GTM Operations

AI & agentic workflows

Use LLMs as per-row functions and research agents inside GTM workflows: classification, extraction, qualification, and first-draft personalization at list scale.

LLMs are what make signal-based personalization economically viable at volume. Used well, they do the mechanical half of research and drafting; used badly, they hallucinate firmographics into your CRM.

AI columns

The workhorse pattern: a natural-language prompt computes each row’s value. “Is this company B2B SaaS?” “Summarize what they do in one line.” “Draft an opener referencing their recent funding.” Cheap model for classification, strong model for reasoning.

Research agents

Agents like Claygent scrape and reason over the web to pull data points providers miss, straight into a cell. This is where you get the non-obvious detail that makes outreach land.

The honest state of autonomy

The “autonomous AI SDR that replaces the rep” pitch has largely not held up: fully hands-off sending burns domain reputation and misses judgment calls. The working model in 2025-2026 is human-in-the-loop, roughly one human supervising a couple of AI seats. AI owns the mechanical half; humans own judgment and deliverability.

Guardrails

Structured, deterministic outputs. Validation before anything writes to the CRM. A human approval step on anything that sends. Treat the model as a fast junior analyst, not an oracle.