AI Didn’t Shrink the Org Chart. It Moved the Bottleneck

Forward deploy engineers, senior-heavy teams, and the junior math nobody’s run yet

I talk to founders and technical leaders across our client base every week, and right now they are running two completely different playbooks.

Some are just getting started with AI. Others have already cut their engineering teams down while getting more output than before, some as much as 10x on certain workflows. On paper it looks like the same story: AI is replacing headcount. Underneath, it’s a more specific story, and it’s the one I think matters for how you hire in the next 12 months.

AI collapsed the cost of building something. It did not touch the cost of knowing whether what got built is actually sound. That second thing, judgment, is where the real hiring decisions are happening right now. Here’s what I mean, based on four patterns I’m seeing repeat across very different companies.

Forward deploy engineers are back in demand

We’re seeing a real uptick in requests for forward deployed engineers. This isn’t nostalgia for a role that never really left. It’s a direct response to what AI can’t do: sit inside a specific client’s messy, half-documented environment and figure out what actually needs to be built there. A model can write excellent code in the abstract. It can’t yet walk into a client’s stack and know which of their five conflicting systems is the source of truth. That’s judgment applied to a specific mess, and it’s exactly the kind of work that’s getting more valuable, not less.

The staff/principal-only strategy

Some companies have made a clean bet: stop hiring mid-level, hire only Staff and Principal level people, and let them direct AI to do the rest. The logic is sound. If judgment is the scarce resource, concentrate it in fewer, more senior hands and let AI handle execution underneath them. Where I’d push back a little as an advisor is that this only works if those senior hires actually have the bandwidth to review what AI produces at the volume it’s now being produced. I’m watching a few clients learn that lesson the hard way, where the bottleneck just moved from “not enough engineers” to “not enough senior reviewers.”

The instinct here is usually to hire even more seniors to close that gap. I don’t think that’s the right fix, and this is coming from a recruiter. Senior engineers are already the scarcest, most expensive part of the market, so trying to out-hire your way past a review bottleneck just runs you straight into the same scarcity problem you started with. The real fix is making senior review scale differently, not making more of it. That looks like staged review, reserving real senior attention for the architecturally risky parts and letting lighter checks cover the rest, building more judgment directly into the harness so less output needs a human pass at all, and letting mid-level people do a first pass with seniors spot-checking rather than routing everything straight to a Staff or Principal. If you’re betting on staff/principal-only, the binding constraint isn’t headcount. It’s review throughput, and that gets solved with process and tooling, not by trying to hire your way past a talent pool that’s already this thin.

The junior-plus-harness math is getting more complicated

The opposite bet is happening too. Some companies let juniors run with AI once a solid harness and guardrails are in place, and it works, to a point. But token costs add up fast, and companies are about to start doing the real math: tokens plus junior salary plus review time, versus one experienced hire who needs less of all three.

A director I spoke with, at a company running fully AI-first, is hungry for juniors right now, cutting against the staff/principal-only trend. Their process is spec-driven: product writes the PRD, it becomes a spec, agents build against it, and engineers review for architecture soundness instead of writing code themselves, in trios of a designer, an engineer, and a product person. His view: models are good enough now that you don’t need to already be a great engineer, just trainable into good judgment inside a strong harness.

Even there, the gap shows up. A junior on his team ran up an enormous credit bill building one website, because no one was overseeing him closely enough. The harness existed. The oversight didn’t, and it hit the bill directly.

I’ve seen this pushed further. One client told me you don’t need to be a developer at all anymore, they planned to hire anyone and have them build AI agents directly in the codebase, trusting the harness to carry all the judgment a developer used to need. A junior with real training can learn to spot when something’s off inside a good harness. Someone with no development background has no baseline for that judgment at all.

Neither extreme is right. Hire the juniors, and actually oversee them while they learn inside the harness, instead of assuming the harness alone will catch what a person would have.

The through-line

Every one of these patterns is a company deciding, consciously or not, where judgment sits in their org. Forward deploy engineers put it in the field. Staff/Principal-only hiring puts it at the top and hopes it scales. Junior-plus-harness puts it in the guardrails and the token budget, which only works if someone’s actually watching the budget too.

If you’re hiring right now, the question worth asking isn’t only “where can AI replace a role.” It’s “where does this org actually need a human making a judgment call, and do we have the right person in that seat?” That’s a different hiring brief from the one most companies are still writing, and it’s the one I’d want to build my team around.

This is the first in a series I’ll be writing on how AI is actually changing hiring and org structure, not the version people talk about at conferences.