November 11, 2026

Claude Code for GTM: Setup, Workflows and Tools (2026)

How to use Claude Code for GTM: what it automates, how to set it up with CLAUDE.md, connect HubSpot and your tools, and where it breaks. A practical guide

Claude Code for GTM: A Practical Guide to Setup, Workflows and Tools

By Akil, Head of RevOps at Leadle · Updated November 2026

‍

Claude Code for GTM means using Claude Code, Anthropic's agentic tool, to run go-to-market work from your own files and tools: researching accounts, sorting them against your ICP, enriching contacts, writing outreach and updating the CRM. You describe the rules once in plain English, and it does the repeatable work.

‍

Key takeaways
‍

  • The structured, repeatable layer of GTM operations can now run on Claude, with people reviewing the output.
  • Start with a written brain of your business: ICP, positioning, signals and rules. Every workflow inherits it.
  • Connect your tools through MCP, one at a time. Our stack is Claude Code, Blitz API, HubSpot, Instantly and HeyReach.
  • We've built 30-plus agents across five areas: pipeline, sales execution, RevOps, client operations and founder support.
  • A person reviews anything that reaches a customer. Ours is a campaign manager.
  • Start with one expensive workflow, prove it, then add the next.

‍

New to Claude Code? Watch our beginner session first, or read the beginner guide [LINK: beginner guide].

Get Started with Claude Code: Build a Working AI Agent for Sales & GTM. A zero-code session led by Akil, Head of RevOps at Leadle: how Claude Code differs from chat, how to set up a project folder, and a live build of an account-segmentation agent.

‍

In this guide: Why GTM changes · The brain of your business · Connect your tools · Workflows end to end · Agents we've built · What it can do and where it breaks · How to start · What it takes · What stays human · Claude Code vs other tools · Cost · Where it goes wrong · Who builds it · Measuring ·
‍‍

A year ago, scaling B2B GTM meant hiring specialists.

You needed a HubSpot expert to keep the CRM clean. A Salesforce admin to build the reports. A RevOps manager to maintain the pipeline. A sales operations hire to chase the data hygiene issues. A marketing operations person to manage the campaign attribution.

Each role was a $60K-$120K annual investment. Each role had a months-long ramp. Each role created a dependency you couldn't easily unwind.

That model is changing.

The same operational work—CRM hygiene, pipeline forecasting, signal research, outbound QA, proposal generation, client health scoring—can now be handled by Claude agents. Not all of it. The work that requires judgment, relationships, and strategy still belongs to humans. But the structured, repeatable layer that consumed 60% of those specialist roles? That's automatable now.

This isn't a "Claude can do your job" story. It's an architecture shift. The specialists you used to hire to do the work are now the specialists you hire to design the agents that do the work. Fewer people. Higher leverage. Better systems.

This blog covers how to use Claude Code for GTM, the 30 agents we've built across five operational categories, and how to start.

‍

What Claude Code is: Claude chat answers a question and the conversation ends. Claude Code is the same model working inside your project folder: it reads your files, follows your rules, uses your tools and keeps going across a task. For the full beginner explanation, see [LINK: beginner guide].


‍

Meme of four brains, from "Do it yourself" to "Use Claude to automate that and everything", with the Claude logo

‍

Why does Claude Code change how GTM teams run operations?


Because the structured, repeatable part of GTM work, the part that used to need a specialist, can now run on Claude Code with a person reviewing the output. Strategy, relationships and judgment stay human. What changes is who does the operational work and how many people you need for it.


Scaling B2B GTM used to mean hiring specialists. A HubSpot expert to keep the CRM clean. A RevOps manager to maintain the pipeline. A sales ops hire to chase data hygiene. A marketing ops person for attribution. Each was a hire, a ramp-up and a dependency.


The work those roles did, CRM hygiene, pipeline checks, signal research, outreach drafts, proposals, client health scoring, is largely structured and repeatable. That layer can now run on Claude, with a person reviewing it. Not all of it. Work that needs judgment, relationships and strategy still belongs to people.


