We’re not anti search.
Search is great at one thing: discovery.
If you don’t know what you’re looking for, or you’re exploring a long tail topic, search is the fastest way to get a foothold.
The problem is using search as the default context engine for an agent that has to do a job repeatedly.
That’s like using “ask the internet” as your database.
Where search fits
Search fits when:
- the user’s question is genuinely open-ended,
- the topic is brand new and your curated streams don’t cover it yet,
- you need to find a new source channel to add to your supply chain,
- you’re doing one-off background reading.
In other words: when you want breadth and you can tolerate noise.
Where search hurts
Search hurts when:
- you need repeatability,
- you need time-aware guarantees,
- you need provenance and evidence typing,
- you need to produce a trail a human can audit.
Those are the “agent workflow” requirements.
Search can’t reliably satisfy them because it isn’t designed to. It’s designed to find pages, not deliver structured context objects.
The pattern that works
Use search to discover sources.
Then graduate those sources into streams.
Once you have streams, agents can operate on context that is typed, time-aware, and addressable. Hanging Context makes the public aggregate layer visible. Synorb provides the detailed retrieval layer.
Search stays in the toolkit. It just stops being the foundation.