The 4 New Claude Code Features for GTM Operators

Here's what shipped, why it matters for GTM, and exactly how to use it

Anthropic has been shipping a whole host of new features in Claude Code that are ‘gold’ for GTM operators, but you’ve likely not heard or used them yet. What I’m going to show you in this post is some of the best features shipped over the past 90 days and how to use them as GTM operator, whether that’s marketing, growth, or sales.

What shipped in the last 90 days:

Routines — Claude Code now runs on a schedule, a webhook, or an API call. Entirely in the cloud. Without you. It’s available in the Claude Code web app.

Chrome integration — Claude can now drive your real, logged-in browser. HubSpot, Gmail, Gong, Notion allowing you to bring way more context into your Claude Code sessions. It’s available in the Claude Code desktop app plus the Claude in Chrome extension, available free from the Chrome Web Store.

Auto-Dream — a background process that consolidates everything Claude has learned about your projects. You’ll start to get much more context between sessions. It’s available in Claude Code v2.1.139 or later. Run claude update in your terminal to get the latest version.

Agent view and /goal — one screen showing every session running across your work, with the ability to define an outcome and let Claude run at it until it’s done. It’s available in Claude Code v2.1.139 or later on a paid plan — Pro, Max, Team, or Enterprise.

None of these are incremental. Together, they describe something that didn’t exist six months ago: a GTM operating layer that runs in the background, lives inside your existing stack, and ships work that used to require a developer, a project manager, and a junior analyst.

Below is exactly how these four features can run across a GTM team, with the commands, the prompts, and the plays you can set up now. These are examples. The purpose of this post is to show how much more valuable they make Claude Code as your GTM tool.

---

1. Routines: scheduling tasks

Requires: Claude Code on the web

A Routine is a Claude Code agent that runs on a schedule, a webhook, or an API call. Entirely in the cloud. Proactive vs. Reactive.

The competitor intel Routine

Routine helps with the workflows your team performs frequently. For example, every company wants competitive intelligence, especially now that so many new competitors are launched daily.

Every Monday, you can now set up a Routine and ask Claude to pull every blog post, product update, and G2 review from your top eight competitors published in the last seven days. It runs that content against your positioning doc. It sends a Slack message to the channel with three things the team should care about that week.

How to set this up:

Open the Claude Code web app. In the left sidebar, click New Routine. You will see two options, Local (runs on your laptop) and Remote (runs in Anthropic’s cloud). Select Remote so it runs without your laptop being open.

Give it a name: “Monday Competitor Digest.”

Every Monday at 6:30am, search the web for blog posts, product updates, and G2 reviews published by [Competitor 1], [Competitor 2], [Competitor 3] in the last 7 days. Compare each piece against the positioning doc at [file path or paste it inline]. Send a Slack message to [channel] summarising the 3 things our marketing team should care about this week, with a link to each source.

Set the schedule to Weekly, Monday, 6:30am. Connect your Slack under Connectors. Hit Create.

You now have a recurring agent. Pretty dope. A couple more.

Lightweight prospecting agent

A high-fit lead fills a form. A HubSpot workflow fires a webhook. Claude reads their company, their role, their recent activity, and any Gong call notes if they’ve spoken to the team before. It drafts a first-touch email in your voice. It drops it as a Gmail draft. A rep reviews and sends in thirty seconds.

How to set this up:

First, set up the HubSpot side. In HubSpot, create a workflow triggered by your high-fit lead form submission. Add a Send webhook action, HubSpot will generate a webhook URL for you.

Then, in the Claude Code desktop app, click New Routine. Select Webhook as the trigger type, you will see a URL generated for your Routine. Copy that URL and paste it into the HubSpot webhook action.

Use this as your Routine prompt:

When triggered, take the lead data from the webhook payload. Search Gong for any past calls with this lead's company — if found, read the last call summary and note any key context. Research the company using web search — funding stage, recent news, likely pain points. Draft a first-touch outbound email in [your voice/paste style brief] that references relevant context from any prior conversations. Keep it under 150 words. Drop it as a Gmail draft addressed to the lead's email from the payload.

