Claude Code can now run an entire LinkedIn outbound campaign from a single terminal.
Othmane Khadri, founder and CEO of Earleads, walked through the exact system his team uses to scrape post engagers, qualify them through a 7 gate pipeline, draft outreach copy, and send connection requests. No jumping between Clay, lemlist, Instantly, or any other platform. One prompt, one terminal, one sequence of skills and APIs doing the work.
Earleads crossed $1M ARR in November and is on track to hit $10M. This is the engine running underneath that growth.
Here is the full breakdown of how it works and what it means for GTM teams in 2026.
1. Why Claude Code Is Becoming a GTM Operating System
GTM teams typically run campaigns across five to seven different platforms. Data sourcing happens in one tool. Enrichment happens in another. Sequencing in a third. Analytics in a fourth.
Each platform has its own learning curve, its own interface, its own way of storing data. Teams spend more time gluing tools together than running campaigns.
Claude Code changes this equation. It acts as an orchestrator that can call any API, run any skill, and hold the full context of your business in one place. You describe what you want in plain language and the system executes the workflow.
The shift is not about replacing Clay or any specific tool. It is about collapsing the entire stack into a single interface where each tool becomes an API call instead of a destination.
Tools struggle to keep up with every niche use case. A terminal that can call any API does not have that limitation.
2. The Four Layer Structure of a Claude Code GTM System
Othmane breaks the operating system into four layers. Each one handles a different part of how the system thinks and acts.
The input layer
The user types or dictates a query. This is the entry point. A strategy, a campaign idea, a specific ICP to target. The clearer the input, the better the execution downstream.
The Claude.md rules file
This file contains the general rules Claude Code follows across every task. Writing style, workflow preferences, what to always check before executing, what to never do. It is the system's constitution.
The skills layer
Skills are containers that bundle one specific action with the rules and API access needed to execute it. One skill scrapes LinkedIn post engagers. Another qualifies leads against your ICP. Another drafts outreach copy. Another sends connection requests through an API.
One skill equals one action. Skills standardize how Claude Code interacts with external systems and make sure nothing happens in a random or unsafe way.
The memory.md file
This is where your business context lives. Your ICP definitions, your value proposition, your best clients, outreach angles that worked in the past, ideas you had at midnight that you sent to Claude to save.
When Claude Code runs a campaign, it reads all four layers. The input tells it what to do. The rules tell it how. The skills tell it which actions are available. The memory tells it who you are and who you serve.
3. Running a Full LinkedIn Campaign From One Prompt
Here is what the demo does step by step.
Michel gives Claude Code one prompt:
"Build a GTM outreach campaign. First, scrape the leads from this LinkedIn post. Second, qualify them based on Earleads ICP. Third, test this outreach angle: 15 minute chat about how we are rebuilding Clay in an open source way."
Claude Code goes into plan mode first. This is important. Plan mode lets it read every file, every skill, every rule, and come back with a full plan before executing. No surprises.
The plan it produces:
→ Scrape post engagers using a LinkedIn skill
→ Run them through the qualification flow
→ Score and tag each lead from 0 to 100
→ Draft the outreach message with one variant
→ Send a test message to one profile
→ Log everything to a unified Notion database
→ Respect LinkedIn rate limits so the account does not get flagged
Michel approves the plan. Claude Code executes.
Five to ten minutes later, the campaign has run end to end. Leads scraped, qualified, and scored. The outreach message sent to Michel directly from the terminal, without touching Clay, lemlist, or Instantly. Every action routed through code calling APIs via skills.
Based on how this system scrapes LinkedIn post engagers and maps them back to decision makers, we built a similar tool for the public.
You can find the right decision makers at your target accounts in seconds, for free:
People Finder Tool
FIND PEOPLE
Type domains, select personas, fetch real contacts.
Type LinkedIn company domain and press Enter to add
Persona 1
Define specific job titles and seniority levels to target the right decision makers
4. The 7 Gate Lead Qualification Pipeline
The qualification skill is where most of the intelligence lives. Earleads runs every scraped lead through seven gates before it gets added to an outreach list.
Gate 1: Deduplication
Checks against the existing database. If the lead is already in a campaign or marked as a customer, it gets filtered out immediately.
Gate 2: Headline and title pre qualification
Reads the LinkedIn headline and current title. Filters out obviously wrong fits before any enrichment credits get spent.
Gate 3: Exclusion list
Runs against clients, partners, competitors, team members, and freelancers. If the lead appears on any exclusion list, it gets dropped.
Gate 4: Profile enrichment
For surviving leads, Claude Code fetches the full LinkedIn profile and pulls the data needed for scoring.
Gate 5: Country and role qualification
Checks geography and seniority against ICP criteria. A founder in a target country stays. A freelancer in a non target region drops.
Gate 6: Company qualification
Filters by company size, type, and industry. Earleads targets B2B SaaS and tech companies with 5 to 500 employees. Google, for example, gets excluded because it falls outside that band.
Gate 7: Best company from experience
This is the creative one. Claude Code reads the context from past successful engagements and identifies pattern matches. If most of Earleads' best clients had marketing co founders, leads at companies with only technical founding teams get down ranked.
Every lead that clears all seven gates gets a score, a tag, and a recommended action. Top tier leads flag for manual review. Mid tier leads route to automated outreach with a specific message variant. Lower tier leads get a different angle or get dropped entirely.
LinkedIn post engagement is one of the strongest intent signals a B2B team can act on. People who engage with specific content are showing exactly what they care about.
