How We Use Claude Code to Run GTM Work

Picture a founder who sells a B2B service and spends 2 hours a day on sales admin. Before every call they look up the company and the person. Every morning they work through LinkedIn. After every call they write a recap. Every week they build a list in a spreadsheet. None of it is hard. All of it is repeated, and all of it is the work that gets skipped in a busy week.

That is the work we moved into Claude Code. Ashwanth runs much of Oriel's go-to-market work from it: research, lists, outreach routines, call prep and recaps.

This article explains what Claude Code is, which GTM jobs we run from it, the guardrails that keep it safe, what it is bad at and how a team could start. There is no code here. The point is the operating model, not the syntax.

What Claude Code Is

Claude Code is an agent from Anthropic that runs in a terminal, the text window developers use to run commands on a computer. You give it instructions in plain English. It can read and write files, run scripts and use tools you connect to it, and it shows you what it plans to do before it does anything that matters.

It was built for writing software. The same abilities make it useful for GTM work, which is mostly repeated research, data handling and writing. GTM means go-to-market: everything a company does to find, reach and win customers.

The useful way to think about it is as a workbench. Each repeated job, such as a call prep brief, becomes a written routine. Claude Code runs the routine, and a person checks the result.

The GTM Jobs We Run From It

Here are the jobs, from the first touch to after the call.

Fig. 1The GTM jobs we run from Claude Code

1. Research

   Profile and recent posts read before any message is drafted.

   LinkedIn profileRecent postsCompany news
2. List building

   Accounts and contacts pulled to the ICP, deduplicated against past outreach.

   ICP filtersDedupeSheet or CRM
3. Daily outreach routine

   The Sales Navigator run: requests, accepts, researched first messages, replies flagged.

   Sales NavigatorFirst messagesReply flags
4. Call prep

   A brief before every sales call: company, person, likely questions.

   Prep briefQuestions
5. Post-call recap

   A recap drafted from the call notes, with next steps.

   RecapNext stepsCRM

1. Profile and post research before drafting

Before any message to a prospect, the person's profile and recent posts are read so the message opens on something real. Claude Code gathers the profile, the recent posts and any company news, then summarizes what is worth opening on. If nothing useful turns up, it says so instead of inventing a hook.

2. List building

Lists are pulled to the ICP, the written description of who we sell to, and checked against everyone already contacted so nobody is approached twice. The output goes into a sheet or a CRM where a person can review it. We explain what makes a list worth sending to in How to Build a B2B Outbound List.

3. The daily Sales Navigator run

Every morning, the LinkedIn Sales Navigator routine runs: connection requests to qualified owners, a check on who accepted, a researched first message to each new connection and replies flagged for a person. The full routine is in The Connect-to-DM Routine We Run Every Morning.

4. Call prep briefs

Before every sales call, Claude Code writes a short brief: what the company does, what changed recently, who the person is, what they have said in public, and the questions worth asking. The same idea, built as a system, is in AI Call Prep, and the meeting prep and follow-up play shows the workflow end to end.

5. Post-call recaps

After a call, it drafts a recap from the notes or transcript: what the buyer said, what was agreed and the next steps with dates. A person edits it before it goes anywhere.

What a Written Routine Contains

The written routine is where most of the value sits. Claude Code follows instructions well, so the quality of the output depends on how clearly the job is described. Each of our routines covers the same 5 things:

• The goal. What a good result looks like, in a sentence. For a call prep brief: "One page that lets the person taking the call open on something specific."
• The inputs. Where the information comes from, in order of trust: the prospect's own posts and site first, news second, databases last.
• The steps. What to gather, in what order, and what to do when a source is missing.
• The output format. Headings, length and where the file or record goes.
• The stop rules. What the routine must never do: send anything, spend on a paid tool without approval, or state a fact without a source.

Writing these down takes more thought than running them. It also makes the work transferable: anyone on the team can read a routine and know exactly what the agent will do.

A Morning in Practice

Here is how a typical morning of GTM work can run this way, as an example.

• First thing. The Sales Navigator routine checks yesterday's accepts, sends a researched first message to each and queues the day's connection requests. Any reply is flagged for a person.
• Before the first call. A prep brief for each of the day's calls is already written. The person reads it in 5 minutes instead of researching for 30.
• After each call. The notes go in, a recap draft comes out with next steps and dates. The person edits it and sends it themselves.
• Mid-week. A list for the next outreach round is built to the ICP and deduplicated. A person approves it before any drafting starts.

