My friend's interview ended at 9:03 AM. Mine started at 1:00 PM.
He texted me the second he walked out: "Bro. They grilled me on inbox placement. SPF, DKIM, DMARC was the warm-up. They wanted code."
I had 3 hours and 57 minutes.
The setup
The role was at a B2B outreach company. I'd applied a week earlier with a resume that said "email infrastructure" without committing to specifics. I figured I'd wing the technical part — I've shipped cold outreach systems for years, I know the theory cold.
My friend interviewed at 8 AM his time. I was scheduled for 1 PM mine. Same panel, same hiring manager, almost certainly the same playbook.
His debrief was three bullet points:
- Explain how you'd detect whether an email landed in inbox, spam, or Gmail's Promotions tab — without access to the recipient's mailbox.
- Walk through a system design for measuring deliverability across 10 mailbox providers in real time.
- Bonus: show code. Any code. They'll respect a hack over a deck.
Theory I had. Code I didn't.
The 4-hour sprint
I opened Claude Code at 9:11 AM with a coffee and a deadline.
Plan:
- Spin up a tiny seed-mailbox network — burner accounts at Gmail, Outlook, Yandex, Mail.ru, ProtonMail.
- Build an IMAP fetcher that polls each seed and classifies where the inbound email landed (inbox / spam / category tab).
- Wrap it in a CLI that takes a "from" address, sends a test message to every seed, and prints a placement report.
Claude Code wrote the scaffolding in about twenty minutes. NestJS + BullMQ for the queue, IMAP via imapflow, classification logic based on Gmail's X-Gm-Labels header and folder names for the other providers.
The real time went into edge cases:
- Outlook's "Focused" vs "Other" inbox — neither is spam, but recruiters confuse them constantly.
- Yandex returns folder names in Russian. Hardcoded fallbacks needed.
- Some providers throttle IMAP polls aggressively. Had to add jittered retries.
By noon I had a working CLI. I sent a test from my personal Gmail to all five seeds, waited 90 seconds, and got back:
gmail.com → INBOX
outlook.com → JUNK
yandex.ru → INBOX
mail.ru → INBOX
proton.me → INBOX
Outlook eating a clean Gmail send was a useful surprise — exactly the kind of thing you'd want a tool to catch before sending 10,000 emails.
Screenshotted the output, pushed the repo to a private GitHub, ate lunch.
The interview
The first technical question was almost word-for-word what my friend warned about. I answered the theory part, then said: "I actually built a small reference implementation this morning. Want me to share my screen?"
The hiring manager paused for two seconds. "Please."
I ran a live test against their own marketing domain. Three providers came back clean, two came back spam. We spent the next twenty minutes debugging why — DMARC alignment was off on their no-reply subdomain.
I didn't pitch myself as the candidate who knew deliverability. I pitched myself as the candidate who could ship a deliverability tool between breakfast and lunch.
Offer arrived three days later. I declined for unrelated reasons, but that's a different post.
What Claude Code actually changed
The honest version: I could have written this prototype before Claude Code. It would have taken two days, not four hours, and I would have spent half of it on IMAP boilerplate I've written six times in my career.
What changed isn't capability. It's the size of the window where preparation is possible.
A four-hour gap used to be coffee and panic. Now it's a working prototype with edge cases handled. That's a different category of professional life.
I kept hacking on the CLI after the interview. It grew into a full inbox placement service — multi-tenant, hosted, with a free tier for one-off checks.
If you want to run the same test on your own sending domain, it lives here: https://check.live-direct-marketing.online. Paste a from-address, get an inbox-per-provider breakdown in about 60 seconds. No signup for single checks.
The four-hour version of this would have been embarrassing to ship publicly. The version after a few months of polish is what I would have wanted that morning.