01/ 06
Signal:a data event that tells you a prospect is in-market before they raise their hand: a pricing page visit, a funding round, a job change, a replyEnrichment waterfall:querying data sources in sequence until you get a value: Clay tries source A, then B, then C. Better coverage than any single providerGTM Engineering:treating your go-to-market motion as a software system: ICP as filter, signals as triggers, sequences as logic branches, Slack as the delivery layerIntent:behavioral evidence that an account is actively researching a problem you solve, before they contact sales or fill a formOutbound infra:the sending domains, mailboxes, DNS records, and warm-up cadence that keep your email out of spam and your main domain cleanICP filter:the exact definition of which companies and personas belong in your pipeline, expressed precisely enough that a machine can execute itWarm lead:a prospect who has taken a signal action (pricing page, reply, meeting request) that moves them ahead of every cold contact in your sequenceClay waterfall:a Clay table that pulls from LinkedIn, Apollo, and 50+ sources simultaneously, scoring and filtering prospects in real time before they reach a repEnrichment cost:the per-row credit spend of running a Clay workflow: easy to burn through budget fast when calling Clay from an AI agent without guardrails; controlling this requires knowing which enrichments to run, in what order, and when to stopSignal:a data event that tells you a prospect is in-market before they raise their hand: a pricing page visit, a funding round, a job change, a replyEnrichment waterfall:querying data sources in sequence until you get a value: Clay tries source A, then B, then C. Better coverage than any single providerGTM Engineering:treating your go-to-market motion as a software system: ICP as filter, signals as triggers, sequences as logic branches, Slack as the delivery layerIntent:behavioral evidence that an account is actively researching a problem you solve, before they contact sales or fill a formOutbound infra:the sending domains, mailboxes, DNS records, and warm-up cadence that keep your email out of spam and your main domain cleanICP filter:the exact definition of which companies and personas belong in your pipeline, expressed precisely enough that a machine can execute itWarm lead:a prospect who has taken a signal action (pricing page, reply, meeting request) that moves them ahead of every cold contact in your sequenceClay waterfall:a Clay table that pulls from LinkedIn, Apollo, and 50+ sources simultaneously, scoring and filtering prospects in real time before they reach a repEnrichment cost:the per-row credit spend of running a Clay workflow: easy to burn through budget fast when calling Clay from an AI agent without guardrails; controlling this requires knowing which enrichments to run, in what order, and when to stop
SORTED.
Book a Call
← All Posts

Clay Just Shipped Agent Plugins. Here's What That Means for Your GTM Stack (and Your Budget).

Clay published clay-run/agent-plugins this week: an official plugin that lets Claude Code, Codex, and Cursor call Clay directly via natural language. Clay solved the connection problem. The cost problem is entirely new.

What actually shipped

The repo ships a shared plugin source and a Clay CLI. Point any supported AI coding agent at it, authenticate with your Clay account, and the agent can run natural language commands against Clay: search companies, enrich people, query tables. No human in the Clay UI required.

The example in the repo shows an agent chaining two calls autonomously: find Series B fintech companies in NYC, then get CEO work emails. Both steps, no human in the loop, resolved in one prompt.

That is genuinely useful. The credit bill comes with it.

What this changes

Until now, the hard part of GTM Engineering was the plumbing: connecting Clay to AI workflows, building the integrations, writing the orchestration layer. Clay now handles that directly.

The question shifts from “how do I get an AI agent to call Clay?” to “what should it actually do once it can?” That is the harder question and the one Clay's plugin does not answer.

Clay gave practitioners a loaded gun. Knowing how to aim it, and how much each shot costs, is still the whole job.

The cost problem nobody is talking about

Clay credits run per-row and per-enrichment. The interface you call them from changes nothing. Calling Clay from Claude Code costs exactly the same as clicking through the Clay UI, and an AI agent will not hesitate the way a human does.

