I design and build AI agent pipelines that plug into the tools and workflows your marketing team already runs — not ones that force you to change how you operate. My background is UX and human-systems engineering, so these systems are built around how your people actually make decisions, not just what's technically possible.
An agent system for a marketing team isn't one clever bot — it's an orchestration layer, a set of human checkpoints, and the actual marketing work it plugs into. I design and build all three together so they hold up in daily use.
I build the pipeline itself — Claude and Claude subagents doing specialized work, coordinated through Hermes and handed off cleanly from one step to the next, instead of one model trying to do everything at once.
This is where the UX and human-systems background matters most. I design where a person reviews, approves, or overrides the agent's work — inside Slack and Discord, where your team already communicates — so the system earns trust instead of getting routed around.
The pipeline connects directly into the marketing work it's meant to support — campaign execution, reporting, and the day-to-day operations that used to eat a team's time before a human ever looked at strategy.
Wiring Claude to a few APIs is the easy part. What actually determines whether a marketing team keeps using an agent system six months in is whether people trust it, understand when to step in, and were never asked to abandon the tools they already know.
My background is UX and human systems engineering, not just prompt engineering. Before I write a pipeline, I map how your team actually makes decisions today — who approves what, where mistakes are costly, where speed matters more than precision. The agent system gets built around that map, with checkpoints in the channels your team already lives in (Slack, Discord), instead of a new dashboard nobody opens. The result is a system that fits into how marketing work already happens, rather than one your team has to be retrained around.
The practical questions behind how I design, build, and integrate agent pipelines for marketing teams.
A small set of AI agents (built on Claude and Claude subagents, coordinated through Hermes) that each handle a specific piece of marketing work — drafting, research, reporting, campaign monitoring — and hand off to each other and to a human at defined points, rather than one general-purpose bot trying to do everything.
Claude and Claude subagents for the core agent work, Hermes for orchestration, ChatGPT where it's the better fit for a specific task, and Slack or Discord as the interface your team actually interacts with. On the marketing-ops side, I connect pipelines into Google Ads and whatever reporting tools your team already relies on.
The goal is to remove repeatable, low-judgment work from your team's plate — not their judgment itself. A human-systems approach means I'm explicitly designing where a person needs to stay in the loop, not just automating everything technically possible and calling it a win.
Through review and approval checkpoints built directly into Slack or Discord — wherever your team already communicates. Where a mistake would be costly, the agent stops and asks. Where speed matters more than precision, it moves on its own. Deciding which is which for your team is most of the design work.
It starts with mapping how your team currently makes decisions and where the real bottlenecks are — not a tooling conversation. From there I design the agent pipeline and the human checkpoints together, build it against your actual tools, and run it alongside your team rather than handing over a black box.
Adoption first — is your team actually using it a month in, not just the week it launched — then time saved on the work it replaced, and the quality of what it hands back. A technically impressive pipeline nobody trusts enough to use isn't a win.
If you're running a marketing team or agency and want to know whether an agent pipeline actually makes sense for how you work — not just whether it's technically possible — send me a note. I'll walk through your workflow and give you a straight answer.
alexd.com