Clay AI looks very different today than it did six months ago. Between July and September 2026, Clay shipped 12 releases in 12 weeks. Tables are still there, but the platform now includes a visual workflow builder, purpose-built agents, a CLI for coding agents, a GTM data layer called Audiences, and a Knowledge Hub that stores your business context.

If you last logged in to Clay a few quarters ago, much of what you knew still works. What changed is how much of your GTM motion Clay can now run, and how you build in it.

This guide covers what is new in Clay AI, what each feature does, where Clay says it is heading, and how to adopt the changes without rebuilding everything at once.

What is Clay AI?

Clay AI is a go-to-market data and orchestration platform. It combines a marketplace of 150+ data providers with AI agents that research accounts, enrich contacts, score leads, and trigger outbound, ads, and CRM updates. Clay is moving from a spreadsheet-style enrichment tool toward an end-to-end system that runs GTM plays and learns from their results.

In practice, Clay AI now covers four layers of a GTM system:

1. Data: first-party CRM and warehouse data plus third-party enrichment, unified in Audiences.

2. Logic: plays built as visual Workflows or classic tables.

3. Agents: Claygent and purpose-built agents for account research and execution.

4. Activation: syncs to sequencers, ad platforms, and your CRM.

For a deeper look at the data layer, see [link: Clay data enrichment guide]. To see how we build these systems for clients, visit [link: Clay services page].

The 5 Clay AI changes that matter most

Five releases change how GTM teams build in Clay AI. Here is the short version before the detail.

1. Workflows

Workflows is a visual canvas where each step of a play is a node. A trigger sits at the top, followed by enrichment steps and conditional logic. You can follow a single lead through every step, and teammates can review a play or take it over.

Tables are still useful, but anyone who has debugged a complex table knows how hard it is to trace conditional runs across dozens of columns. Workflows make the logic visible. Three details stand out:

• Code nodes run rule-based steps as scripts, so they are fast, cheap, and deterministic.

• No row or cell limits. Workflows run at Audiences scale, so a play built for 100 leads works on a million.

• A Runs tab shows status, cost, duration, and version history for each run.

Workflows also give the CLI a cleaner structure to read and write, which matters for the next two features.

2. Claygent and purpose-built agents

Claygent started as Clay's web research agent. It is now an agent builder: you describe what you want in natural language and Clay builds the agent. Around it, Clay is adding agents with specific GTM jobs:

• Account Research Agent reads unstructured data, such as signals and Gong call transcripts, and turns it into structured fields you can write to your CRM or use to trigger plays.

• Account Execution Agent reads that research, prioritizes accounts, and recommends the next play.

Clay also launched a skills marketplace in September 2026. Skills are plain-language instructions that teach coding agents how to run a GTM task, with Clay supplying the data. They are free to install and use Clay credits when they pull data.

The direction is clear. Claygents are moving from "find this on the web" toward agents that complete full GTM tasks.

3. The CLI and Agent Plugin

The Agent Plugin gives Clay a CLI, an API, an MCP server, and skills. You can build in Clay from coding agents like Claude Code, Cursor, and Codex without opening the Clay interface. Supported tasks include sourcing from data vendors, building prospect lists, routing inbound leads, and creating workflows in natural language.

This may be the most important change for new users. Clay has always had more features than most people could learn, and getting real value usually took an experienced GTM engineer. The CLI lowers that barrier. It also surfaces capabilities that are easy to miss in the UI.

A related release, MCP for Reps, brings Clay data and account research into tools reps already use, including Claude, ChatGPT, Microsoft Copilot, and Glean. For more on this way of working, read [link: Claude Code and skills for GTM and RevOps].

4. Audiences

Audiences is a data layer built for GTM. It unifies first-party data from your CRM and data warehouse with third-party enrichment from 150+ providers, plus signals like job changes and web intent.

What it replaces is the manual work of exporting CRM segments into tables and reworking them by hand. With Audiences you can:

• Build dynamic segments with simple filters that scale past 15 million records and refresh continuously.

• Sync segments to sequencers and ad platforms.

• Write clean data back to your CRM.

For teams that have hit table row limits on large enrichment runs, this is a major shift.

5. Knowledge Hub

Knowledge Hub stores your business context so Clay's agents can use it. Every workspace gets five standard sections: Company Description, Ideal Customer Profile, Buyer Personas, Interest Signals, and End Market. You can add up to three custom sections and upload up to 10 context documents.