A human operator building a waterfall manually thinks twice before adding a phone number lookup to 10,000 rows. An AI agent running a natural language instruction like “find mobile numbers for all of these contacts” will execute that without a second thought. At $0.10–$0.30 per phone lookup, 10,000 rows is $1,000–$3,000 in one run.

That is not hypothetical. The teams that get burned first will be the ones who treated the CLI like a conversational interface and forgot there was a meter running.

How to keep costs under control

The same waterfall discipline works here. You just need to enforce it more deliberately when an agent is calling.

Run cheap providers first

Apollo basic company data: fractions of a cent per row. LinkedIn profile enrichment: similarly cheap. Work email lookup via Prospeo or Hunter: a few cents. Mobile number lookup: $0.10–0.30. VOIP detection: add-on cost on top. Build your waterfall cheapest-to-most-expensive and stop as soon as you have what you need.

Most expensive Clay waterfalls share one cause: a high-cost enrichment on every row, ungated. Conditions fix it: “only look up mobile if work email is missing,” “only run VOIP detection if phone was found.” Those conditions cut cost by 40–70% on most lists.

Set row limits before any agent-triggered run

Know your row count and per-row cost before any agent-triggered run. The enrichments you have configured set that cost. Check them first. A 500-row test with a 5,000-row full list behind it is a different risk profile.

Never let an agent trigger expensive enrichments without approval

Phone lookups, VOIP detection, and verified personal emails are the three categories that blow budgets. If you are using an AI agent to orchestrate Clay, explicitly exclude these from autonomous calls. Make them gated: require a human confirmation before any run that includes them.

Monitor the credit dashboard

Clay shows per-table credit usage. Check it after every agent-triggered run until you have confidence in what the agent is doing. Budget surprises are always traceable in Clay's usage logs. The discipline is looking before and after, not just after.

What the plugin does not replace

Clay's agent plugin is infrastructure. It removes friction. That is genuinely valuable. Less time wiring integrations means more time running plays. But infrastructure does not make decisions.

Teams that already know which plays work for their ICP get the most here: they execute faster. Teams that burn budget are the ones who confuse “easy to call” with “tells you what to do.”

Which waterfall to build, which signals to act on, how to combine Clay with the rest of the stack, how to keep credits predictable: that is GTM Engineering. The plugin removed one excuse for not doing it.

FAQ

What are Clay agent plugins?

Clay agent plugins are official integrations that let AI coding agents (Claude Code, Codex, Cursor) call Clay directly via natural language. The clay-run/agent-plugins repo ships a shared plugin source and CLI tools so agents can search companies and people, run enrichment workflows, and query Clay tables without a human navigating the UI.

How do I use Clay from Claude Code?

Install the Clay CLI from the clay-run/agent-plugins repo, authenticate with your Clay account, and point Claude Code at the plugin. From there, Claude can run natural language commands like "find Series B fintech companies in NYC and get CEO work emails" and it will chain the Clay calls automatically.

How much do Clay credits cost when called from an AI agent?

The same as when called from the Clay UI. Credit costs are per-row and per-enrichment regardless of interface. The risk is that AI agents chain enrichments automatically without the cost-awareness a human operator would have. A waterfall that runs phone lookups on 5,000 rows can cost $500–$1,500 in a single agent run without guardrails.

How do I control Clay credit spending when using AI agents?

Three rules: run cheap providers first and gate expensive ones behind conditions, set row limits before any agent-triggered run, and never let an agent trigger phone number or VOIP-detection enrichments without explicit approval.

Is Clay agent integration replacing GTM engineers?

No. Clay agent plugins solve the connection problem. They don't solve the strategy problem: which waterfalls to run, which enrichments to use for which ICPs, how to keep costs down, how to combine Clay output with Instantly and Attio. That judgment layer is where GTM Engineering still lives.

SortedGTM

We run the plays, not just the pipeline.

Clay waterfalls, credit-efficient enrichment, Instantly cadences, Attio routing. Built, documented, and running inside your stack. Book a call and we'll show you the exact setup.

Book a Strategy Call