He Subnetted Perfectly. By the Third Office, the Routing Table Had 40 Unmergeable Lines.

A network technician gets promoted to engineer after two years in the NOC queue, and subnetting is second nature by now, /24s and /27s carved out without needing a calculator. First real design project: the company is opening a second office, then a third, each one needing a site-to-site link back to headquarters and to a shared AWS VPC. He does what got him promoted in the first place, carves a fresh /24 off the same flat 10.0.0.0/16 for whichever site is opening next, whichever block happens to be free. It works. Site one and two come up clean, traffic flows, nobody notices anything.

By site three, the WAN edge router's routing table has forty individual /24 routes in it, because none of those blocks were ever chosen to line up on a boundary that summarizes into anything larger. Every new site is another standalone advertisement, another line in every ACL and route-map that has to reference "everything behind site C" one /24 at a time instead of one summary route. Troubleshooting stops being "check the region" and becomes "check which specific /24 this ticket is even about." A senior engineer finally asks the question nobody asked at the start of the project: did you plan the address space so it summarizes, or did you just carve out /24s as sites came online. He hadn't heard the phrase "summarizable address block" before that conversation, and subnetting had never once let him down before, so it never occurred to him it was only half the skill.

The fix isn't a smarter subnetting formula, it's a planning step that happens before any device exists at a site. Allocate address space top-down instead of bottom-up: reserve a block per region or tier first, sized with real headroom, aligned so the whole region collapses into one summary route at the WAN edge. A /16 for the company splits into /20s per region, each /20 splits into /24s per site as they open, and because every site's /24 falls inside its region's /20, the WAN edge only ever needs to advertise the /20, not forty individual /24s behind it. Fewer routes means faster convergence, lighter router CPU and memory at scale, and filtering that actually works, because a route-map or prefix-list written against "this region's /20" doesn't need updating every time a new site opens inside it.

That same discipline is exactly what shows up again, in a different shape, once you're the one designing the BGP edge for a multi-site company talking to an ISP or a cloud provider: prefix filtering and route summarization at the data-center boundary aren't a separate BGP-specific skill, they're the same hierarchical-planning habit applied one layer up. Subnetting tells you the network is technically correct. Address planning is what keeps it operable once it has more than two sites in it, and almost nobody teaches the second half alongside the first.

The Network Engineer Book Bundle covers that arc deliberately, L1 through L3 in one 765-page set, 70+ labs and 450+ interview questions, subnetting and VLANs through enterprise routing through the senior-level BGP and data-center architecture that address planning was always quietly setting up for: https://resources.codelivly.com/product/network-engineer-complete-bundle-l1-l2-l3-foundations-to-advanced-architecture/