You can see which companies are actively researching solutions in your space right now, for free:
Intent Signals Tool
Fields marked with * are required
Quick examples:
5. How the System Refines Your ICP After Every Campaign
The point Othmane keeps coming back to is the feedback loop. This is where Claude Code does something Clay cannot.
After every campaign, Earleads feeds the results back into the system and asks Claude Code to refine the qualification criteria.
The logic goes like this:
→ Define the success KPI (connection acceptance with a salesy note, positive reply rate, meetings booked)
→ Compare the winners to the losers
→ Ask Claude to identify the hidden patterns
Example. A campaign targeting HR leaders across multiple titles gets a 30% reply rate overall. But when Claude looks deeper, it finds that profiles with "people" in the job title hit 67% reply rate while "employee wellbeing" titles hit 12%.
Without Claude Code, this analysis takes a team of analysts hours of manual work. With Claude Code and the memory file holding all past campaign data, it takes one prompt.
The system does not stop at pattern recognition. It uses the new pattern to rewrite the qualification rules for the next campaign. Then it runs the next campaign. Then it refines again.
Clay cannot do this. A Clay table is a static artifact. You build it, you run it, and if you want to change the logic you duplicate the table and rebuild every column. There is no automatic loop from campaign output back to campaign setup.
Claude Code closes that loop automatically.
There is a limit. Claude does not have taste. It can spot quantitative patterns but it cannot always tell you when an offer is wrong or when a completely different angle would convert better. That is where human intervention still matters.
6. Building Custom Skills for LinkedIn Actions
The skills are the building blocks that make the whole system work. Here is how Othmane approaches building one.
Step 1: Find the SDK for the API
Every action needs an external environment. Scraping LinkedIn, sending messages, enriching contacts. Each environment has an API and each API has an SDK that documents every available action.
Before writing the skill, Claude Code reads the SDK so it understands how to talk to that API.
Step 2: Define the input and the output clearly
Vague skills produce vague results. A scraping skill that says "get the commenters" will sometimes stop at 50. A scraping skill that says "get every single commenter on this post" will not stop until it has all of them.
Explicit success criteria matter.
Step 3: Give the skill access to the right environment
This is where the API connection gets wired in. Credentials, endpoints, rate limits, error handling. All of it lives inside the skill.
Step 4: Test it
Skills fail the first time. Claude Code tries to comment on a LinkedIn post and posts random text instead of the intended reply. The fix is inside the skill, not inside the prompt.
Once a skill works, it becomes reusable. Every future campaign that needs to scrape LinkedIn engagers calls the same skill. No rebuilding from scratch.
7. The API Stack That Powers the System
Three APIs do most of the heavy lifting.
Unipile handles all the LinkedIn actions. Scraping post engagers, sending connection requests, sending messages, replying to comments. It also covers WhatsApp and email inboxes, which means a team can standardize outreach across all three channels through one provider.
Firecrawl handles web fetching. When the system needs to pull context from a company's website, news articles, or funding announcements, Firecrawl does the scraping.
Apify acts as an aggregator of specialized scrapers. Google Maps, Instagram, niche platforms that do not justify their own subscription. Apify lets one account access hundreds of scrapers through a marketplace model.
For contact enrichment, the same approach scales through APIs like Prospeo, FullEnrich, OpenMart, and PredictLeads. Each one plugs into a skill. Each skill gets called when the campaign needs that specific data type.
Contact discovery sits at the heart of most outbound workflows. The same APIs Earleads uses for enrichment also power the tools on our site.
You can find verified email addresses for your target accounts here, for free:
Email Finder Tool
Your target's information
Fields marked with * are required
Enter just the domain (e.g., company.com) or paste a website URL
Please enter a professional email address
Please fill in all required fields
8. Why This Changes How GTM Teams Will Work
The deeper shift Othmane flagged is about what work even means when one person can run five or six agents in parallel.
The developer era rewarded people who could stay focused on one deep task for hours. The Claude Code era rewards people who can hold multiple contexts at once, hop between agents, give feedback, and keep the whole operation moving.
Othmane called it the ADHD era. When every task you start runs in the background, the skill shifts from deep focus to high bandwidth orchestration. You give context, you review output, you redirect, you move to the next.
GTM teams that treat Claude Code as an automation layer miss the point. The value is not automation. The value is compression. Five different platforms collapse into one terminal. Five different roles collapse into one person who can direct five agents.
The warning Othmane left with was direct. AI slop is not automation that helps. It is automation that removes the craft. The commodity layer is growing fast. The value is moving up the ladder toward creativity, taste, and speed of iteration.
Use the tools to do better work. Not just more work.
Claude Code gives teams visibility into every lever of their GTM stack. The same diagnostic thinking scales up to the strategy level, where founders can stress test their current motion against what is working elsewhere.
If you want to understand where your GTM motion stands today, see how your current approach compares to these specialized models:
GTM Report Tool
9. Conclusion
The Earleads system shows what becomes possible when a GTM operating system collapses into a single terminal. One prompt. Seven gates of qualification. Three APIs. Dozens of reusable skills. Full context stored in memory and referenced on every run.
The competitive edge does not come from the automation itself. That is becoming a commodity. The edge comes from how fast a team can test creative angles, refine them based on real results, and move to the next hypothesis.
Claude Code does not replace taste. It frees up the time that taste needs to develop.
What LinkedIn action would you automate first inside Claude Code?
---



