How I'm building

Normally in this newsletter, I'm wearing my analyst hat. Sharing data on the market and my takes on where things are going.

But since making Adam's GTM Report my focus, I'm also wearing a builder hat. 

Over the last year, I've learned to build with agents and radically changed how I work. 

My outputs are the live market tracker (software + data), this newsletter (content), client deliverables (services), and my frequent posts on LinkedIn. Every output is my own. It's my voice, my judgement, my curation. But the agents behind the scenes are helping me ship more quality in less time than I'd ever imagined. 

Today I want to show you some of the core pieces in my system.

If you work in a bigger company, your situation is different. You've got legacy systems, teammates, and a multiplayer environment to think about. I'm in a privileged position as a solo operator: I can move fast and be very experimental.

My details probably won't fit your environment. But some of the principles and lessons likely will. To help make that connection, paste the prompt below into your favorite AI system:

Subscribers are the center of my universe (that's you), so it makes sense to start here.

If you're building a modern B2B brand, you are probably doing similar activities to build an audience and/or community. I believe that every brand should have a newsletter, should publish on social media, and be thoughtful about how they engage their target audience wayyyy before the demo request. It's that old 95/5 rule.

That means I need more than a place to send emails. I need to know who's here, what they care about, and when a more personal invitation would be useful. Here's how I see the audience ops system.

The center of this is beehiiv. That's where I manage subscribers, publish, and run workflows to different segments at key moments in the journey. I used to manage my newsletter on Substack and was frustrated that I couldn't have more control over how I manage and communicate with my audience. 

I use Codex and Claude Code with beehiiv's API and MCP to build workflows that segment subscribers, integrate other tools, and create different special engagement paths when relevant.

Here's one example.

I serve ~20K GTM leaders, builders, and practitioners. That's a pretty large diverse audience and the TAM there is wide.

Within that audience is a much smaller group of marketers and execs at the GTM Tech brands in my Index. This is a narrow, defined TAM. When they subscribe, I have an additional objective: ask them to claim and verify their company's profile through the Brand Portal.  It's free and the benefit to them is usually very clear -- a more accurate profile in my index which helps their visibility.

My agents use ZoomInfo and FullEnrich for subscriber enrichment (both via API). The workflow matches subscribers to Index companies (custom code), checks verification status and contact history, and prepares a Gmail draft. My job is to review and send it. Once they take action, an automated workflow takes over based on the steps they've taken. Enriched subscriber details get sent back to beehiiv. I keep tailoring content and messaging as the relationship develops. 

Here's what I see in my Drafts folder after a week of running this. I can personalize the emails a bit and then send at my convenience. 

These audience ops workflows give me richer subscriber profiles in beehiiv that I can use for personal engagement and content going forward. I now have 22 custom fields and 44 subscriber tags. Best part: I didn't configure a single field or tag by hand. My agents built it all (using beehiiv's API and MCP). Here are some of the fields behind the verified-brand use case, including claim status and interest in partnerships.

I love these kind of "find the needles in the haystack" workflows. I think they apply for any B2B marketer, not just a solo operator. Imagine your field team is running a dinner for C-level leaders in Chicago. Find subscribers with the right titles, at your target accounts, in that city. Tee up personal invitations and track who replies and attends.

In my opinion, your audience ops should be a separate system from your lead management and CRM. You want to integrate these systems, but avoid SDRs pushing demos on subscribers before you've had a chance to really engage and earn trust. 

As you know beehiiv sponsors this newsletter. I was building on it before we became partners. They are kind enough to offer my subscribers 30% off for three months when you use code ADAM30.

Keeping my research and market tracker current takes a lot of recurring work. I wanted that work to run while I sleep (literally). 

My /night-gtm skill is the entry point to a scheduled process that starts at 10:30 p.m. It's a big skill that orchestrates other skills and scripts inside it. A runner controls the order, timing, and dependencies. Here's a simplified view of what's inside. 

