Matthew Prince pointed to Cloudflare Radar showing that bot traffic has just passed human traffic for worldwide HTML requests.
That is a strange line to cross.
The important part is not the chart. It is what the chart says about the reader.
The web we know was built around a person with a browser. Someone searches when they need something, opens a few links, skims, copies a sentence somewhere, forgets the source, and comes back later to search again.
That is still how most of us use the internet.
Looking for a hotel. Checking a company. Trying to remember which regulator said what.
A lot of the internet adapted to that behavior. Pages were written to rank. Headlines were written to get clicked. Analytics counted visits. Companies learned to explain themselves to someone moving fast, half-paying attention, with too many tabs open.
That worked because the reader was human.
Agents do not read that way.
An agent can keep checking the same source. A changed sentence on a company page is visible. Today’s transcript can sit next to last quarter’s. A person, a filing, a product, a rule, or a market can stay on watch without anyone remembering to come back.
It also reads things people usually skip. The footnote. The correction. The timestamp. The policy update. The page nobody linked to.
Not all of this traffic is agentic. Some of it is crawlers, monitors, and old automation. But agents make the shift feel different, because software is becoming a reader that can actually do work.
That is not really search anymore.
Search is a person asking a question. This is more like a standing watch.
And once the reader changes, the material has to change too.
Humans are forgiving readers. We can deal with messy pages. We skim around clutter, make guesses, and usually figure out what matters from context.
Machines can process more, but they depend much more on the structure underneath.
A claim without a source is hard to trust. A vague summary becomes weak context. If the same fact shows up ten times with no clear origin, it turns into noise. If a record is stale, the mistake can keep moving.
This is why we care so much about the basic pieces at Synorb: claims, sources, timestamps, stable IDs, provenance, taxonomies, and records that can be followed over time.
Those can sound like database details. They are not.
They are what make information usable when an agent is actually trying to do work.
An analyst watching public companies does not need another pile of scraped pages. A procurement team tracking supplier risk does not need a hundred loose summaries. A legal workflow monitoring policy changes needs to know what changed, when it changed, and where the change came from.
Same for founders trying to keep up with competitors, customers, investors, and markets.
The useful thing is not just more information. There is already too much of that.
The useful thing is context that can be traced.
What happened. Who it relates to. Where it came from. When it changed. Whether the claim can be followed back to the source.
That is the work.
Synorb turns public information into records machines can follow, compare, and reason over.
The web was built for people who search.
A lot of what comes next will be built for machines that keep up.
Source: Cloudflare Radar, Bot vs. Human worldwide HTML traffic.