamp the brief's actual sources and topics — not just tag the output after the fact — so its coverage now matches both V&A's current client base and its ideal-client profile, and cross-tags every relevant item against that same archetype framework for the firm's existing Trigger Intelligence Monitor. The first full live run under the rebuilt pipeline (August 5, 2026) generated the brief, scored it, wrote it to Airtable, and tagged it correctly against both new fields — no manual step anywhere in the chain. - Tools: Claude (research, veracity scoring, direct Airtable writes), Airtable (News Data base — Daily Briefings and Recipients tables), Make.com (automates the original morning distribution to my partners and me, unaffected by this work), and next: Claude Code Routines, to fully automate the daily run. - Methodology: Most news, particularly anything touching government, isn't traditional fact-first journalism — it's spin, from whichever direction it's coming from. I used to handle that myself by reading one left-leaning and one right-leaning paper, both fairly centrist, and doing the discernment in my own head. About a year ago, independent of AI Mastery, I decided AI could do that filtering better than I could manually, with two specific goals: strip out spin and rhetoric from either side, and separately score the veracity of what's left. The brief started as a ChatGPT project — I'd manually approve the output before it went into Airtable, and a Make.com automation handled sending it to my partners and me every morning. When ChatGPT introduced agents, I automated the generation step to run at 5 a.m., but the Airtable authorization would time out before I got to it — I usually wasn't available until 6 or 7 a.m. — so I still had to manually re-run the authorization every day, which defeated the purpose of automating it in the first place. That's where AI Mastery actually changed the project, in two distinct ways. Working through the coursework on MCP reliability, I realized MCP integrations aren't interchangeable — the same Airtable connection behaves very differently depending on which platform is driving it, and Claude's was noticeably more reliable than ChatGPT's. I migrated the whole pipeline to Claude a couple of days ago, and the entire process is now automated with no manual step anywhere. Separately, the program taught me to use AI to build out a real ICP, client-archetype set, and business development plan for the firm. Rather than treat that as a separate exercise, I had Claude go back into the brief's underlying instruction set and revamp the actual sources and topics it searches, so coverage is now built around both our current client base and our ideal-client profile — not just whatever seemed broadly relevant to government relations. On the output side, I also had it tag every brief item against that same archetype framework in two new Airtable fields, so the same system that used to just summarize the news now flags when something is actually relevant to a specific client relationship or business-development opportunity, feeding the firm's existing Trigger Intelligence Monitor. That required a real schema change in Airtable — the two new multi-select fields, ICP Archetypes Triggered and Client Relevance — which I confirmed and then had built directly. The first full live run, on August 5, 2026, generated the brief, applied the veracity and source-triangulation scoring, wrote it to the table, and tagged both fields correctly. - Learnings: The MCP layer matters as much as the model behind it — the same task, the same intent, produced a system that quietly needed daily human intervention on one platform and didn't on another, and that difference came directly from something the coursework pointed me toward, not something I'd have caught on my own. The other lesson, from the ICP work: intelligence tools get more valuable when they're built around who the information is actually for, not just tagged for relevance after the fact — revamping the sources and topics upstream produced a noticeably more useful brief than tagging the old output ever could have. - Relevant URLs: Internal tool, not public-facing.