Collect and calculate. Scripts pull posts, jobs, feeds, and website changes, preserve the evidence, and update metrics. The predictable parts can run the same way each time. This uses Apify, web search, and custom web crawlers that Claude Code built on Playwright. It's all purpose-built for my exact goals. 

Interpret and update. Agent skills enrich links I've saved, look for interesting hiring patterns, update wiki pages, and edit the new research. They work from the collected evidence and my instructions.

Check and report. The system checks freshness, links, failures, and unfinished work. It prepares a research summary and the morning emails. Telegram gives me a completion update; my admin page holds decisions that need me.

In the morning, I can see what changed in the market, which of my research threads advanced, how my content performed (including beehiiv opens and clicks), what happened to things I sent in, and where I need to make a call. I can follow the links into the underlying work.

It keeps improving through changes we make to the machinery. A repeated mistake becomes a better instruction, a check, or a script fix. My feedback can change what the next report emphasizes. Running it again doesn't magically teach it anything; saving the correction does.

For your team, the inputs might be customer calls, campaign results, and CRM changes. Decide what you want waiting in the morning, then work backward: what can code handle, what needs interpretation, and what still needs a person?

Coding agents aren't just for coding. I'm not an engineer. I hadn't touched a terminal until October last year. I didn't have a GitHub account until January.

But once I started using Coding Agents (Claude and Codex) in a local workspace, I found a next gear working with AI. And most of my output is NOT coding.

The environment has three main pieces:

Local workspace. Research, drafts, graphics, and code live in project folders, with history synced to GitHub. The work stays with me when I change agents.

Multiple agents. I started with Claude Code. Then added Codex. Both use the same context and skills. They can work on different tasks, review each other's work, or give me another approach when one gets stuck.

Context. My goals, business strategy, voice, and project decisions live in files. My market research lives in a wiki based on Andrej Karpathy's LLM Wiki. For example, the CRM page connects company launches, sources, and earlier research so the next project can build on what we've already learned.

I use each deliverable to improve the system. I come in with something real: “Make a graphic to explain my local system.” The agent makes a version. I react to what's unclear, what's missing, and what works. We iterate.

Then we keep what will help next time: instructions in the skill, preferences in my context files, and good outputs as examples. The next graphic starts with those improvements.

That's a setup you could build around one recurring campaign, report, or research project. Keep its context, outputs, and lessons together so the next attempt has a better starting point.

Under all of this is a stack of APIs. My agents reach these tools through APIs, CLIs, and MCP connections.

My test for any new tool: can my agents connect, manage it, and use it to help our goals?

I never want to log in to these tools unless I need to upgrade my account or get another API key. That's the goal I'm building toward.

These are tools I use or have built with across the system.

Here's a GTM example. For my Dreamforce and UNBOUND research, Codex used Apify to collect public posts and Jev to classify them. We saved the evidence locally, analyzed the conversation, and turned it into a dashboard and downloadable build kit. Netlify hosts the experience; beehiiv handles subscription and resource delivery.

I shared the analysis and my Apify + Jev experiment on LinkedIn. If you want to try the collection workflow, here’s my Apify partner signup link.

The same approach could help a marketer understand the conversation around an industry event and create something useful for that audience. The APIs supply the capabilities. The agent connects the work. I still decide what question to ask and what is worth publishing.

I'm excited about how much I can build now and the opportunity in GTM teams. I don't think any of us has this figured out. We're all learning as we go. 

Right now I'm testing Instinct, OpenAI dots, and Grok Bot to see if/how/when they fit into this world. So far Grok Bot wasn't for me, but dots have added an interesting way to literally talk to my system (via voice). 

The graphics in this newsletter took a lot of back and forth with the agents. I had to decide what mattered, explain what wasn't working, and keep pushing until the pictures made the ideas clearer. That's been a big part of learning to work this way. Each turn makes the system better. 

As all the tools and models change, I expect that core idea to stick. Start with something you already need to ship. Bring your context, work through the rough versions, and save what you learn so the next project has a better starting point.

What are you building? 

I'd love to hear what's working for you and where you're getting stuck. 

I read all replies.