TL;DR
- AI orchestrated now 80% of our GTM work
- Claude Code rebuilt our entire client onboarding from weeks to days
- Sending infrastructure: 40 minutes → 5 minutes
- LinkedIn gets 10x more replies than email — same message, same person
- Clay's new pricing changed the agency landscape
- The future GTM Engineer role is about orchestrating AI agents, not just tools
Three months ago, we were wrong about AI
In our first podcast episode back in December, we made a hypothesis: AI was overused in B2B sales. Most agencies were leaning too hard on automation, and the human element of outreach would always win.
Three months later, we've reversed our position completely.
Today, over 80% of our GTM work is done by AI — orchestrated by our team, but executed by AI agents. We're not just using AI as a copywriting tool anymore. We're using it to rebuild our entire operational stack: client onboarding, infrastructure setup, campaign creation, prospecting, and reporting.
This isn't a small shift. It's a fundamental change in how a GTM agency operates in 2026.
Claude Code: the biggest GTM shift since Clay
The single biggest change since our last episode is Claude Code — Anthropic's coding agent that runs locally and writes custom code based on natural language prompts.
For agencies, this changes everything.
Before Claude Code, when we needed to scrape a database that didn't have a clean API, we'd hire Python freelancers from Fiverr. Custom scraping projects took days and cost hundreds of euros per task. Now, we describe what we need in plain English, and Claude Code writes the entire scraper from scratch.
One concrete example: ShopTalk, a major ecommerce conference, doesn't have a public attendee list. Attendees are visible only after logging into the platform with active filters. Using Claude Code combined with a browser cookie editor extension, we scraped 400 targeted attendees in under an hour. Without it, this would have required hiring a developer or attending the event in person.
This isn't theoretical. We're using Claude Code today across the entire client lifecycle.
How we automated client onboarding end-to-end
The moment a new client lands in our Slack channel, the automation begins.
Three AI agents run in parallel: one analyzes the client's industry, one maps the competitive landscape, and one generates predictions about market direction. We feed in transcripts from pre-sale calls so the agents have full context before producing any output.
Within minutes, the system creates a custom Typeform onboarding document — pre-populated with hypotheses about the client's ICP, decision makers, pain points, and value propositions. A separate AI agent posts the form link directly into Slack with instructions for the client.
When the client fills it out, the data flows into what we call our **copy and offer bibles** — a database of our highest-performing campaigns and offers. Within hours, we have **80–90% of campaign copy ready** for review.
Total time from contract signature to first campaign draft: days, not weeks.
Building sending infrastructure in 5 minutes
This was the moment we knew the shift was real.
Setting up email sending infrastructure used to take 40 minutes per client: creating accounts, configuring DNS records, setting up sender names and signatures, pushing accounts into warmup tools.
Now it takes 5 minutes.
We feed Claude Code the relevant client data — sender names, signatures, profile pictures — through our Typeform pipeline. It builds the entire infrastructure end-to-end, integrates with our partners at Salesforge (whose API documentation made this possible), and pushes everything into the warmup queue.
Credit where credit is due: this only works because Salesforge built proper API access. We've tested similar workflows with other tools where Claude Code hits a wall because the API documentation doesn't allow the necessary steps. Tool quality is the new bottleneck for AI-native operations.
Email vs LinkedIn vs Multi-channel: the data
One client recently asked us to break down their results across channels. Their TAM is around 2,500 companies — small but specific.
Same message. Same decision maker. Different channels.
- Email reply rate: 2–3%
- LinkedIn reply rate: 20–30%
That's a 10x difference for identical content.
This doesn't mean email is dead. For broad targeting (think 10,000+ ecommerce decision makers), email-only campaigns still work because you need volume. But for focused outreach at 1,000–2,000 prospects, multi-channel beats single-channel by miles.
The key insight: treat every touchpoint as the first one. Even within a multi-channel sequence, your value proposition should be visible in every message — because the reply rate suggests recipients often see only one of them.