Connect Gmail and Gong under Connectors. Save. Every qualifying form submission now produces a researched, personalised draft in the rep’s Gmail inbox.

The weekly content brief Routine

This is one of my favorites. Every Friday at 4pm: Claude reads my five best-performing SubStack posts, identifies the patterns, and drops a me a Slack message with five new content angles that map to what’s been working.

Click New Routine. Select Remote. Set the schedule to Weekly, Friday, 4:00pm. Paste this as your prompt:

Read the last 30 days of posts from [SubStack URL]. Identify the 3 patterns that made the highest-performing posts work, format, topic type, opening structure. Then generate 5 new content angles for next week using those same patterns. Post them to [Slack channel].

Connect Slack. Hit Create. Your Monday morning content block now starts with a brief already waiting.

** Note: Substack is a little trickier, I actually do this by having a repository of all my content in a folder and having it read that vs. the URL.

2. Chrome integration: the operator inside your stack

Requires: Claude Code desktop app plus the Claude in Chrome extension (free, Chrome Web Store). Paid Anthropic plan required.

Claude Code can drive your real, logged-in Chrome browser. This is two things working together: Claude Code and the Claude in Chrome extension. You need both installed. Once connected, Claude Code can click buttons, fill forms, read page content, and navigate across any site you’re already logged into, all driven from a single prompt.

How to connect it:

Install the Claude in Chrome extension from the Chrome Web Store. Sign in with your Anthropic account. Then open Claude Code and run:

/chrome

That’s the connection live. From this point, any prompt that involves your browser runs through your real Chrome session, with all your existing logins intact.

The stuck pipeline audit

“Pull every deal that’s been in stage three for more than 21 days. Read the last five notes on each one. Tell me which ones have a clear next step, which ones are waiting on us, and which ones need a manager to look at this week.”

This is a common task RevOps would do, now it’s just a prompt.

How to run this:

Make sure you are logged into HubSpot in Chrome. Open Claude Code and type:

Open my HubSpot CRM. Go to the deals pipeline view. Filter for deals in Stage 3 with a last activity date more than 21 days ago. For each deal, open the record and read the last 5 activity notes. Then give me a table with: Deal name, days in stage, last note summary, and a recommendation — Clear next step / Waiting on us / Needs manager review.

Claude opens HubSpot, navigates the pipeline, reads each record, and returns the table. You do not touch a single screen.

The Gong coaching brief

Before a team review: “Read the last ten Gong calls where we lost at the pricing conversation. Identify the three most common objections. Tell me which ones our reps handled well and which ones they stumbled on.”

How to run this:

Log into Gong in Chrome. Open Claude Code and type:

Go to my Gong account. Search for calls from the last 60 days tagged with ‘lost’ or ‘pricing objection’. Open the 10 most recent. For each, read the transcript section where pricing came up. Then summarise: the 3 most common objections raised, how reps typically responded, and where the conversation broke down. Give me a coaching brief I can use in our next team review.

This is a lightweight coaching use case. If you plugged in your sales methodology, it could start to easily grade these calls.

3. Auto-Dream: the system that gets sharper the longer you use it

Requires: Claude Code v2.1.139 or later. Run claude update in your terminal to get the latest version, then claude --version to confirm.

This new feature is one of my favorites. One of the problems with Claude is you spend twenty minutes at the start of a session explaining your brand, your audience, your context, your preferences. Then you close the session and have to start again. I showed how to solve this with my Secondbrain post.

Auto-dream is a background process that periodically reviews everything Claude has learned across your sessions, decisions made, context built up, patterns in how you work, and writes the most important parts to memory files Claude reads at the start of every new session. Stale information gets removed. Duplicates get merged. What’s left is a clean, current picture of your projects that loads automatically every time you open Claude Code.

The name comes from how REM sleep works. You don’t consciously decide what to remember from your day. Your brain reviews it, extracts what matters, and commits it to long-term memory while you sleep. Auto-Dream does the same thing for your AI sessions.

That is all it does but that one thing has a significant compounding effect. Context you built up in week one is still accurate and usable in week four, rather than buried under contradictions and outdated notes.

