How I use Claude Code in client GTM work

Most people who’ve heard of Claude Code think of it as a coding assistant. That’s the smallest part of what it does in my client engagements.

I’ve embedded it into three parts of every GTM engagement I run: outbound intelligence, content operations, and CRM data. The result is a delivery model that didn’t exist two years ago — and one I can’t imagine going back from.

This isn’t about replacing the thinking. It’s about eliminating the overhead between thinking and shipping.

---

The three jobs it does

I use Claude Code for three distinct workloads, each with its own setup and logic.

Job 1: Research and outbound intelligence

I’ve documented the GTM agent stack in detail. Short version: Claude Code, four MCP tools, a CLAUDE.md operator brief, and five named skills handle the enrichment-to-draft pipeline. A 50-account list goes from cold CSV to scored accounts with drafted sequences in under an hour.

The piece I didn’t cover there: the compounding effect.

After three months on an engagement, the CLAUDE.md is dense with client-specific ICP signals, objection patterns, and sequence learnings. The fifth run is measurably better than the first. That doesn’t happen in Clay or Apollo. Those systems process rows — they don’t accumulate operator context.

Job 2: Content operations

Every GTM engagement touches content: the website, case studies, LinkedIn positioning, email nurture tracks. The gap between “we need to update the homepage” and “homepage is updated” is usually weeks of stakeholder cycles and writing back-and-forth.

With Claude Code, the workflow collapses:

1. Audit phase: /audit reads the current site, the competitor set, and the ICP brief. Returns a structured gap analysis with specific recommendations per section.
2. Draft phase: /draft [section] — not from a template, from the audit findings. Each draft is grounded in what the audit surfaced, not what the writer assumed.
3. Iteration: I edit the draft. It observes what I changed and why. The next draft starts closer to the right answer.

The output isn’t perfect out of the gate. The baseline is research-grounded, and the iteration cycle is hours instead of weeks.

Job 3: CRM data operations

This is the most underrated use case.

Most B2B CRMs I walk into have the same problems: duplicate companies, contacts without companies, deals stuck at the wrong stage, custom fields nobody filled in. Cleaning this manually takes a researcher weeks and still comes out inconsistent, because humans apply the same rules differently on row 12 and row 412.

Claude Code with the HubSpot MCP does it differently:

• Reads contact and company records in batches
• Applies ICP scoring criteria to classify and segment — consistently, at scale
• Flags duplicates and missing data by pattern, not by checking each row individually
• Generates the update payloads for human review before anything touches the CRM

It’s not one-click. You still review and approve. But the classification logic is consistent, and you can inspect the reasoning — “scored low on ICP because headcount under 50 and no enterprise tech signals” — before committing anything.

---

The engagement model this enables

Before: A GTM engagement meant scoping deliverables, assigning research tasks, writing briefs for writers, reviewing drafts, building sequences, managing CRM cleanup in parallel. Six weeks minimum, with constant orchestration.

Now: I scope around operator outputs — the enriched account list, the sequenced outreach, the positioned website copy, the clean CRM — and I deliver them faster with fewer dependencies on external resources.

This doesn’t mean I work alone. Clients are still in the loop on ICP definition, sequence tone, content approval. The judgment is mine and theirs. What’s gone is the research overhead and the writing latency.

---

What clients see — and don’t

Clients see faster delivery and higher-quality first drafts. Sequences that reference actual account-specific signals rather than generic personalization. A CRM that reflects the ICP rather than historical chaos.

They don’t see the agent stack. That’s intentional.

The output speaks for itself. The tool is an implementation detail. A client doesn’t need to understand MCP servers. They need sequences that book calls. That’s the contract.

---

The constraints worth knowing

It’s not frictionless. A few things to be clear about:

It requires real upfront investment. The CLAUDE.md takes time to write properly. The client memory structure takes a session or two to scaffold. First-run output is good but not great. Compounding starts at run two, not run one.

It’s better on information-dense tasks. The richer the ICP definition, the more detailed the site content, the cleaner the CRM export — the better the output. The agent is only as good as the context you give it.

It’s not a replacement for operator judgment. The scoring logic, the sequence strategy, the positioning decisions — those come from the operator. Claude Code executes. The operator decides.

I scope it, I run it, I review the output. The speed improvement is real. The judgment layer is still mine.

---

If you’re running GTM engagements and the bottleneck is research, writing, or CRM hygiene rather than strategy, this is worth looking at seriously.

Book a call if you want to walk through how it maps to your stack.