TL;DR
- The split: sales teams buy their AI agents, while marketing teams are starting to build their own. Nooks and Claude Code show both paths.
- Sales side: Nooks secured $500M in growth capital and launched Node, an assistant that drafts follow-ups a rep approves before anything sends.
- Marketing side: an Anthropic growth marketer with no coding background used Claude Code to cut ad creation from 30 minutes to 30 seconds.
- The shared rule: both keep a human on the send button. The agent does the prep work and the person keeps the judgment.
- The start: pick one workflow with a clear input and a clear output, write the rules down, then automate it.
Every vendor now sells an AI agent for sales and marketing. Almost none can say what the agent does on a Tuesday.
That is the real problem with the category. The word agent has turned into a label instead of a job description. It gets stapled onto chatbots, sequencers, enrichment tools and dashboards alike. Buyers end up comparing products that do completely different work. Two tools cut through that noise for me this month: Nooks on the sales side and Claude Code on the marketing side.
They come at the problem from opposite directions. Sales is getting agents it buys. Marketing is getting agents it builds.
Why sales and marketing need different agents
Sales work repeats in a predictable shape. Dial, connect, log the call, send the follow-up, chase the buying committee. The steps barely change. They look almost identical at a 50-person startup and a 5,000-person enterprise. That sameness is what lets a vendor productize the work and train on huge volumes of real sales conversations.
Marketing work repeats too, but the shape changes at every company.
Your brand voice, product claims, channel mix and reporting stack belong to you. A packaged agent built to write on-brand copy for 2,000 customers ends up writing generic copy for all of them. Generic copy sells nothing. So the buying decision splits at the middle of the funnel.
| Dimension | Sales agents (Nooks) | Marketing agents (Claude Code) |
|---|---|---|
| How you get it | Buy a platform | Build your own workflows |
| What it learns from | Large pools of real sales interactions | Your files, rules and data |
| Who writes the rules | Vendor defaults plus your playbook | You, in written skills and instructions |
| Human control | Rep approves every send | Marketer reviews every output |
| Best first job | Deal follow-up and dialing | High-volume production with strict rules |
The sales side: what Nooks is betting on
Nooks started in late 2020 as a virtual office for remote teams. Its founders, Dan Lee, Rohan Suri and Nikhil Cheerla, saw that sales teams were the most engaged users and pivoted to AI sales tools. The first product was a parallel dialer that skips ringing and voicemail, logs calls and takes notes.
The pitch rested on a number from its Series A announcement: SDRs spend 90% of their time on manual work like research, emails and lead finding. Vendor statistic or not, every SDR manager knows that shape.
On November 4, 2026, Nooks announced $500 million in growth capital from General Catalyst’s Customer Value Fund. This is not a typical equity round. The fund backs sales and marketing spend and gets repaid from the customer value that spend creates. Alongside the money, Nooks launched Node, which it calls the first AI personal assistant for sales.
Node’s first job is deal management. It tracks the promises made on calls, drafts follow-ups, flags buying-committee members who go quiet, and drafts moves to restart stalled deals.
Reps approve every send.
Nooks says 96% of emails drafted by its agents go out without edits. It also says more than 2,000 companies use the platform, including Notion, Deel and Replit. Treat those as vendor numbers until your own pilot confirms them. That design choice is the part worth copying. Node does the remembering and the drafting. The rep still owns the relationship.
Earlier this fall I wrote about how AI SDRs are destroying outbound when they blast without a signal or a human in the loop. Node is built the other way around. That design deserves attention.
The marketing side: what Claude Code changes
Claude Code is Anthropic’s agentic tool that runs in the terminal. It reads files, runs commands and writes code on your own machine. It was built for developers. That is changing fast. Marketers are now using it to build the tools their vendors never made for them.