Clay's global agent starts every task from the Knowledge Hub. That context shapes which accounts it looks for, how it scores them, and the copy it drafts. Agents can suggest edits, but changes only take effect after an Admin approves them.

This matters because generic plays produce generic results. The plays that win are built around your specific offer, ICP, and signals. Knowledge Hub gives agents that context by default. If your ICP is still fuzzy, start with [link: GTM strategy service page].

Where Clay AI is heading: a self-improving revenue engine

Clay's stated goal is to become a "self-improving revenue engine" that remembers which GTM efforts worked and suggests what to do next. Clay CEO Kareem Amin walked through this roadmap with the product team in July 2026.

Clay frames GTM competition in three stages:

1. Data. Better data wins. Clay's answer is its 150+ provider marketplace and Claygent.

2. GTM infrastructure. Integrated systems win. Clay connects the CRM, data warehouse, call recordings, enrichment providers, sequencers, and ad platforms in one place.

3. Learning loops. Agents run plays, learn from the results, and improve them. Teams can run these loops by hand today. Clay's stated goal is for the platform to run them on its own.

Every release above maps to that plan. Audiences and Workflows are the infrastructure. Knowledge Hub is the context layer agents need. The Account Research and Execution agents are the first purpose-built agents for the learning loop.

The practical takeaway: Clay AI is becoming less of a point tool and more of a GTM operating layer. That raises the stakes on how it connects to the rest of your system, especially your CRM. See [link: RevOps service page].

How to adopt the new Clay AI features without rebuilding everything

You do not need to migrate every table on day one. Start with the foundation, then move plays over one at a time.

1. Fill in your Knowledge Hub first. Agents are only as good as the context they get. Write a clear company description, ICP, buyer personas, and the signals that matter to you. Have an Admin review Clay's suggested edits.

2. Connect your data to Audiences. Bring in CRM and warehouse data before you build new segments. Clean data in means clean data written back to the CRM.

3. Rebuild one high-value play in Workflows. Pick a play you run often, like a signal-based outbound sequence. Use code nodes for rule-based steps to keep costs down.

4. Test agents on sample records. Clay lets you test on sample records and set credit limits before running at scale. Use both.

5. Try the CLI if your team already uses coding agents. If Claude Code or Cursor is part of your workflow, the Agent Plugin lets you build and edit plays without switching tools.

The common thread is system design. Clay AI can now run more of your GTM motion, which means gaps in your CRM structure, ICP definition, or routing logic show up faster. Fix those first. A RevOps leader or fractional CRO can then run and improve a system that was built to scale.

Partner UP is a Clay Advanced Artisan Partner. We design and build GTM systems on Clay, from Audiences and Workflows to CRM write-back and outbound infrastructure. Founder Leila Ergul Demir brings 20+ years of B2B sales leadership, including enterprise roles at Microsoft and Spotify. If you want help moving your team onto the new Clay, [link: book a call / contact page].

Clay AI FAQ

What does Clay do?

Clay helps GTM teams find, enrich, and act on account and contact data. It pulls from 150+ data providers and your own CRM and warehouse data, then uses AI agents to research accounts, score leads, and personalize outreach. Clay can sync segments to sequencers and ad platforms and write clean data back to your CRM. In 2026, Clay added Workflows, Audiences, Knowledge Hub, and a CLI, moving it from an enrichment tool toward a full GTM orchestration platform.

What is Claygent?

Claygent is Clay's AI agent builder. You describe a research or enrichment task in natural language, and Clay builds an agent to run it across your records. Claygent began as a web research agent. Clay now pairs it with purpose-built agents: the Account Research Agent turns unstructured data like call transcripts into structured fields, and the Account Execution Agent prioritizes accounts and recommends the next play.

Does Clay still have row limits?

Clay tables still have row limits, which vary by plan. Workflows and Audiences remove them for most large-scale work. According to Clay, Workflows run at Audiences scale with no row or cell limits, and Audiences segments scale past 15 million records with continuous refresh. For teams running enrichment on large account lists, this removes the need to split data across many tables.

What is the Clay CLI?

The Clay CLI is part of Clay's Agent Plugin, which also includes an API, an MCP server, and skills. It lets you build and run Clay plays from coding agents like Claude Code, Cursor, and Codex without opening the Clay interface. You can source data, build prospect lists, route inbound leads, and create or edit workflows in natural language. It makes Clay easier to learn for teams without a dedicated GTM engineer.