From e925fc134f8e863dad5f4fcd1299c8417f934258 Mon Sep 17 00:00:00 2001 From: Hermes Agent Date: Sun, 23 Aug 2026 10:09:02 +0000 Subject: [PATCH] Schedule (PR): module-boundary-not-network post for senior specialists/CTOs - proposed 2026-09-01 17:00 CEST --- .../personal/module-boundary-not-network.md | 50 +++++++++++++++++++ 1 file changed, 50 insertions(+) create mode 100644 linkedin-posts/personal/module-boundary-not-network.md diff --git a/linkedin-posts/personal/module-boundary-not-network.md b/linkedin-posts/personal/module-boundary-not-network.md new file mode 100644 index 000000000..70980e495 --- /dev/null +++ b/linkedin-posts/personal/module-boundary-not-network.md @@ -0,0 +1,50 @@ +--- +title: "You probably don't need a microservice. You need a folder with rules." +lane: personal +pillar: teach +author: paul-keen +voice: peer +icp_test: Does the "a module boundary, not a network boundary" take land with senior engineers, tech leads, and CTOs, and do they share the service split they are tempted to make this quarter? +image: "" +first_comment: | + (value post - no link; reply-CTA only) +utm_campaign: "" +utm_content: module_boundary_not_network +status: draft +stage: future +proposed_for: "2026-09-01 17:00 CEST" +notes: | + Pillar: teach (the 70% value; NO course link). Audience for this task: technical + senior specialists, tech leaders, CTOs. Reuses the finalized Maya-segment draft + (t_019ef725) but swaps the [HYPOTHETICAL] use case for a VERIFIED story-bank case + (vibe-coded MVP -> rewrite, not refactor: AI regenerates code, same logic ends up + in 5-6 places, one bug hides in several) so the sourcing gate is cleared before + scheduling. Peer register, no em dashes, close is a real question, no CTA/link/JT. + Image: left empty on purpose - text-only is a valid Holub-register choice for a + blunt take, and the repo's native-upload attach step is Paul's manual gap per the + README known-tooling-gap note. Paul can attach the boundary exhibit PNG at compose + time if he wants one. Queued (stage: future) for pre-verify; scheduled_for stays + Paul-gated per doctrine. +--- + +Most teams that say they need microservices actually need a module boundary, not a network boundary. The network part is the expensive part, and it is the part most teams are not ready for. + +A service is a deployment and a failure domain, not a folder. The conference talks show the clean diagram. They skip the distributed write you will hand-roll at 2am when two services need one consistent update. + +Here is the test I run before any extraction. Write one sentence: "Today, without this split, we are blocked because ____." If the blank is team coordination or deploy frequency, a boundary helps. If the blank is "the code is messy," extraction makes it messier, just farther apart. Run that sentence on your next refactor proposal. + +A team I worked with let an AI tool regenerate a feature. The same validation logic ended up in five or six places. One bug hid in several of them, and nobody could see it from any single file. We drew a hard module boundary first: one owner, one interface, one place the logic lived. The duplication collapsed. Eighteen months on, they still had not needed the network split. + +What is working: + +- Soft boundaries. Owned modules, clear interfaces, one repo. This bought most of the benefit for most teams I have seen. +- Strangler-style extraction done against a measured pain, like one service that genuinely needs independent scale. + +What is not: + +- Jumping to a message queue and separate deployables because a talk said coupling is the enemy, when the real bottleneck was a review backlog. The queue added a failure mode and fixed nothing. +- Treating "too big" as a reason to distribute. Size is not coupling. + +One claim I would not stake my name on yet: my read, not a measurement, is that teams under-estimate the operational tax of a distributed system by roughly an order of magnitude. Test it on your own setup before you quote it. + +What is the one service split you are tempted to make this quarter, and what sentence would prove you actually need it?