Building Moneda as an agentic media company
Moneda AI is the public face of an experiment: can a small operator maintain trustworthy buying decisions better and cheaper than a traditional comparison publisher — with agents in the loop, and humans only where trust requires it?
What the site is
moneda-app.com is decision infrastructure for AI tools. Jobs → compares → tool entities. The matcher at /find/ is guided traversal; the compare page is the compiled verdict.
What the agent harness is for
Alongside the Astro site we run an operator harness (ai-media-company) whose job is not to invent winners for commission. It:
- audits portfolio quality and freshness,
- proposes claim updates with provenance,
- fans source changes into affected Golden decisions,
- recomputes recommendations with a frozen criterion engine,
- stops for human approval on material or winner-changing diffs,
- keeps affiliate economics out of scoring (editorial–commercial firewall).
Agents prepare work. Editors gate truth.
What we are not doing yet
We are not rewriting the stack for Temporal, LangGraph, or a Next.js rewrite. We are not expanding into travel, learning, or trading verticals until the AI-tools wedge proves autonomy and commercial intent.
Where this goes
If closed-loop maintenance works on a small Golden set, the same pattern — source → claim → entity → criterion → decision → recommendation — can eventually power embeds, matchers, and other surfaces without five separate content trees.
Until then: fewer pages, maintained harder. Start with a decision or read how Moneda maintains a buying decision.