The ladder was always context
Matteo Collina maps out three tiers of software engineering emerging as AI collapses implementation costs. Digital-native tech companies where senior engineers review AI output. Large enterprises supplemented by fractional senior expertise. Small businesses getting custom software for the first time via AI-assisted developers acting as "software plumbers."
The tiers are different but the pattern is the same — at every tier, the bottleneck moved from "can we build this?" to "does someone understand what we're building and why?"
Context is the bottleneck at every tier#
- Senior engineers reviewing AI output at Tier 1.
- Architects making platform decisions at Tier 2.
- At Tier 3, someone still has to understand the local business problem well enough to specify the right solution (that's my read, not Collina's, but it follows from the pattern).
The implementation is increasingly automated. The context isn't.
This is the thread connecting planning over prompting, context curation, and comprehension debt: code that works fine until someone needs to change it and nobody remembers why it's shaped the way it is.
It was never about writing code — it was about codifying#
Collina flags the talent pipeline risk: AI handles "junior-level" tasks, junior hiring slows, the supply of future seniors dries up. He frames this as the career ladder losing its bottom rungs.
The deeper issue is what the ladder was always teaching. Years of building, maintaining, breaking, and fixing systems were never just about learning to code; they were about accumulating context. How systems behave under load. Why that module is shaped the way it is. What the business actually needs versus what it asks for.
The context work was bundled with the implementation work. They looked like one thing.
AI unbundled them. The ladder wasn't an implementation ladder. It was a context-acquisition pipeline. That reframes the risk. It's not only that juniors won't become seniors. It's that at every tier, people are shipping code without the context to understand it.
A study on AI-assisted skill formation already shows this happening: developers who fully delegated to AI gained some productivity but lost conceptual understanding, code reading ability, and debugging skills. What Osmani calls comprehension debt is context loss by another name. That's not a future pipeline problem. It's a present-tense context crisis, and the broken ladder is a symptom.
The organizations that recognize context as infrastructure — not overhead, not a side effect of seniority — are the ones that will still be able to maintain what they build.
Related writing#
- The power of the pause: How planning beats prompt tuning (Sep 2025) — if the bottleneck is judgment, planning is where the value lives
- Developers are becoming context technical writers for AI (Sep 2025) — the context-holding work Collina describes across all three tiers is a form of context curation
- Context engineering beats prompt tricks for AI agents (Oct 2025) — framing the input well matters more than generating the output fast
- I AIn't known what's in my code anymore (Mar 2026) — what happens when teams stop understanding what they shipped