How Anthropic's GTM Engineering Team Uses Claude

1. The administrative burden that consumes sales teams

Modern account executives face an uncomfortable reality: the job title says “sales” but the daily reality is closer to “email administrator.” A typical enterprise AE managing a large territory can have hundreds of accounts, each with its own history, technical requirements, open tickets, pricing discussions, and relationship context. Jared Sires, a GTM Product Manager at Anthropic, describes managing 600 to 700 accounts as a startup account executive — a portfolio so large that keeping it all organized is itself a full-time job, on top of the actual selling.

The evening hours become a kind of administrative overflow valve. When the calendar clears and the calls stop, the real work begins: working through an inbox that accumulated during the day, each email requiring the rep to recall or reconstruct context about that particular account. Is this customer on an enterprise contract or a startup plan? Did they ask about a specific feature last quarter? What is the current status of their technical evaluation? Getting those answers means digging through Slack conversation history, hunting through Google Docs that may or may not be named consistently, checking Google Drive folders that may have been organized by whoever owned the account before. Jared describes working until 8:00 or 9:00 at night just answering emails — time that should have been spent with family, on rest, or on forward-looking sales strategy.

This is not a time-management problem or a productivity problem in the conventional sense. It is a structural information-access problem. The knowledge an AE needs to write a good customer email exists in the organization — it is in product documentation, in past email threads, in internal wikis, in Slack channels. But it is distributed across a dozen systems with no unified interface. Every email response requires a mini-research project before any writing can begin.

2. What “Class” is and how it works

Jared’s response to this problem was to build something. “Class” — short for Claude Draft — is an internal tool he built that hooks the Claude API up to the context sources an AE needs, then uses that context to generate email drafts directly from the inbox. The name captures the function precisely: it is a system that generates drafts using Claude, pulling in the relevant knowledge before it starts writing.

The architecture of Class reflects how well-structured AI systems are generally built. There are three core components working together. First, a system prompt that establishes who the user is, what their role is, and what they are trying to accomplish — this is the “persona” layer that gives Claude a framework for writing in an appropriate voice and with relevant priorities. Second, a context retrieval layer that pulls in live information from Google Docs and web URLs at the time a draft is being generated — this ensures that the draft reflects current product capabilities, current pricing, and current policy rather than whatever the model learned during training. Third, pattern matching that helps the system understand what kind of email it is looking at and what kind of response is appropriate.

What makes this architecture powerful is that none of these components requires a software engineering background to configure. Jared built Class himself before transitioning into a technical role. The system prompt is written in plain English. The Google Docs integration points at documents that already exist. The result is a tool that any AE could set up and use, even without knowing what an API is. When Jared demos Class, the workflow is: enter your name and role, select “account executive” from a dropdown, describe what you help customers with, generate a system prompt, connect your Google Docs and any relevant URLs, then open an email and click “generate draft.” Claude retrieves the documentation, reads the email, and produces a response that sounds like a knowledgeable human who has actually read the relevant docs.

3. The outcome: time savings and a career transformation

The numbers Jared reports are striking but believable once you understand how the tool works. Class saves him at least two to three hours per day on email — and that is likely a conservative estimate, since he describes thousands of emails having been sent through the system over its lifetime. At two to three hours per day, that is roughly 500 to 750 hours recovered annually for a single user. Across a sales team of ten or twenty AEs, the aggregate productivity gain becomes a significant organizational advantage.

But the more interesting outcome is not the time savings — it is what Jared did with the time and capability that building Class gave him. The experience of designing and shipping a real tool that his colleagues used transformed his sense of what was possible for him professionally. He describes having lacked the technical confidence to participate in architecture conversations before, and emerging from the Class project with the ability to “design more products and have senior engineers help me implement this to the final stretch.” He ultimately moved into a GTM architect role — a position that would not have been accessible to him through the traditional path of accumulating engineering credentials.

This career trajectory is worth dwelling on. The conventional assumption is that building software requires learning to code, and that learning to code requires significant time investment in studying programming fundamentals. What Jared’s experience suggests is that AI has effectively reordered that dependency. The ability to describe a system clearly in natural language, to understand what problems it needs to solve, to iterate on a design based on user feedback — these skills transfer directly into AI-assisted software development. Technical implementation details that would previously have required months of study can now be handled collaboratively with AI. The gap between “person with a good idea” and “person who shipped a working tool” has narrowed dramatically.

pip install anthropic
export ANTHROPIC_API_KEY="your-key-here"

import anthropic

client = anthropic.Anthropic()

def build_system_prompt(name: str, role: str, focus: str) -> str:
    return f"""You are {name}, working as {role}.

Your communication focus: {focus}

When drafting email replies:
- Match the tone of the incoming email (formal/informal)
- Be concise — get to the answer in the first sentence
- Always reference specific product capabilities when relevant
- End with a clear next step or call to action
- Do not make up features or pricing — only use the context provided

If the context documents do not contain enough information to answer a question,
say so explicitly rather than guessing.
"""

def load_context(context_text: str) -> str:
    """
    In a real system, this would fetch from Google Docs or web URLs.
    For this example, we pass context directly.
    """
    return context_text

def generate_email_draft(
    incoming_email: str,
    system_prompt: str,
    context: str
) -> str:
    """
    Generate an email draft using Claude with provided context.
    """
    message = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=1024,
        system=system_prompt,
        messages=[
            {
                "role": "user",
                "content": f"""Here is relevant context from our documentation:

---
{context}
---

Please draft a reply to this incoming email:

{incoming_email}

Write only the email reply text, ready to send."""
            }
        ]
    )
    return message.content[0].text

MY_SYSTEM_PROMPT = build_system_prompt(
    name="Alex",
    role="Account Executive at Acme SaaS",
    focus="helping enterprise customers get value from our analytics platform"
)

PRODUCT_CONTEXT = """
Acme Analytics Platform — Key Facts:
- Pricing: Starter ($99/mo), Growth ($499/mo), Enterprise (custom)
- Enterprise plan includes: SSO, audit logs, custom data retention, dedicated support
- Data residency options: US, EU, APAC
- API rate limits: 1,000 req/min on Growth, unlimited on Enterprise
- Typical onboarding: 2-4 weeks with dedicated success manager
- SOC 2 Type II certified, GDPR compliant
"""

SAMPLE_EMAIL = """
From: procurement@bigcorp.com
Subject: Re: Acme Analytics evaluation

Hi Alex,

Thanks for the demo last week. Our security team has a few questions before we can move forward:

1. Do you support EU data residency?
2. What is your SOC 2 status?
3. Can you accommodate a custom data retention policy for our 7-year compliance requirement?

We are also wondering about API limits at the Enterprise tier.

Best,
Sarah
Procurement Lead, BigCorp
"""

draft = generate_email_draft(
    incoming_email=SAMPLE_EMAIL,
    system_prompt=MY_SYSTEM_PROMPT,
    context=PRODUCT_CONTEXT
)

print("GENERATED DRAFT:")
print("=" * 60)
print(draft)

python email_assistant.py

Where to go next

• Watch the original video by Jared Sires to see the Class system demonstrated live.
• Explore How Anthropic’s Marketing Team Uses Claude for more examples of non-engineers building AI-powered productivity tools.
• Read the Anthropic API documentation on context windows to understand how to pass large documents effectively.