Last week I sat down with Jacob, founder and CEO of Relay, to talk through how small GTM teams can actually use AI agents without drowning in complexity. If you're a solo marketer or founder wearing 10 hats, this ones for you. Jacob runs marketing, sales, support, and pretty much everything else at Relay with a team of nine people and he does it by stacking simple AI workflows on top of each other. Want to see the full breakdown? Check out our step by step guide on how to build a webinar content engine with Relay or grab the playbook and book a call.

Top Takeaways

1. Start with stuff you repeat every week, then automate one piece at a time Most useful agents arent fancy. Theyre boring small workflows that handle the tasks you do over and over. Jacob automates webinar signups, reminder emails, promo posts, and newsletter drafts because he was doing those things manually every single week. And heres the thing, you dont need to build everything at once.

"In aggregate this looks very impressive and scary. Oh my god, I have 50 different AI agents working for me in marketing. But if you look at any individual one, theyre super simple." - Jacob

Pick one annoying task. Build the trigger. Add steps as you notice gaps. 2. Keep prompts simple and example based Forget the mega prompts. Jacob uses Claude by Anthropic for writing tasks and his entire prompt is basically "draft a LinkedIn post promoting this webinar, heres an example of a previous post." Thats it.

"I think people are really over complicating prompting. This is all the prompting you need. Given an upcoming webinar, please draft a LinkedIn post promoting it. Heres an example of a previous promotional post." - Jacob

Give the AI one good example and the inputs, then iterate from there. Dont waste time engineering perfect prompts when simple ones work fine. 3. Add human approval for anything public facing

Jacob has human in the loop steps baked into almost every workflow. For LinkedIn posts and emails that go out under his name, he reviews drafts in Slack before they publish. For lower stakes stuff like brand account posts, he skips the review because theres less risk.

Two types of human steps work best. One is review inside the AI step where you can edit back and forth. The other is a separate approval checkpoint for go or no go decisions.

4. Use templates for visuals instead of AI image generation AI image tools are great for cinematic stuff but terrible for simple text heavy posters. Jacob uses Google Slides templates with placeholders that get filled automatically by the workflow. Way more reliable than trying to get an AI to spell "webinar" correctly. 5. Content ideas come from customer conversations Every customer call has a nugget that other people care about. Jacob built an agent that sends him analysis emails when YouTubers he follows post new videos so he can study their hooks and topics. But his best content always comes from real conversations with users.

"Generate content based on customer conversations because always every single customer conversation Ive ever had has had a nugget that is valuable to many more people." - Jacob

6. Attribution tracking can waste time early on Jacob doesnt track ROI scientifically. He used to try measuring which LinkedIn posts led to paid customers but the data was mostly BS and added friction to onboarding. Now he just stays close to customers, watches engagement patterns, and understands the vibes of what resonates.

"Modern marketing especially like PLG self serve bottom up marketing is less of a funnel and more of like youre a planet building gravity." - Jacob

For organic content at early stage, being 100% vibes based is honestly fine. Save the attribution math for when you start spending money on paid.

Key Learnings

Maintain simplicity as you scale. Each individual agent should stay simple. Thats how you avoid building a brittle mess. Start with boring stuff like signup processing and reminder emails, then stack more workflows over time. Separate learning agents from publishing agents. Inspiration emails that summarize competitor content or customer feedback should marinate in your subconscious. Publishing agents that draft posts or send emails need tighter human review loops. Tune content goals to the outcome you want. Webinar promo posts should attract the right 100 to 200 people, not go viral. Founder account posts can aim for big reach. Brand account posts just need to show a heartbeat that your company exists.

Jacob and his team are proof that you can run a full GTM motion with AI doing the repetitive work while humans focus on strategy and relationships. The winners in 2026 wont be the teams with the biggest headcount. Theyll be the small tight teams that figured out how to get leverage from AI without losing their voice or shipping garbage content.

Ready to build your own system? Start with one workflow that annoys you this week and go from there.

• (0–3 min) Cold open + promise: how a small team gets GTM leverage with agents

• (3–6 min) Guest intro: Jacob’s background (Timeful → Google → Relay) and the “9 humans / 300 agents” model

• (6–12 min) Define the building blocks: triggers, workflows, agents, and HITL checkpoints

• (12–25 min) Workflow breakdown #1: Webinar ops engine (signup → reminders → follow-up)

• (25–35 min) Workflow breakdown #2: Content engine (recording → clips/summary → newsletter + LinkedIn drafts)

• (35–42 min) Workflow breakdown #3: Research/inspiration agents (customer convo mining, competitor/video summaries)

• (42–48 min) Guardrails & quality: structured outputs, approvals, on-brand knowledge, escalation paths

• (48–53 min) What’s newly possible now: Tables for logging/lookups, templates, model choice per step

• (53–60 min) Q&A + implementation checklist: what to build this week + how to avoid workflow debt