The short version
- The role is content infrastructure. Generators, publish gates, CMS migrations and verification scripts, and almost none of it arrives as a brief.
- The screening question: how would you enforce a house style rule across 200 published posts? An audit of ours found 2,801 violations, and roughly a third could not be fixed by find and replace.
- Marketing inherits production risk on client systems. On one account the publish path was a git push and the obvious deploy command would have overwritten the live site.
- The bias arrives with the hire. Engineers build, so the build or buy test has to run before anyone produces an estimate.
Three AI engineers work inside our marketing team, and the work is content infrastructure rather than web development. A generator that turns one spec into 255 pages. A validator that refuses to publish a page below the standard, and the migration tooling for the day 387 records have to leave somebody else’s CMS.
What does an engineer actually do on a marketing team?
Builds the machinery content survives at volume on. The largest build is a 255-page programmatic engine, where planning and generation ran as one pass and verification and the de-AI sweep as another. Generation is fast and confident. Verification is slow, and running them together lets the confident half win.
The rest is enforcement and repair. A publish gate guarding 44 content pages on one account. A migration lifting 42 routes, 387 entries and 545 assets out of a hosted CMS, where roughly one record in six carried a legacy field or a reference with no destination in the new model.
The work, at the size it actually arrives:
| The job | Scope on one account | What it is not |
|---|---|---|
| Page generation | 255 pages | Writing 255 pages |
| Publish gate | 44 pages guarded | A rule stated in a brief |
| CMS migration | 42 routes, 387 entries, 545 assets | A redesign |
| Style sweep | 2,801 em-dashes across 200 posts | Find and replace |
| Verification | Counts on the rendered page | Reading the draft |
How do you interview for this role?
Hand over a real repair job with a trap in it. Ours: a documented house rule of zero em-dashes, and an audit of 200 published posts that found 2,801 of them.
The fast answer is a global find and replace, and it is wrong. Roughly a third of those dashes joined two independent clauses, so a blanket comma would have left a comma splice everywhere one sat. Whoever reaches for classification before replacement is the hire.
What risk does marketing take on?
Production access to somebody else’s site. On one rebuild the project document listed a staging URL that had been correct a year earlier and by then served a different application. Production ran a different framework from the repository we were handed.
Publishing was a push to a branch rather than a deploy command, so the obvious command for that stack would have put a half-built version over the live site. We found it by making one trivial visible change and following it end to end. Learning how an account publishes now happens before anyone has write access.
How many engineers does a marketing team need?
We run 8 client accounts and 11 brands with 3 AI engineers inside the marketing team. That is a ratio from one agency rather than a rule, and it holds because most of the work is shared. A generator built for one account’s page set is the generator for the next.
What does not transfer is the per-account knowledge. Each account carries its own deploy target and its own approvals, plus at least one command that cannot be undone, and that gets learned once per relationship.
Three engineers cover eleven brands because the infrastructure carries across accounts. Nothing about the writing does.
What does hiring engineers not fix?
The writing. Nothing in the tooling decides what a page should say, and the account knowledge behind that decision is still carried one person at a time.
Then there is what the team starts building. Engineers build, so a marketing team with engineers in it will build things it should have bought. The question that settles it: does a company already exist whose entire job is maintaining this thing? A course platform, yes. A publish gate encoding one client’s editorial rules, no.
The two pieces underneath this: what removing the handoff changed and why a written rule needs a script behind it.
Frequently asked questions
What is the difference between a marketing engineer and a web developer?
A web developer builds a site from a brief. A marketing engineer builds what the content runs on and holds an opinion about the content: page generators, publish validators, CMS migrations, and the verification that proves a rendered page meets the standard.
How do you interview an engineer for a marketing role?
Give them a repair job with a trap in it. Ours is a library of 200 posts carrying 2,801 violations of a documented house rule. A global find and replace looks correct and is wrong, because roughly a third needed classifying first.
How many engineers does a marketing team need?
We run 8 client accounts and 11 brands with 3 AI engineers on the team. The number does not scale with accounts, because the infrastructure is shared. A generator built for one account's page set becomes the generator for the next.
Do you need an engineer for programmatic SEO?
Past a few dozen pages, yes. Our largest build is 255 pages generated from a spec, with generation and verification run as separate passes. Doing that by hand is not a smaller version of the same job.
What are the risks of giving a marketing team deploy access?
Overwriting a live site. On one client rebuild the publish path was a push to a branch, and the standard deploy command for that stack would have put a half-built version over production.
What goes wrong after you hire engineers into marketing?
Two things. The team starts building what it should have bought, because a build looks cheap from inside a team that can do it. And production risk on client systems moves into marketing, where nobody held it before.
Sources
- Chua Network delivery data across 8 client accounts (internal fact bank)
- Chua Network engagement records, anonymized (internal experience bank)