It's an architecture shift, not a "Claude can do your job" story. The specialists you used to hire to do the work are now the specialists you hire to design the agents that do it. Fewer people, more leverage, better systems.


What Claude Code is, in one paragraph.
Claude chat answers a question and the conversation ends. Claude Code is the same model working inside your project folder: it reads your files, follows your rules, uses your tools and keeps going across a task. For the full beginner explanation, see [LINK: beginner guide].

‍

How do you give Claude a brain of your business?


You write down what Claude needs to know about your business as a set of short files it reads before every session: your ICP, positioning, competitors and signals, plus the rules for the work. Every agent you build on top then starts from your business, not from a blank page.


Without that, you're prompting Claude from scratch every morning and getting generic output. With it, each workflow starts aligned.


A working GTM brain typically includes:

  • Company profile. What you sell, who you sell to, current priorities.
  • ICP definition. Specific ranges, technographic signals and anti-ICP exclusions. Not "Series B SaaS".
  • Positioning. Core value pillars, how you differ, and the language you don't use.
  • Competitive notes. Win and loss patterns, common objections, where you beat them and where you lose.
  • Signal library. The buying signals worth acting on, how to score them, hook angles and a log of what worked.

‍

How the pieces fit. The main file is CLAUDE.md, the rules Claude reads at the start of every session. Memory is separate: a notebook Claude builds as you work, capturing preferences and mistakes you corrected. You write the rules once, and memory grows on its own. Keep the files short, add rules as you spot gaps, and never put keys or tokens in them. The details, with a worked example, are in the beginner guide [LINK: beginner guide].

‍

We define the ICP with Claude. Give it the evidence you have: your best customers, deals you lost and what you know about who buys. Ask it to propose the rules, test them on a real list and correct what's wrong. Then write the result into the brain. This is the first thing to build, before any agent, because every agent depends on it.

‍

How do you connect Claude Code to your CRM and GTM tools?


You connect each tool through MCP, an open standard that lets Claude read and write live data. Connect the CRM first, then the data tool, then the sending tools, one at a time. Start with read access and widen it once the output is clean.

‍

MCP is the Model Context Protocol. We use it because the alternative is writing a specific API integration for every task, which is a huge job to build and to keep running. MCP is the better solution: you connect each tool once, and Claude can then use it for any task. Without a connection, Claude gives advice about data it can't see: you paste a spreadsheet in, it analyses it, you implement the result by hand. With one, a single instruction can reach several tools.

‍

What we connect, in order

  1. HubSpot, our CRM. It's the system of record, so it's the first connection. See our guide to connecting Claude Code to HubSpot.
  2. Blitz API, our data layer. Blitz publishes its own MCP server. In the desktop app, add it with Blitz's one-click "Add to Claude" connector.You sign in with your Blitz account, so there's no API key to paste.
  3. Instantly and HeyReach, our sending tools. Both are available through the built-in connectors in Claude itself. We connect them last, because they act on the outside world. Email goes out on Instantly/Smartlead and LinkedIn on HeyReach/Aimfox.

‍

User scope. By default, a server you add in Claude Code is only available in the current project. Add --scope user and it's available in all your projects. That's what you want for a tool the whole team uses.

‍

What does a Claude Code GTM workflow look like end to end?

‍

Workflow diagram: Blitz API data, Claude Code drafts, campaign manager approves, Instantly and HeyReach send, HubSpot records


Workflow 1: Outbound, from a list to a sent message
‍

  1. Clean the list. Claude validates the input, fixes obvious errors and removes duplicates.
  2. Segment the accounts. Claude applies the ICP rules and sorts each account into sequence, review or exclude, with a reason for each. In our webinar demo, 75 accounts came out as 39 to sequence, 8 to review and 28 excluded.
  3. Find and enrich the people. For accounts that qualify, Claude finds the target personas and enriches their contact details and basic signals through Blitz API.
  4. Write the messages. Claude writes each one from the ICP, the signal and the contact. It doesn't paste a company name into a template.
  5. Review. Our campaign manager reviews the messages before anything sends.
  6. Send. Approved messages go out through the Instantly and HeyReach connectors.