Every reply goes to a person, and in client work nothing sends until the client approves it. The agent removes the preparation, so the person's time goes to reading, deciding and talking to buyers.

The Guardrails We Keep

An agent that can run scripts and use connected tools can also do damage quickly. These rules are why we trust it with daily work.

Client outreach waits for approval

This is the rule that matters most in client work. The agent drafts, and the client approves every lead list and every first message before anything sends, on email, LinkedIn or any other channel. In Ashwanth's own Sales Navigator routine, the first message to a new connection goes out after the research step, and every reply goes to a person. We cover the reasoning in Human Approval in AI Outreach.

Keys kept out of files

API keys, the passwords that let software use a paid service, are kept in one protected place and never written into routines, notes or shared files. A routine that gets copied should never carry a key with it.

Cost limits on paid APIs

Some tools charge per lookup or per call. An agent running in a loop can run up a bill before anyone looks. Any paid call needs a person's yes and a cost estimate first.

A person reads every brief and recap

Even research briefs that never reach a buyer get read before they are used. If a brief contains a wrong fact, the message built on it will too.

What It Is Bad At

Claude Code is very good at repeated, well-defined work. It is weak at a few things that matter in sales.

Fig. 2Where Claude Code helps and where a person decides

Claude Code doesA person decides

ResearchReads sources, writes the briefWhether the brief is accurate

ListsPulls and cleans contactsWhether the list is right

MessagesDrafts from the researchWhether it sends

RepliesFlags and sorts themWhat to answer

CallsPrep brief and recapThe conversation itself

SpendRuns within set limitsWhether to raise them

• Knowing what is stale. It will report a job title or a company fact with the same confidence whether the source is from last week or 2 years ago, unless the routine checks dates.
• Reading a buyer. It does not know that this prospect was annoyed by a pushy vendor last month, or that a warm referral needs a different tone. A person does.
• Pushing back. It carries out instructions efficiently, including poor ones. If you ask for 500 messages to a loose list, it will help you write them.
• Judging what is worth saying. It can find 10 facts about a company. Deciding which one is a good reason to reach out is still a judgment call.
• Running unattended for long. Websites change, tools change, inputs drift. Routines need someone watching the output.

None of these are reasons to avoid it. They are reasons to keep a person on the decisions that matter: targeting, approvals and every reply.

Most of these weaknesses can be reduced with better routines. Checking dates on every source catches stale facts. Feeding in notes from past conversations gives the agent context on a buyer. Writing stop rules into every routine limits what a poor instruction can do. None of them reach zero, which is why the review step stays.

How a Team Could Start

If you want to try this, start small and stay away from anything that touches buyers until the basics are reliable.

1. Pick one internal task. Call prep briefs are a good first choice: high value, easy to check, no buyer contact.
2. Write the steps down. What does a good brief contain? Where does each part come from? This written routine is the real asset.
3. Run it with a check on every output for 2 weeks. Note every error and fix it in the written routine so it does not come back.
4. Set the guardrails before connecting anything that sends. Approval, keys, cost limits.
5. Add the next task. Research before drafting, then list building, then outreach drafts for approval.

If this sounds like a role rather than a side project, it is. The discipline of building these systems is often called GTM engineering, which we explain in What Is GTM Engineering?.

When This Is the Wrong Approach

• Nobody on the team will review outputs. Without a person checking, the guardrails do not exist.
• You have no written process to automate. If the way you sell is not defined, an agent will automate a guess.
• The volume is tiny. If you take 2 calls a month, writing your own prep notes is faster than building a routine.

FAQs

Run Your GTM Work This Way With Oriel

Oriel runs its own go-to-market from routines like these and builds similar systems for B2B companies above $100k/month in revenue. We write down your research, list building, outreach, call prep and recap steps, set the approval and cost guardrails, and run them inside tools you own, with you approving every list and first message before it sends.

On a 30-minute call, walk us through the sales admin that fills your week. We will show you which parts we would move into routines first and where a person should stay in charge. If you are not ready for this yet, we will tell you that too.