How do AI agents track new bills in Congress?

How AI agents track new bills in Congress

An agent tracks congressional bills by querying a bill feed at runtime and filtering by document type, then deduplicating by bill number rather than by document. Each bill exists as a chain of published versions, so treating every new file as a new bill will inflate the count several times over.

Legislative monitoring is an established paid market, which tells you the underlying problem is real. What changes with an agent is that the filtering, the part a human analyst charges for, happens inside the task rather than in a weekly digest nobody reads.

The version chain is what breaks naive tracking

A bill is not a document. It is a sequence of documents, each a published version of the same proposal as it moves through the process. GPO publishes every one of them.

The scale of that is easy to check. For the second session of the 119th Congress, GPO's bulk data repository lists 3,376 distinct House bill numbers spread across more than 3,900 published version files. An agent that counts files reports roughly 16% more legislative activity than actually happened, and an agent that alerts on files will alert repeatedly on the same bill.

The fix is boring and it works: key on chamber plus bill number, keep the most recent version, and treat a new version of a known bill as an update rather than a discovery.

Most bills never get past the first row. That is the single most useful prior an agent can hold about this data: introduction is cheap and common, and it is not news on its own.

Eight document types, and only some of them are laws

The bulk data collection splits into eight types, and conflating them produces confident nonsense about what Congress just did.

• hr, s are bills proper. These can become law.
• hjres, sjres are joint resolutions. These also can become law, and

they carry constitutional amendments.

• hconres, sconres are concurrent resolutions. Both chambers, no

President, no force of law.

• hres, sres are simple resolutions. One chamber, internal matters.

An agent asked "did Congress pass anything about X" that returns a simple resolution as an answer has technically found a document and substantively misled the person who asked.

What to query, and what to ignore

Coverage runs from the 103rd Congress (1993 to 1994) forward, so anything older needs a different source. For live monitoring, three filters carry most of the value: the search terms, a type split between current bills and enacted statutes, and a time window wide enough to survive a quiet week.

That last point matters more here than on a news wire. Congress does not publish continuously; it publishes in bursts around session days and goes quiet during recess. This wire defaults to a 7 day window for exactly that reason, and a narrow window returning nothing is usually a calendar fact rather than a bug. The same reasoning applies across every sparse government source, as covered in how recent agent data needs to be.

# newly published bills mentioning a topic, 14 day window
curl -i "https://thebotwire.com/bills/latest?q=artificial+intelligence&type=bills&since=14d"

# enacted statutes rather than pending proposals
curl -i "https://thebotwire.com/bills/latest?q=export+control&type=statutes&since=30d"

# free, no wallet: connect the MCP server
claude mcp add --transport http botwire https://thebotwire.com/mcp

The call costs $0.01 and needs no signup. Full parameters for every wire are at /docs. Bills are one of three federal document streams worth watching together: the rulemaking that follows a statute is in how agents track federal regulations, and the litigation that follows both is in how agents read federal court opinions.

FAQ

How soon after introduction does a bill appear?

Once GPO publishes the introduced version, which is typically the next business day rather than the moment of introduction. Congressional data is Event class, not Alert class: being an hour late costs nothing, being a week late means someone else already briefed the client.

Can an agent tell whether a bill is likely to pass?

Not from this data, and it should not pretend to. The feed reports what was published, not what will happen. What it does support is a useful proxy: a bill that has reached a reported or engrossed version has cleared a filter that most introduced bills never clear.

Why does a search for a well-known bill return several results?

Because each published version is its own document, and a bill that moved through committee has several. Deduplicate by chamber and number, then sort by version stage to see where it actually stands.

Is this the same as tracking the Federal Register?

No, and the distinction is the whole legislative-to-regulatory pipeline. Congress passes statutes; agencies then write the rules that implement them, and those appear in the Federal Register months or years later. Tracking one and calling it the other misses the lag entirely.

Sources

Collection scope, coverage from the 103rd Congress forward, the eight bill and resolution types, and the published version chain are documented in GovInfo's Congressional Bills collection. The file and bill-number counts quoted above were taken on 6 August 2026 from GPO's bulk data repository for the 119th Congress. Poll cadence and per-call pricing for this wire are published at thebotwire.com/health.

Related: How do AI agents track new US federal regulations? · How do AI agents read new federal court opinions?