The "all-bound" approach
We're going beyond multi-channel into what we call all-bound: email, LinkedIn, and cold calling combined.
Cold calling reply rates aren't dramatically higher than email (around 3% in the US, more in markets where cold calling is still socially acceptable). But once you get someone on the line, the conversion to a booked meeting is significantly higher because objections get handled live.
For B2B teams seriously entering new markets, the holy trinity is email + LinkedIn + cold calling on top accounts. Plus emerging plays like LinkedIn ads (Clay added ads recently) and outbound-led-inbound — sending personal invites to webinars or events to your ICP through LinkedIn.
Open Claw and autonomous prospecting
Claude Code isn't the only AI agent reshaping GTM. Open Claw, a free open-source autonomous AI agent, is becoming the second tool in our stack.
While our GTM unit hasn't fully adopted Open Claw yet, our broader RevOps team built something incredible with it: an autonomous prospecting bot that runs through Slack.
We have around 70 cold callers. Each needs fresh, accurate prospect lists daily. Manually building those lists was a full-time job. Now, anyone in the company can ping the bot in Slack with a target profile, and within 10 minutes get a complete list with verified contact details — phone numbers included.
The bot uses local Finnish databases, Prospeo, and other enrichment sources to find both the right decision makers and their contact info. This is happening autonomously, at scale, replacing what used to require dedicated headcount.
A note on risk: don't run agents like Open Claw on your main work device. Use a separate machine. Don't give it access to vulnerable data. The privacy and confidentiality risks are real.
Clay's new pricing: what changed and what it means
On March 11th, Clay rolled out a new pricing model. The old model charged credits based on enrichments. The new model has two tiers: **data credits** (for emails, phone numbers) and action credits (for any operation inside the platform — including third-party API calls).
This is a significant change for agencies.
Under the old model, we could build a Serper integration and scrape 75,000 US businesses from Google Maps for zero Clay credits, then enrich the list using our own OpenAI API keys for around $200. Under the new model, that same workflow would burn through credits in days.
To be clear: Clay isn't doing anything wrong. They've been transparent about the changes and they still allow legacy users to keep the old pricing. But the message is clear — Clay is moving toward enterprise customers, where one workspace serves one company. The agency model of running 10–20 clients off a single Pro plan is no longer viable on the new pricing.
For now, we're keeping our legacy plan and scaling it bigger. But we're also diversifying. Tools like Bitscale and Databar have credit systems that look more agency-friendly. And for many use cases, Claude Code can do what we used to do in Clay — for free.
The lesson: don't put all your eggs in one basket if the basket might change underneath you.
The future of the GTM Engineer role
If 2025 was about GTM Engineers orchestrating tools (Clay, Apollo, Heyreach), 2026 is about orchestrating AI agents.
Everything that was done manually should be automated. Every workflow that could be productized should become a bot — an infrastructure bot, a copy and offer bot, a dashboard bot, an analytics bot. The GTM Engineer becomes the conductor, not the player.
This creates a clear divide:
Manual agencies still doing everything by hand. Slow execution, inconsistent process, limited scalability.
AI-native agencies like ours that have systematized everything. Faster execution, repeatable process, compounding feedback loops from data.
For B2B companies choosing a GTM partner, the benefits of going AI-native are concrete:
- Speed: campaigns live in days, not months
- Systems: every process mapped, documented, repeatable
- Data feedback: more campaigns = more learning = better results
Three predictions for GTM in 2026
To close, three theses about where this is heading:
1. AI usage will only grow. Not just in GTM — across every business function. If you're not building AI into your operations now, you'll be behind in 12 months.
2. Competition will get fiercer. Inboxes are saturated. Volume alone doesn't work. You need to differentiate through process, channel mix, and offers — and through tactics that don't scale easily (like cold calling).
3. Distribution becomes the moat. Anyone can build a product with AI now. The hard part is getting it in front of the right people. The agencies that win in 2026 will be the ones who build distribution machines, not just lead lists.
.jpeg)
.jpeg)
