Blog
Sessions that actually ran.
The guides are the reference layer — written to still be true in a year. These are dated and opinionated: a real session, a real artefact at the end, and a link into the guide that generalises it.
Guides
Looking for the reference version?
Long-form practitioner guides, grouped into five categories and nine reading paths.
Point it at your stack
6 itemsFor someone already running lifecycle in Braze, Klaviyo or Iterable who wants Claude holding more of the work. Every post is a demonstration with an artefact at the end, not an explainer.
What a read-only Braze key already buys you — and where its ceiling actually is
Read-only access to a live Braze workspace covers most of the work worth handing a model: inventory, Canvas and template review, segment analysis, naming enforcement, data-model validation. The ceiling is real, but it is not the permission checkboxes. It is the API surface itself.
9 September 2026 · 7 min read
Which Canvas checks a machine should own, and which it shouldn't
Split a Braze pre-launch checklist by whether each item has a decidable answer. Template bindings, Liquid fallbacks, link integrity and conversion-event count are machine work. Whether the conversion event is the right one is not, and never will be.
9 September 2026 · 8 min read
The handoff packet: seven parts, and the build question each one kills
A lifecycle handoff is measured in the questions asked after it lands, not the pages in it. Here are the seven parts of a packet a builder can ship from without you, the two everyone skips and why the packet now has a second reader who forgets everything between sessions.
9 September 2026 · 8 min read
Stop writing the email design system. Derive it from what you already send.
A written design system describes the emails you meant to build. Your sent HTML describes the ones you actually built. Machine ingestion reads the second — here is what it can genuinely recover from a Figma file or a shipped template, what it cannot and the line where a module catalogue becomes a system.
9 September 2026 · 8 min read
Previews show you the email. They never measure it.
A preview screenshot answers whether the email looks like the design. It cannot answer whether a tap target is 41px, whether contrast lands at 4.2:1, or whether the body crosses Gmail's clipping limit. Those are measurements. A review thread full of approvals never takes one.
9 September 2026 · 8 min read
Your test profile is one of sixty-four emails
A personalised email is not one email. It is 2^n of them, and the pre-send check almost everyone runs looks at exactly one — whichever state the author's own profile happened to hold. Here is how to enumerate the rest before the send, and what a coverage tool has to refuse to do to be worth trusting.
9 September 2026 · 9 min read
Before you have a lifecycle team
6 itemsFor a startup with no lifecycle function, no analyst, and usually no ESP yet. Same depth, shorter on-ramp — these end with something shipped rather than something understood.
The first three emails you can ship before you buy any tooling
No ESP, no designer, no automation. Three emails you can write and send in an hour from the mailbox you already have — and the reason sending them by hand is the right first move rather than a compromise.
9 September 2026 · 7 min read
Your first 100 signups don't need automation. Here's the line where that changes
Automating too early is the most common early-stage lifecycle mistake. Signup count is the wrong trigger for the decision. Four tripwires that mean an automated flow now beats a founder typing it, and what automating early actually costs.
9 September 2026 · 8 min read
What has to be true about your domain before the first campaign leaves
Six decisions you make once, in order, before your first bulk send — and the specific thing each one costs when you get it wrong. Not an explainer on how authentication works; the sequence, the consequences and the record you publish today.
9 September 2026 · 8 min read
What a model needs from you before it can write as you
Asking for "friendly and professional" is why your AI-drafted emails sound like everyone else's. Swap the adjectives for evidence and you get your voice back — here's the one-page brief that does it.
9 September 2026 · 8 min read
What to do with 4,000 emails you collected and never mailed
You can mail part of a cold list, in a specific order, starting much smaller than you want to — and part of it you should delete without sending. Here's the triage, the send order and the numbers that stop it.
9 September 2026 · 9 min read
The weekly lifecycle review, when you're the only person who'd read it
At startup volumes your weekly email numbers are mostly noise, which means the report can't be about explaining movement. It's a five-line note to yourself in November — and this post ends with you having written the first one.
9 September 2026 · 8 min read
For everyone
1 itemThe argument that does not change with experience level. Both tracks link up to it.
Want this methodology executable inside your Claude sessions? Get Orbit free.