"Pyramids to columns." That is how Peter Diamandis described what is happening to engineering teams. The old model was three juniors for every senior. The new model, according to Diamandis, is one senior, maybe one junior, and AI filling the gap.
It is a clean metaphor. It is also a dangerous one to act on.
The hollowed-out middle
The trend is real. 23 companies have eliminated their junior and mid-level engineering pipelines in the last year, according to recruiter Soleyman Ibrahim. Amazon divisions have gone from hundreds of engineers to double digits. The "pyramids to columns" framing makes this sound intentional, like a strategy. In most cases it was a cost cut dressed up as an org design choice.
What actually happened is companies hollowed out the middle. They kept the seniors who could ship and let go of everyone whose value was harder to measure quarter-over-quarter. AI gave them cover. "We do not need as many people now that the tools are better." That sounds like a plan. It is not. It is what happens when you optimize for short-term output and forget that engineering organizations are pipelines, not just production lines.
The barbell sounds good in theory. Load the ends, seniors and juniors, skip the middle. In investing, that means high-risk and low-risk, nothing in between. In engineering teams, it means architects who can make system-level decisions and apprentices who are learning, with nobody in the layer that actually translates architecture into shipped code.
That layer is your mid-level engineers. They are the ones who take a technical direction and execute it across a codebase without hand-holding. They review pull requests. They onboard the juniors. They are the connective tissue between strategy and execution. Removing them does not create a lean team. It creates a team that looks fast on paper until a senior burns out or leaves and there is nobody ready to step up.
The AWS story everyone is citing
In March, a story circulated that AWS had a 13-hour outage caused by AI-generated code that was pushed without senior review. Whether the details are exact or exaggerated, the response was telling: multiple companies announced that junior and mid-level engineers can no longer push AI-assisted code without a senior signing off.
That is not a policy. That is a bottleneck. If your seniors have to review every line of AI-assisted code, and all your code is now AI-assisted, your seniors are not building. They are gatekeeping. And if your team has no mid-level engineers who can share that review load, you have just made your most expensive people your biggest constraint.
The barbell strategy assumes AI fills the middle. It does not. AI generates code. It does not exercise judgment about whether that code belongs in your system. Someone has to do that, and it cannot only be the three people at the top of your org chart.
What the right mix actually looks like
The question is not "juniors or seniors?" It is "what does each layer actually do now?"
Seniors set technical direction, make architecture decisions, and review the hardest problems. Their job changed less than people think. What changed is that they now spend more time reviewing AI-assisted output and less time writing boilerplate themselves. That is fine, unless reviewing AI output is all they do.
Mid-level engineers are the ones who should be scaling judgment across the team. They can review code, mentor juniors, and take a senior's design and implement it without needing every decision made for them. AI makes them faster, not unnecessary. A mid-level engineer with good AI habits is the highest-leverage hire most teams are not making right now.
Juniors are the pipeline. Not because you need them to write CRUD endpoints, AI does that now. But because in three years you will need mid-level engineers, and they have to come from somewhere. Ran Aroussi said it plainly: "Junior developers in 2026 should not be hired for their coding ability. They should be hired as the pipeline to becoming architects." The companies that froze junior hiring in 2023 are the ones struggling to fill mid-level roles today.
The right mix is not a shape. It is a function of what you are building, how fast your codebase is growing, and how much review overhead your AI-assisted workflow creates. A five-person team building a new product might genuinely work as a column: experienced engineers moving fast with AI tools. A 30-person engineering org maintaining a production system needs every layer, or the seniors drown in review work and the juniors never develop.
What this means for hiring
If your org is all seniors and AI, you have a fragile team. One departure creates a hole nobody can fill without a six-month search. If your org is all juniors and AI, you have a team that ships fast and breaks things nobody notices until production. The mix matters.
The harder question is how you evaluate for the right level. A mid-level engineer who can review AI-assisted code, catch model-introduced bugs, and make judgment calls about what belongs in your system is not the same as a mid-level engineer who was good at writing code from scratch in 2022. The skills shifted. Your screening process needs to reflect that.
At Fairground, the AI Coding Screener lets you see how candidates actually work with AI, whether they are reviewing output critically, validating edge cases, or accepting suggestions blind. That signal matters differently at each level. You are looking for architectural judgment in a senior, review discipline in a mid-level, and learning velocity in a junior. Same screener, different lens. Start free, 100 credits.
Get started with Fairground in just few mins.
Plug and Play. Works well with your existing ATS.
100 Free Credits