Auto-Dream is not on by default. You have to enable it manually. And it is still rolling out, some users will see the option immediately, others may need to update Claude Code to the latest version first.

Here is exactly how to check and enable it.

Step 1: Open your memory settings

Inside any Claude Code session, type:

/memory

You’ll see an option to toggle on or off.

Once turned on you can run /memory again and review what Claude is carrying. You will see your brand context, active project notes, and decisions, now cleaned up and organised. Edit anything that is wrong. Add anything it has missed.

And if you want proactive insights rather than just memory consolidation:

Run /insights. This is a separate command that analyses your last 30 days of sessions and surfaces patterns, where you are losing time, which prompts are producing weak outputs, which workflows are working well. Think of /dream as the janitor that keeps the memory file clean, and /insights as the analyst that tells you what to do differently.

4. Agent view and /goal: your GTM control tower

Requires: Claude Code v2.1.139 or later on a paid plan — Pro, Max, Team, or Enterprise. Run claude update first, then claude --version to confirm your version.

** Note: /goal currently works in the terminal version of Claude Code. If you're using the desktop app and don't see it, switch to the CLI.

Agent view and /goal allow you to run agents simultaneously to complete tasks.

/goal lets you define what “done” looks like, not just the task, but the specific completion condition. Claude then works autonomously across as many turns as it takes until that condition is met.

Agent view is the screen that shows you everything running across your work simultaneously, what is done, what is still running, and what needs a decision from you.

These commands allow you to be way more productive. For example you can kick off:

A launch email sequence being drafted against the campaign brief.

A competitive teardown of every announcement competitors made over the weekend.

A Q3 OKR draft pulled from last quarter’s results and current pipeline data.

A customer interview summary condensing the last month of sales calls into themes and objections.

A board update first draft pulled from CRM data, analytics, and last quarter’s deck.

You check Agent view between meetings. You approve, redirect, and ship. Pretty incredible.

How to set it up:

Step 1 — Update Claude Code.

claude update

Then confirm your version:

claude --version

You need v2.1.139 or later. If your version is lower, the commands below will not work.

Step 2 — Start your first session with /goal.

Open Claude Code and type /goal followed immediately by your completion condition. The goal itself is the instruction, Claude starts working the moment you hit Enter. You do not need to add a separate prompt after it.

Here is what a GTM /goal looks like:

/goal Draft a 5-email launch sequence for our Q3 campaign. 
Done means: all 5 emails written, subject lines included, 
personalisation tokens marked, reviewed against the ICP brief, 
saved to /drafts/q3-emails.md. Stop after 20 turns if not complete.

Notice the last line, always include a turn limit. This stops the session running indefinitely if something goes wrong.

Once running, you will see a /goal active indicator on screen with elapsed time updating in real time. To check status without interrupting it, type:

/goal

Step 3 — Send it to the background.

Once the session is running, send it to the background so you can start the next one:

/bg

Step 4 — Repeat for each workstream.

Start a new Claude Code session for each parallel task. Set a /goal for each one. Send each to the background with /bg.

Step 5 — Open Agent view.

Once your sessions are running, open the dashboard:

claude agents

You will see every session listed with its current status, running, waiting on you, or done. You can also open Agent view from inside any active session by pressing the left arrow key.

To check in on a session without fully entering it, press Space on that row, a peek panel opens showing the latest output or any question it is waiting on. Answer inline and it picks back up. To go deeper into a session, press Enter to attach to it fully.

Three practical things to know before you run this:

Running five parallel sessions uses five times the tokens of a single session. A Max plan handles this comfortably, Pro plan will hit limits faster with several complex sessions running at once.

Keep goals specific and bounded. “Draft a 5-email sequence for [campaign]” works. “Help with our Q3 marketing” does not give Claude a clear enough condition to know when it is done.

Agent view is currently a Research Preview, available on paid plans only and still being refined. Expect the interface to improve over the coming months.

---

Claude Code’s latest features really does turn it into an always on GTM team you can setup and start working across all parts of your business. Stay tuned for more in-depth tutorials on how these could blend into an overall marketing/growth/sales system for your business.

Until Next Time,

Happy AI’fying

Kieran