The human check:
the review bucket for borderline accounts, and the campaign manager's review before sending.


Workflow 2: Post-call CRM updates
‍

  1. Input. A call transcript goes into the project folder.
  2. Extract. Claude pulls the structured fields: next steps, objections, open questions and anything agreed.
  3. Propose. It drafts the call note and the CRM field updates.
  4. Review. The rep approves the changes before they write to HubSpot.


Post-call admin is one of the most expensive recurring tasks in sales, and it's one of the best places to start because it needs no outbound risk.


The human check:
the rep approves before anything writes to the CRM.
‍

Workflow 3: CRM hygiene and pipeline checks
‍

  1. Scan. Claude reads the CRM and flags duplicates, missing required fields and deals whose stage doesn't match their activity.
  2. Report. It produces a short list ranked by impact.
  3. Fix in batches. Someone approves the changes batch by batch.


Start this one read-only. Let it report for a few weeks before you let it change anything.


The human check:
read-only first, then batch approval.

‍

GTM agents we have built with Claude:


We've built 30-plus agents across five areas, each built around a real GTM problem. They cover pipeline intelligence, sales execution, RevOps automation, client operations and founder support. Each one handles a structured job that used to need a specialist, with a person reviewing the output.
‍

  • Pipeline intelligence. Knowing which accounts to talk to, when and why: signal detection across firmographic, technographic, intent and behavioural data, ICP enrichment, account scoring and buying-window detection. Outbound becomes targeted instead of speculative.
    ‍
  • Sales execution. Running outbound, preparing for calls and handling proposals: outreach drafting with a review step, meeting prep briefings, discovery call preparation, and proposal and SOW drafts.
    ‍
  • RevOps automation. Keeping the CRM trustworthy and the pipeline visible: call intelligence that turns transcripts into structured CRM updates, CRM hygiene monitoring, pipeline stage validation and lifecycle stage automation. This is where the shift is most direct.
    ‍
  • Client operations. Keeping customers healthy, onboarded and renewed: client health scoring, churn-risk detection, onboarding pack generation, QBR preparation and renewal forecasting.
    ‍
  • Founder support. The work only the founder can do but shouldn't do alone: daily operational briefings, SOP and playbook generators, context pulls and preparation for high-stakes conversations.


Browse the full map of 30 agents →

‍

What Building This Actually Takes


Setup work is something that most teams underestimate.

Claude Code for GTM isn't a plug-and-play deployment. The reason it produces durable operational leverage is the same reason it requires real setup work to get right.
‍

Setting up the context layer takes 1-2 weeks. Writing the company profile, defining the ICP with specificity, documenting positioning, and building the initial signal library is the foundation. Done casually, the agents that depend on it produce generic outputs. Done well, every agent that runs on top of it produces aligned outputs from day one.


Each agent takes 1-3 weeks to build and stabilize.
The first working version is usually quick. The next 30 days are spent watching it run, identifying edge cases, and tightening the logic. Skipping this stabilization phase is the most common reason agents fail in production.


Integration with your stack adds time.
Connecting Claude to HubSpot, Salesforce, Slack, Lemlist, or Clay through native integrations or MCP servers requires engineering—either yours or ours. The agents only produce leverage if they're reading from and writing to the systems your team already uses.


Maintenance is ongoing.
Context files need updates as your ICP evolves, positioning sharpens, and campaigns generate new signal data. Agents need refinement as edge cases surface. This isn't set-it-and-forget-it work. It's living infrastructure.


The teams getting durable value treat this as operational architecture, not an experiment. The teams that try to shortcut the setup usually rebuild within six months.


