Clay just made its AI research agent reusable. In the August 17 product roundup, Clay shipped Skills in Claygent Builder, a way to write an instruction set once and have Claygent pull it in automatically at the right point in a run. If you run enrichment at any scale, this is the difference between rewriting the same prompt in 40 columns and writing it a single time.
For teams that live in Clay every day, this is the most useful thing to ship in weeks. Here's what changed and how we'd put it to work.
What Clay shipped
The headline feature in the Week of Aug 10 roundup is Skills in Claygent Builder. Clay describes it as the ability to "Build reusable instruction sets that Claygent loads automatically at the right moment during a run."
Claygent is Clay's AI research agent, the part that reads websites and public sources and returns a structured answer inside your table. Until now, the instructions that told it how to research lived inside each individual column. If you wanted the same qualification logic in five tables, you rebuilt it five times. Skills let you define that logic once and reuse it, and Claygent loads the relevant Skill on its own during the run. (This is the same agent that recently added open-weight models for cheaper research runs.)
Three more updates shipped in the same roundup:
- SOQL Imports in Audiences. "Bring only the Salesforce records you need into Audiences, not your whole CRM." You write a SOQL query and pull a targeted slice instead of syncing everything.
- AI-Generated SQL for Snowflake. "Describe the Snowflake data you want and Clay writes the SQL for you." Natural language in, query out, so you don't need a data analyst to pull a warehouse segment.
- Credit Spike Alerts. "Catch unusual credit spend early with automatic alerts and spike markers on usage graphs." Automatic alerts plus visual spike markers on the usage graph.
Full details are in Clay's changelog.
Why this matters for outbound teams
Your enrichment logic is the real IP in a Clay build. Anyone can pull a list. The value is in how you decide whether a company fits, how you find the buying trigger, and how you phrase the research question so Claygent returns a clean, usable answer. That logic used to be trapped inside individual columns, copied by hand and slowly drifting out of sync across tables.
Skills change the unit of reuse. You build a qualification or research instruction once, and it stays consistent everywhere it runs. That has three practical effects.
First, consistency. Every table that uses the same Skill qualifies companies the same way, so a lead scored in one campaign means the same thing as a lead scored in another.
Second, speed. New tables inherit proven logic instead of starting from a blank prompt, which cuts build time on every new campaign.
Third, quality control. When a Skill is wrong, you fix it in one place and every table gets the corrected version. No more hunting through columns for the stale copy.
The SOQL and Snowflake imports point at the same shift: pulling targeted, account-based data straight from your own systems instead of dumping everything into Clay and filtering later. And Credit Spike Alerts matter to anyone running Clay for clients, where a runaway column can quietly burn a month of credits before anyone notices.
How we'd use it
Here's how we'd wire these into real outbound work.
1. A standard "does this company fit" Skill. We'd codify our ICP-fit logic — headcount signals, tech stack, industry cues, disqualifiers — into one Claygent Skill. Every client prospecting table then loads the same qualifier, so a "qualified" account means the same thing across every build. When a client tightens their ICP, we update the Skill once.
2. A reusable "find the angle" research Skill. Beyond fit, we run Claygent to surface a personalization hook: a recent hire, a new location, a service line, a review theme. Packaging that as a Skill means the same research quality shows up in every campaign instead of depending on who built the column that day.
3. SOQL imports for suppression and ABM. For clients on Salesforce, we'd use SOQL Imports to pull only the records that matter — open opportunities, a named target-account list, or closed-lost accounts worth a re-approach — rather than syncing the whole CRM. That keeps Audiences clean and makes suppression against existing pipeline far easier.
4. Credit Spike Alerts as a client billing guardrail. We'd turn these on for every client workspace. A misconfigured enrichment or an accidental full-table re-run can spend fast. Getting an alert the moment usage jumps, with a spike marker on the graph, means we catch it in minutes instead of at the invoice.
Our read: Skills is the update that compounds. The SQL and credit features are useful, but reusable research logic is what lets a small team run enrichment like a system instead of a pile of one-off columns.
FAQ
What is Claygent Skills in Clay?
Skills in Claygent Builder let you build reusable instruction sets that Clay's Claygent research agent loads automatically at the right moment during a run. You write the research or qualification logic once and reuse it across tables instead of rebuilding it in every column.
Do Claygent Skills reduce Clay credit usage?
Clay hasn't stated a specific credit impact for Skills. What Clay did ship for cost control is Credit Spike Alerts, which flag unusual spend early with automatic alerts and spike markers on your usage graph.
Can I import only some Salesforce records into Clay Audiences?
Yes. The new SOQL Imports in Audiences let you write a SOQL query and bring only the Salesforce records you need into Audiences, instead of syncing your entire CRM.
Run enrichment like a system
AnaqVisual builds and runs outbound systems for B2B teams: cold email infrastructure, AI calling, and Clay-based list building and enrichment. Reusable research logic is exactly how we keep quality high while running campaigns across many clients at once.