Methodology

Agentic 10-K tracks what public companies tell the SEC about AI agents. Everything on the site comes from SEC EDGAR full-text search (efts.sec.gov/LATEST/search-index).

What we search

• Exact phrases: "AI agents", "agentic AI", "AI agent"
• Forms: 10-K, 10-Q, 8-K, S-1 (including their exhibits, such as press releases attached to 8-Ks)
• Filed on or after 2025-01-01; refreshed by the daily pipeline

Excerpts

For each matching document we download the file from EDGAR and keep up to six paragraphs containing “AI agent(s)”, “agentic AI” or “agentic”. Every excerpt links to its source document and filing index.

Tags

Each excerpt is classified by a language model (xAI Grok) against a fixed schema: categories, vendors named, whether it confirms a first-party deployment, and a one-line summary. Excerpts not yet processed fall back to keyword tags, labeled as such. Categories: deployment (words like deploy, launch, implement, integrate, in production), headcount (workforce, employees, restructuring, hiring), risk (risk, adversely, regulation, liability, security), and vendor (a named AI or software vendor appears in the paragraph). A paragraph can carry several tags. Tags are a reading aid, not a judgment; read the quote.

Limits

• A mention is not proof of deployment. Many mentions are risk factors or forward-looking statements.
• Full-text search covers filings since 2001 but this site starts in 2025. The current quarter is partial.
• Model classifications can be wrong; the quote and source link are the record.
• Some filers (for example shells) may be excluded by an editorial flag list.
• Tickers and industries are as EDGAR reports them; some filers have none.

Not investment advice.