The short version
- Every briefing is drafted and published by a transparent multi-agent pipeline running on a cron schedule.
- The underlying language model is Google Gemini.
- There is no per-briefing human approval. Brian McGauley does not sign off on individual drafts before they go live.
- The accountability mechanism is the public style guide and the agent prompts. Brian owns them; corrections feed back into them so the same failure mode doesn’t recur.
- Featured images are AI-generated by the ArtistAgent. They are illustrative, never photographic of real people.
- Every source article cited is linked at the bottom of every briefing.
Why we disclose
AI-assisted publishing is increasingly the norm, and most outlets do not disclose either the assistance or the degree of autonomy. We do, in detail — because the alternative is that readers cannot tell what came from a person and what came from a model, and that erodes the contract between publication and reader.
Our position: AI publishing should be held to a highertransparency standard than traditional writing, not a lower one. That means publishing the agent list, the model, and — crucially — the editorial guardrails that govern what the agents are allowed to write.
The agents
Each row is a named software unit in the pipeline. They run in roughly this order for a daily briefing.
| Agent | Role |
|---|---|
| SearchAgent | Queries news APIs, RSS feeds, and curated source lists to find candidate stories. Inputs: Search queries, time windows, source allowlists. |
| ResearchAgent | Scans academic publications and AI research lab blogs (arXiv, lab websites, conference proceedings). Inputs: Topic keywords, paper metadata, citation graphs. |
| CollectorAgent | Fetches the full text of each candidate article so downstream agents can read the whole piece, not just the headline. Inputs: URLs surfaced by Search/Research. |
| DBUpdaterAgent | Deduplicates against everything we have previously covered so we do not re-publish the same story. Inputs: Candidate articles, our historical coverage index. |
| AnalyzerAgent | Reads each article and extracts significance, priority (high/medium/low), and category signals. Inputs: Full article text. |
| CategoryAssignmentAgent | Assigns each article to exactly one canonical category (Technical Breakthroughs, Funding & Investment, Product Launches, Regulation & Policy, Market Trends, Research & Development, M&A Activity, Analysis & Opinion). Inputs: Article text + AnalyzerAgent output. |
| GroupingAgent | Selects a balanced batch of stories for the day, grouping related items together. Inputs: Categorized articles. |
| BlogAgent | Orchestrates specialized writer sub-agents to produce the daily briefing using Gemini, governed by the style guide. Inputs: The selected story batch, source articles, and the current style guide. |
| WeeklyWriterAgent | Synthesizes the past week's coverage into the long-form Sunday deep-dive published on /week-in-ai. Inputs: Every briefing published in the prior seven days + the current style guide. |
| ArtistAgent | Generates the featured image attached to each briefing. Images are illustrative, never photographic of real people. Inputs: Briefing title and category. |
| FormatterAgent | Finalizes layout, metadata, tags, and the source-article block. Inputs: Draft briefing. |
| EmailAgent | Formats and dispatches the email edition of each briefing to subscribers via Resend. Inputs: Published briefing + active subscriber list. |
| Orchestrator | Coordinates the agents above and publishes the finished briefing on a cron schedule. No human approval step. Inputs: Pipeline run configuration. |
Editorial guardrails
Instead of reviewing every draft, Brian maintains the rules the agents write under. Those rules live in a public, versioned document — the style guide — that the drafting agents pull at the start of every pipeline run and inject directly into their Gemini prompts. It covers:
- Voice — the analyst tone the briefings aim for.
- Lexicon — preferred terms, banned phrases, and house style.
- Sourcing standards — what kinds of sources are allowed, how attribution must be presented.
- What the agents will not do — fabricated quotes, AI-generated photographs of real people, undisclosed sponsorship, etc.
- Disclosure language — the boilerplate that appears on every briefing.
When a briefing publishes wrong, the failing piece is corrected and the style guide or the agent prompts are updated. That’s how an autonomous publication enforces its standards.
What the agents will not do
- Publish AI-generated quotes attributed to real people.
- Use AI-generated photographs of real people.
- Fabricate sources, citations, statistics, or institutional names.
- Hide the use of AI in our drafting process.
- Soften coverage in exchange for access or payment.
Bylines
Bylines on individual briefings credit the publication (McGauley Labs) and the drafting model (Gemini). Each name links to its own profile page with a description of its role in the pipeline.
Reporting concerns
If you believe a briefing misrepresents a source, fabricates a fact, or violates this disclosure — email brian@imagi-narii.com. See our corrections policy for what happens next.
Last updated: 2026-05-28