Forwarded this email? Subscribe here for more.
👋 Hey there, I'm Nick. Each week, I share field notes from the people building the future of marketing. For more: LinkedIn | The Marketing Engineer | Profound University.
Lucy Hoyle might be the truest definition of a marketing engineer I have met so far.
She is a content marketer by background then she became Carta's first content engineer, helping support a company-wide Claude rollout to 1,800 people. She went to Profound's first Marketing Engineer Hackathon and used the time to start building a Git-based marketing brain for her team.
Then I went and stole her idea 😂
Her idea was so good that it made me want to go build.
The best marketing engineers are not just 100x marketers. They help other people become 100x marketers too.
That is what makes Lucy so valuable. She builds systems, but she also trains, enables, explains, and creates shared infrastructure. She is using technical fluency to multiply the output of the whole company.
1. Token discipline is a strategy, not a leaderboard.
Lucy was the first guest who talked about token efficiency in a way that felt mature.
At Carta, they monitor Claude usage and have Slack alerts if someone spikes too hard. That could sound restrictive in the wrong culture. Lucy described it differently.
Let's not try to token max. That has probably been more useful for us understanding how to really work with AI.
The goal is not to compete on some arbitrary token leaderboard. The goal is to maximize leverage.
That means asking better questions:
Does this task actually need an LLM?
Is the model adding judgment, or are we using it because it is the tool in front of us?
Could a repeatable Python script do this better?
Where should we spend tokens because the work requires reasoning, and where are we just burning context?
That is the difference between AI adoption and AI discipline.
A lot of teams are still in the phase where usage itself feels like progress. Lucy's point is more advanced. Usage is not the goal. Better work is the goal.
2. A marketing brain makes context reusable.
At Profound's hackathon, Lucy started building what she called a marketing brain.
The problem was familiar: everyone gets excited about Claude, then starts building skills, plugins, scripts, prompts, and campaign workflows. Soon the team has useful things scattered across Google Drive, local machines, and one-off files. Context gets hard-coded into SKILL.md files. Version control breaks. Nobody knows which skill is current.
Lucy wanted a shared context layer in Git.
I wanted to see if we could build a shared context layer, which I termed the marketing brain, and then build skills and scripts that could reference that context layer.
I am building my own GitHub-based context layer for skills at Profound because Lucy's idea made immediate sense. Skills, scripts, campaigns, and source-of-truth context should not live in a pile of disconnected files. They should live somewhere durable, versioned, and easy for agents to reference.
This is one of the big themes across these episodes.
Matt Swulinski built a personal marketing OS that became team infrastructure. Reema Batta talked about Figma's context layer. Holly Chen described Samsara's agent marketplace and governance model.
Lucy gives the builder version: put the shared brain in Git, then let skills and scripts pull from it.
3. Context rot is why more context can make AI worse.
The most technical idea Lucy shared was context rot.
Most people start with the assumption that more context creates better output. Give the model more docs, more files, more notes, more examples, more everything.
Lucy learned the failure mode. The context window is finite. When you overload it, the model has to compress information, and that compression can degrade the working memory of the session.
Her answer was retrieval discipline.
The BM25 script will match the keywords to context that lives in the marketing brain, and then it will pull down the relevant chunks of information, so Claude is only using the context that it needs for that specific use case.
She did not just say "we need better context." She researched the problem, found BM25, wired it into the system, and used it to retrieve only the chunks Claude needed.
This is why Lucy is such a strong example of the role.
She is a marketer first. She understands the content, campaign, and team problems. Then she is technical enough to build the system that solves them.
That combination is rare.
4. The soft skill is part of the system.
Lucy's advice on executive buy-in is also practical.
A lot of builders get excited about the implementation. BM25. Git. Claude Code. Skills. Context layers. Scripts. Retrieval.
Executives usually need a different entry point.
Think about what they care about. What is the cost of inaction? What are we losing by not doing this thing?
Your CMO probably does not care about BM25 at first. They care about why the system matters, how it changes the business, what time it saves, what risk it reduces, and what the team can do next.
Give the summary. Give the takeaway. Give the why. Give the next step.
If they want the implementation details, they will ask.
That is also part of marketing engineering. The job is not only building the system. The job is making the system legible enough that the business will support it.
What I am taking from this episode
First, I want to be more disciplined about token usage. The question is not "how much AI did we use?" The question is "where did AI create leverage?"
Second, I am building toward a Git-based marketing brain. Skills and scripts need a durable context layer, and Git is the most obvious place to start.
Third, I think Lucy is the clearest example so far of the marketing engineer ethos. She has become technical without starting as an engineer. She builds systems, teaches the team, increases her own output, and helps everyone around her get better.
That is the bar.
The best marketing engineers will not just be high-output individuals.
They will become force multipliers for the business.
Lucy already is one.
Watch or listen
If that's you, or you want it to be, stick around. Hit reply and tell me what you're building right now. I read everything.
See you next time,
Nick