The good news: this work compounds. Every agent you ship makes the next one faster to build. Every context refinement sharpens every output downstream. The first three months are an investment. The next three years are return on it.

‍

How do you measure whether it's working?


Compare against the manual baseline: the hours each task took before, the error rate and the quality of the output. Judge each workflow separately, and decide whether to scale it, fix it or stop it.

  • Time per task, before and after.
  • Error rate, such as accounts in the wrong bucket or wrong facts in a message.
  • Output quality, judged by the person who reviews it.
  • Downstream results, such as replies and meetings, once messages are going out.

‍

What can Claude Code do for GTM teams?


It handles the structured, repeatable work: researching accounts, sorting them against your ICP, enriching contacts, drafting outreach, summarising calls and checking CRM records. People keep strategy, relationships and the final call on anything that reaches a customer.

‍

GTM jobs, and where each one breaks

‍

GTM jobWhat Claude Code doesWhere it breaksHuman check
Account researchReads a list or the web and summarises each accountStale or wrong factsSpot-check a sample
Account segmentationApplies your ICP rules and sorts accounts into sequence, review and excludeVague rules give vague bucketsReview the borderline bucket
Contact enrichmentFinds people and verified contact data through a data toolCoverage varies by region and roleCheck a sample before sending
Outreach draftingWrites messages from the ICP, the signal and the contactReads as generic if the inputs are weakA campaign manager reviews every message
Call notes to CRMPulls next steps, objections and fields from a transcriptMisreads names or figures on poor audioReview before it writes to the CRM
CRM hygieneFlags duplicates, missing fields and stage mismatchesFixes the wrong thing if rules are unclearApprove changes in batches
Competitor and market researchGathers and summarises public informationTreats opinion as factCheck the sources
ReportingPulls figures together into a summaryGarbage in, garbage outCompare against the source

‍

How do you start building GTM agents with Claude?


Build the brain first, then pick the one workflow that costs the most hours, build the simplest working version and put a person in front of it. Measure against the manual baseline, then scale, fix or stop. Don't start with the most ambitious agent.
‍

  1. Set the foundation. Build the brain of your business first: ICP, positioning, signals, rules. Without it, every agent starts from scratch.
  2. Audit recurring friction. List what eats the most team hours weekly: post-call admin, research, proposals, CRM hygiene.
  3. Pick the highest-cost workflow. Not the most strategic one. The most operationally expensive one. The agent that recovers the most hours proves the model.
  4. Build the simplest working version. Don't over-engineer. Start with one agent, using the brain.
  5. Deploy with a person in front of it. A human reviews every output for the first 30 days.
  6. Measure the lift. Time saved, quality, error rate, against the manual baseline.
  7. Scale, fix or stop. Most agents survive and get extended. The ones that don't should be stopped before they become operational debt.


For the first few weeks, work in accept-edits mode on Sonnet, so you see each step.

‍

What does it take to build and run?


Real setup work, and more than most teams expect. Plan on one to two weeks for the brain, one to three weeks per agent to build and stabilise, engineering time to connect your tools, and ongoing upkeep. It isn't plug and play, and the teams that shortcut it usually rebuild within six months.

  • The brain: one to two weeks. Done casually, the agents on top produce generic output. Done well, every agent starts aligned.
  • Each agent: one to three weeks. The first version is quick. The next 30 days go on watching it run and tightening the edge cases. Skipping that is the most common reason agents fail in production.
  • Connecting the stack: connecting Claude to HubSpot, your data tool and your sending tools needs technical work, either yours or ours.
  • Maintenance: the brain needs updating as your ICP and positioning change, and agents need refining as edge cases appear. It's living infrastructure.


The work compounds. Every agent you ship makes the next one faster, and every improvement to the brain sharpens every output.

‍

Claude Code vs Clay, Zapier and other GTM tools: when to use which?

