Lifecycle email is product logic. Treat it like product logic.

Most companies don't have a lifecycle "codebase." They have a silo.

It starts reasonably: someone owns marketing, and they pick the platform they already know — Customer.io, Brevo, Klaviyo, whatever they used at the last job. The account lives on their email address. They wire up the welcome series, the trial nudges, the win-back flow, clicking it together in the visual builder. It works. Nobody else ever looks at it.

That's the whole system. Not a second codebase — a second world, with one inhabitant.

The silo fails in predictable ways

It lives and dies with its owner. The workflows are on their login, named in their private shorthand, documented nowhere. When they leave — or just go on holiday during a billing incident — the company discovers that nobody else can even read the lifecycle layer, let alone change it.

Nobody ever sees it. Product changes ship through review; everyone can read the diff. The trial-expiry flow changes when one person drags a node on a canvas, and the change history is a modified-date. There is no PR to comment on, no test to fail, no diff to read. Problems surface as customer replies.

The owner becomes the bottleneck. Every lifecycle change — new trigger, new condition, new message — queues behind one person clicking through a UI. Meanwhile the product team ships features continuously, each one changing what users do, which events fire, what "activated" even means. The silo doesn't just move slower; it falls further behind every sprint, because its inputs are moving and its throughput is fixed.

AI made the gap embarrassing

This was survivable when everyone was slow. It isn't anymore. An engineering team with AI-assisted development ships in an afternoon what used to take a week — new features, new flows, new events. The lifecycle layer's clock speed didn't change: it's still one person, clicking.

The result is a widening mismatch at the exact seam where product meets revenue. The product outruns the messaging about the product. Onboarding emails reference screens that were redesigned two sprints ago. The upgrade nudge fires on an event the product stopped emitting. Nobody notices, because nobody can see it.

And the thing AI accelerates is code. An agent can write, review, and test a TypeScript journey; it cannot click through Customer.io's UI on someone else's login. Locking lifecycle logic in a dashboard doesn't just keep humans out — it keeps your fastest new contributor out too.

What treating it like product logic means

Move the lifecycle layer into the repo, next to the product it serves:

• Shared, not owned. Journeys are files. Anyone on the team can read them, and the person who knows the billing edge cases reviews the trial-expiry change before it ships.
• Versioned. When reactivation tanks, git log says what changed and git revert fixes it. No archaeology in a vendor UI.
• Same vocabulary as the product. The Events.TRIAL_STARTED constant the journey triggers on is the one the product emits. Rename it and the compiler finds every break — the silo's "event that quietly stopped firing" class of bug becomes impossible. (Your event names matter more than you think.)
• As fast as the rest of the team. A lifecycle change is a PR, which means it moves at the same speed as everything else — including when an AI writes it.

Marketing still needs to see what's running — that's a read surface (dashboards, send logs, journey state), and it can be excellent without being an editor. But authorship belongs where the speed is.

The silo made sense when email tools were the only way to send email. Now it's a single point of failure attached to your revenue, moving at 2015 speed while everything around it compounds. That's the gap.