I watched Google name our category last week.

On May 19 at I/O 2026, Liz Reid stood on stage and announced "information agents." They run in the background, watch the open web, synthesize updates, and notify you when something changes. Rolling out this summer to AI Pro and Ultra subscribers, twenty to two hundred dollars a month.

That announcement is the most important thing Google said all week. Most of the coverage missed it. WIRED led with "Search goes agentic". TechCrunch wrote three different posts on Spark, Halo, and Daily Brief. Antigravity and the new model launches ate the keynote chatter. The architectural story, the one that matters to anyone building or buying in this space, sat in a single product slide and a four-paragraph blog section.

I want to write the architectural story down.

Google named the category. Visualping has been running it for nine years.

The lineage Google did not draw

Watching information agents become a named consumer category took twenty-three years. Here is the timeline most coverage skipped.

For two decades the work of watching the web sat inside scattered subscription tools, internal corporate platforms, and a few stubborn consumer products. The category had no shared consumer name. I/O 2026 gave it one.

Naming a category is a real act. It tells reporters what to write about. It tells buyers what to budget for. It tells engineers what to ask their CTO for. A named category is a market.

What Google named is the demand side of the problem. What Google did not draw is the supply side.

Two architectures, not one

I keep reading takes that position Visualping as a "monitoring agent" somewhere underneath Google's information agents in a hierarchy. The right picture is two architectures next to each other, both serving the same job, both load-bearing.

Discovery agents scan the open web for a topic. You hand them a question or a theme. They roam, synthesize, and surface what looks relevant. Google's information agents, ChatGPT Pulse, Exa, and Perplexity all sit in this layer. Their input is a topic. Their freshness budget is wide and shallow.

Page monitors verify state on a specific URL. You hand them a page. They check it on a schedule, capture a visual or structural diff, classify the change, and route an alert. Visualping has sat in this layer since 2017. Their input is a URL. Their freshness budget is narrow and deep.

These layers do different things, and both are needed.

A discovery agent is the right tool when you do not yet know where the answer lives. It is the wrong tool when you already know the page that matters and need to be told the moment it changes. A discovery agent that broadly samples the web cannot reliably watch five thousand specific pages every five minutes with visual diffing, AI-driven importance classification, and a timestamped audit trail. That is a different system.

In a same-day snapshot of active Visualping monitors, roughly three in ten check their target page more than once an hour. That cadence is the working definition of the precision layer. In the same snapshot, 52,452 active monitors ran on a five-minute-or-faster cadence. SEC filings, regulatory dockets, tender boards, pricing pages, status pages where minutes matter.

The architecture story is simple. Discovery agents find the page. Page monitors verify the page. Both feed downstream agents that take action on what changed.

Two architectures: discovery scans the web, precision verifies the page.

What a sample of 1.87 million monitors knows

I should explain what the precision layer has been doing while the rest of the press caught up.

In a same-day snapshot on May 24, 2026, the sample of active Visualping monitors reached 1,874,911 jobs. In that same snapshot, 58.8 percent were owned by a team or workspace rather than a personal account. The precision layer is mostly an organizational capability.

In a separate profiled sample of 152,221 accounts, 11.4 percent come from companies with ten thousand or more employees. The largest enterprises were already paying for this work before Google named it.

In a thirty-day sample of 19.6 million completed checks, roughly one in ten returned a change. That is the rate at which automated, scheduled page monitoring earns its keep over manual refresh. It is also the rate at which any downstream agent (consumer or developer) gets a fresh, verified signal to act on.

I will share the more telling number from the customer side, not the platform side. One prospect, talking with our team in May, described what happened when their company tried to run broad LLM-based change detection on their own:

"When we've done this with, like, ChatGPT, it doesn't not give us answers. It will give us a change of changes, and it will say, oh, this was changed to this, from this to this. But then when we go and manually look it up, it's wrong."

That sentence is the substrate thesis in plain English. A general-purpose LLM running a broad scan can produce a story about what changed. Verifying that story, on a specific page, on a defensible schedule, with a screenshot and a timestamp, is a different system.

Another prospect, an engineering services firm, put the economics in one line:

"Right now we've got multiple project managers all spending time looking at these sites when we could have one person doing that."

Information agents move that "one person" down to zero. The precision layer is what makes the zero honest.

The same gap is opening for coding agents

Consumer information agents are not the only place this architecture matters. Anyone who has spent a weekend with Claude Code, OpenAI Codex, or Google Antigravity has hit the same wall.

You ask the coding agent to "watch this dashboard and tell me when the build status changes," or "page me when the pricing on this competitor's docs page moves," or "trigger this workflow when the regulator updates the docket." The agent reasons well. The agent writes code well. The agent has nowhere to put the actual watching. It needs a monitoring primitive it can call.

Visualping is that primitive. As of this week, the Visualping MCP is in public beta and works as a custom connector in both Claude and ChatGPT.

Add Visualping to your AI agent in four clicks.

In Claude: Settings → Connectors → Add custom connector → paste https://visualping.io/mcp/sse. In ChatGPT: Apps → gear icon → Create app → paste https://visualping.io/mcp/sse.

The agent can then create monitors, list monitors, and read recent changes by name. The REST API remains available on every plan, including Free, for webhook and code-first patterns.

Once that connector is wired, the architecture becomes practical. The discovery layer finds the page. The agent asks Visualping to watch the page. Visualping returns verified changes. The agent takes action.

I keep writing the word "primitive" on purpose. Information agents need monitoring primitives the same way browsers need TLS, databases need indexes, and search needs a crawler. The category is real. The substrate has to be real too.

The category formed in private notebooks long before Google named it.

What comes next for Visualping

Three commitments worth writing down in public.

We hold the precision layer. We keep building the things that make per-page monitoring durable: visual diffing on JavaScript-rendered pages, AI-classified importance, sub-minute cadences for the use cases that need them, timestamped evidence for the customers who need to defend a finding in court or in front of an auditor.

We make Visualping callable from any agent stack worth caring about. The MCP endpoint is public beta this week. The REST API is on every plan. We will keep adding to both.

We publish the architectural framework openly so the category forms with shared vocabulary. This piece builds on earlier work like agent search optimization and opens a short series: a definition of open-web agents versus page monitors, a comparison of the named players, a glossary for the information-agent era, a developer cornerstone on calling Visualping from coding agents. If the category Google named is going to mean anything in two years, the language has to be sharp before the marketing departments water it down.

Naming a category is not the same as owning it. The work of watching the web at the precision a real customer needs has been happening for nine years. It will keep happening.

The only thing that changes is that we no longer have to explain what the category is called.

Eric Do Couto is Head of Marketing at Visualping. He writes about the architecture of monitoring, AI agents, and the slow business of category formation.