They do different jobs. Data tools supply contacts and signals, workflow tools run fixed triggers, sending tools deliver messages, and Claude Code does the reasoning and the writing. Most teams use Claude Code alongside the others, not instead of them.


Tool types and what each is for
‍

Tool typeExamplesBest forFalls short
Claude CodeClaude CodeReasoning, rules, writing and multi-step work on your filesNo built-in data or sending
Data and enrichmentBlitz API, ClayFinding and verifying companies, people, emails and phonesNeeds something to decide what to do with the data
Workflow automationZapier, n8nFixed trigger-and-action jobsWeak at judgment and writing
Sales engagementInstantly, HeyReachSending and managing sequencesDoesn't decide who to contact or what to say

For a full comparison with 2026 prices, see our guide to the top Clay alternatives.

‍

How much does GTM with Claude cost, and how do you stay within usage limits?


The cost has four parts: the Claude plan, the other tools you connect, the time to build, and review time.

  • The Claude plan: usage limits depend on the plan.
  • A data tool: Blitz API's Start plan was $99 a month in November 2026, on a flat subscription with every endpoint included.
  • Sending tools: Instantly and HeyReach have their own plans.
  • Build and review time: the part teams underestimate.

To stay within usage limits: start small, build one piece at a time, start on Sonnet, plan the project before you open Claude Code, and compact long conversations. The beginner guide has the detail [LINK: beginner guide].

‍

Do you still need a GTM engineer, or can you build this yourself?

A first agent doesn't need an engineer. Connecting several tools, keeping them running as your stack changes and designing the whole workflow is where most teams want help.

  • Do it yourself if you have one clear workflow, the time to learn and a CSV to start from.
  • Hire if you have enough ongoing GTM work to keep a person busy.
  • Use an agency if you want the connected workflow built and maintained without hiring. That's the GTM engineering work we do.

‍

How do you measure whether it's working?


Compare against the manual baseline: the hours each task took before, the error rate and the quality of the output. Judge each workflow separately, and decide whether to scale it, fix it or stop it.

  • Time per task, before and after.
  • Error rate, such as accounts in the wrong bucket or wrong facts in a message.
  • Output quality, judged by the person who reviews it.
  • Downstream results, such as replies and meetings, once messages are going out.


The Bottom Line


The work that used to require specialist hires is becoming agent work, with human oversight. CRM admin. Pipeline operations. Signal research. Proposal generation. Client health monitoring.

The teams building this well aren't trying to replace their teams. They're freeing their teams to focus on what only humans can do - strategy, relationships, judgment, while agents handle the structured layer underneath.

We've built 30 agents across the five categories above. Each one reduces the operational load that previously required dedicated headcount. Setup takes real work. The leverage compounds for years.

Get a GTM workflow scoped for your team →

Or browse the full map of 30 GTM agents: AI Agents Map →

‍

FAQs:

‍

1. What is Claude Code for GTM? It's using Claude Code to run go-to-market work: researching accounts, sorting them against your ICP, enriching contacts, writing outreach and keeping the CRM clean, from rules and context you write in plain English.

2. How does Claude connect to HubSpot and our other tools? Through MCP. Each tool is connected once, and Claude can then read and write to it. We use MCP because the alternative is writing a specific API integration for every task.

3. Do I need to know how to code? Not to build a first agent. Connecting external tools is the step where some teams want technical help.

4. What GTM tasks can Claude automate? Account research, segmentation, enrichment through a data tool, outreach drafting, call notes into CRM fields and CRM hygiene checks. People keep strategy and relationships.

5. Is Claude Code better than Clay? They do different jobs. Clay is a data and workflow platform. Claude Code does reasoning and writing on your files. Many teams use both, or pair Claude Code with a data API. See our Clay alternatives guide.

6. Can Claude replace SDRs, AEs or RevOps managers? No. It removes the structured work those roles spend much of their time on. Relationships, strategy and judgment stay human.

‍

in this article: