← Back to CredentialsPoint of View — Print to save as PDF
Point of View · Platform Engineering · Q2 2025

The Platform Engineering Trap: When IDPs Create More Toil Than They Eliminate

Internal developer platforms that aren't treated as products become expensive maintenance burdens. The four anti-patterns we see repeatedly — and how to avoid them.

Platform engineering has crossed the chasm from early adopter to mainstream enterprise practice. But with mainstream adoption comes a new failure mode: organisations building IDPs as infrastructure projects rather than product investments. The result is platforms that reduce toil for the team that built them and increase it for everyone else.
CELESDEV Solutions · Q2 2025 · 3 PAGES
PLATFORM ENGINEERING
The Platform Engineering Trap
CELESDEV Solutions · celesdev.com
Our Position

Platform teams that build for themselves instead of their users

The best-intentioned platform teams fall into the same traps repeatedly. They build what they find technically interesting rather than what engineers actually need. They measure platform adoption by the number of services onboarded rather than by the reduction in cognitive load experienced by the teams they serve. They underinvest in documentation and onboarding because the platform team already knows how it works.

Platform engineering succeeds when it is treated as a product discipline with real users, a real backlog, and real success metrics measured from the user's perspective. The technical work is straightforward — Backstage, Crossplane, and ArgoCD are well-documented. The product discipline is the hard part.


The Four Anti-Patterns
1
The golden path that fits nobody
A single opinionated deployment pathway designed around the team that builds it, not the service archetypes that need to use it. HTTP APIs, batch jobs, ML endpoints, and frontends have fundamentally different needs. One golden path fits none of them well.
2
The catalog that nobody trusts
Backstage catalogs populated by heroic manual effort during launch and never validated again. Stale catalog-info.yaml files propagate. After six months, the catalog is 40% inaccurate. Engineers stop trusting it. An inaccurate catalog is worse than no catalog.
3
Self-service that requires a ticket
"Self-service" infrastructure provisioning that requires a platform team approval step — because the Crossplane compositions aren't trusted to enforce policy automatically. The ticket queue defeats the purpose of the platform.
4
No developer experience metric
Platform teams that measure their own output (services onboarded, templates created, PRs merged) rather than their users' experience (time to first deployment, cognitive load survey scores, support ticket volume). Output metrics don't surface the friction that kills adoption.

The fix is product thinking, not technical work. Assign a product owner to the platform. Run quarterly developer experience surveys. Define golden paths per service archetype. Automate catalog validation in CI. Measure adoption by outcomes, not outputs.

Author: CELESDEV Solutions — Cloud & Platform Practice. Based on IDP design and adoption programmes serving 200–1,200 engineers.

CELESDEV Solutions · celesdev.com · business@celesdev.com
2 of 2