The clearest example comes from Anthropic’s own growth marketing team. Austin Lau, a growth marketer who had never written a line of code, built two workflows with it. The first is a Figma plugin that generates dozens of ad variants from pasted headline copy, and it took about an hour to build. The second is a Google Ads workflow, triggered by a custom /rsa command, that drafts responsive search ad copy and exports upload-ready CSV files. Anthropic reports that what used to take 30 minutes per ad now takes 30 seconds.
Most marketers skip the part that kept the quality up.
The Google Ads workflow cross-references every input against Agent Skills that Austin wrote for Anthropic’s brand voice, product accuracy and Google Ads best practices. Those skills are plain written rules. The agent stays on-brand because the instructions are specific, and Claude Code applies them on every run without anyone retyping them.
The lesson for every marketing team: stop learning prompts and build AI workflows that carry your rules with them.
What it looked like on my own site this week
I run kokasexton.com with a crew of AI agents that handle my marketing ops. This week I used Claude Code to rebuild the pipeline that publishes this blog. I asked for one new article a day, researched and reviewed to my own standards. Claude Code read my standard operating procedures before it touched anything. It found a line in my own instructions file that still said the daily article job was retired. That stale line had made my agents refuse to write four times in October.
Then it added six review gates that every article must pass before it can publish.
The gates caught something I did not enjoy hearing. A post published the day before failed them. It had three lines of AI filler and a run of paragraphs that all had the same shape. Claude Code fixed the post, then found that the voice check itself had been called wrong in three scheduled jobs.
No vendor sells that agent, because no vendor knows those rules. They took years of publishing to write, and Claude Code turned them into daily checks.
Sales teams should buy their AI agents. Marketing teams should build theirs. The test is simple: does the job look the same at every company, or only at yours?
The rule both sides share
In a Stack Overflow pulse survey of 1,100 developers and working professionals from April 2026, 59% said they use agents at work, up from 31% a year earlier. Usage nearly doubled. Trust did not. Yet 63% still rarely or never let agents run on autopilot. These are the people closest to the technology, and they keep a hand on the wheel.
Nooks and Claude Code land in the same place from opposite ends. Node drafts the follow-up and the rep approves it. Claude Code builds the ad variants and the marketer reviews them. The agents producing results in 2026 take the prep work off your plate and leave the judgment with you.
Autonomy is something an agent earns with a track record.
How to start with AI agents for sales and marketing
You do not need a transformation program. You need one workflow that works, then the next one.
Follow-up emails after discovery calls. Ad copy variants from a keyword list. A weekly report from three exports. Skip anything as vague as “run my marketing.”
Brand voice, product facts, claims you cannot make, formats that must not change. In Claude Code these become skills and instruction files. In a platform like Nooks they become your playbook and approval settings.
Agents draft and people send until the edit rate drops. Track how often you change the draft. That number tells you when the agent has earned more room.
More emails and more ads are easy to produce and easy to ignore. Track hours returned to the team and pipeline created, as I laid out in how to measure AI marketing impact instead of output.
If you run sales, start with the platform and test the vendor’s numbers in your own pilot. If you run marketing, start with one tool you build yourself. Your brand rules were never going to ship in a box. Either way, the first win should be small enough to finish this month.
Buy the agent when the job looks the same at every company. Build it when the job only looks like yours. Either way, keep a human on the send button until the agent earns more.
Frequently asked questions
What is Nooks? Nooks is an AI sales platform that started as a parallel dialer. It now runs deal, outbound, expansion and coaching agents, plus Node, an assistant that drafts follow-ups for reps to approve.
Can marketers use Claude Code without coding experience? Yes. Anthropic’s growth marketer Austin Lau had never written code and built a Figma plugin and a Google Ads workflow with it. Expect a learning curve in the terminal, and start with one small, well-defined task.
Should sales and marketing use the same AI agent? Usually not. Sales workflows are similar across companies, so a purpose-built platform wins. Marketing depends on your own voice and rules, so a tool you configure yourself fits better.
