בדקתי את מערכת הדיוור שלי ומצאתי 17 בעיות

Last weekend I pointed AI agents at my own newsletter infrastructure with one instruction: show me all gaps, no sugarcoating. It came back with 17 findings. Three were bad enough that they were silently burning leads.

The three that actually hurt

1. Unsubscribe did not exist. The welcome email promised one-click opt-out. There was no link, no page, no List-Unsubscribe header. Resend injects RFC 8058 headers for Broadcasts, not for transactional sends, and the welcome was transactional. If you send marketing-ish transactional email, you have to build the opt-out path yourself.
2. Every reply bounced. Mail went out from a send-only subdomain with no MX and no replyTo. The booking recap said just reply to this email. Anyone who did got a bounce. For a solo consultant that is warm leads dying silently.
3. No Israeli spam-law layer. Section 30A wants opt-in proof, an advertising label, sender identity, and same-channel opt-out, stricter than CAN-SPAM. I had none of it. Now: double opt-in with stored consent timestamps, and a labeled subject line.

The loop, not the gaps

The interesting part is not the findings. It is the workflow. One agent audited the codebase, another researched 2026 sender requirements, and the two merged into a ranked table with file-and-line evidence. Fixes shipped as one PR the next morning. Then the same loop that found the gaps verified them against production: subscribed, confirmed, unsubscribed, cleaned up.

flowchart LR
  A["Audit the stack"] --> B["Research sender law"]
  B --> C["Ranked findings"]
  C --> D["One PR"]
  D --> E["Subscribe / confirm / unsubscribe"]

The one-line fix that stopped the bounces

await resend.emails.send({
  from: "Yonatan <[email protected]>",
  replyTo: "[email protected]",
  to,
  subject,
  html,
});

Total wall-clock from audit my stack to verified in prod: under a day. That is the pattern I keep seeing with production AI systems: the value is not the code generation, it is the audit, fix, verify loop running against real infrastructure with real evidence at every step.

If you want this kind of loop pointed at your own systems, book an intro call at yonyon.ai, or subscribe at yonyon.ai/subscribe to get the next issue first.