Hiring the graduates who can actually ship

A short note to employers, parents, and the occasional confused recruiter.

We are now hiring out of the Moonlabs Academy at salaries that look senior on paper. The reason is not that we are overpaying juniors out of some sentimental attachment to our own graduates. The reason is much more straightforward: the work has changed, and the graduates who can do the new work are worth what we pay them. The market is in the process of figuring this out and recruiters are about eighteen months behind.

This essay is intended for the buyers — the heads of engineering and CTOs we get notes from, asking how to think about hiring AI-fluent graduates in 2026, when the salary bands their HR teams have are calibrated for a 2022 market.

What “junior engineer” used to mean

In 2022, hiring a junior engineer meant hiring someone with limited production experience who would, with mentoring, become useful in roughly twelve to eighteen months. The expected output in their first year was modest. They were a long-term bet, paid accordingly: £35,000 to £45,000 in the UK regional market, £50,000 to £60,000 in London.

The job they were going to grow into looked like this: writing code in a familiar stack, getting reviewed by a senior, being assigned to features that the team had already scoped. The graduate’s value was largely in their trajectory. They were going to be senior in three to five years and the firm wanted to lock in the relationship early.

That model is broken in two specific places.

First, the productivity ratio between a junior and a senior has compressed dramatically. A senior engineer with the modern AI toolchain is roughly 2-3× more productive than they were in 2022. A graduate who is fluent in those same tools — and this is the operative phrase — is roughly 5-7× more productive than the 2022 baseline graduate. The senior pulled away, but the well-trained junior pulled away much faster, because the ceiling on their output was always their typing speed and the ceiling has lifted.

Second, the judgment gap between junior and senior has widened in the absolute, even as the productivity gap has narrowed. The new failure modes — model hallucinations that look plausible, AI-generated code that compiles but does the wrong thing, evals that test the wrong axis — require taste to catch. That taste does not yet come from the model. It comes from the human. Graduates who have been trained to develop this taste early, by reviewing AI output in a structured way from the first week of their professional life, develop it faster than the cohorts who learned it incidentally over five years of debugging legacy code.

The result is that the modern junior is no longer the modest first-year bet they used to be. The right modern junior can take on work that, in 2022, would have been comfortably senior. We pay accordingly because the alternative is paying a senior twice as much for the same work and waiting two months for their notice period.

What we pay, and why

We hire out of the Academy at a starting salary in the £55,000 to £75,000 range, depending on prior work and how the demo day went, with a profit-share component on the small portfolio of products they touch. Some graduates with stronger commercial signals start higher.

The numbers shock recruiters. They should not. Run the maths.

A 2022 senior engineer was paid roughly £85,000 to £110,000 in the UK regional market for output of, conservatively, 1× a 2022 baseline. A 2026 Academy graduate, with the modern toolchain and the structured commercial training, produces output of roughly 4-5× a 2022 baseline on engineering work, plus they can run a discovery call, plus they can sit in front of an investor.

We pay them roughly two-thirds of what a 2022 senior used to cost, for roughly four times the unit output. The cost per unit of useful work is therefore roughly six times better. We would, in principle, pay considerably more than we do. We do not, because the wider market is not there yet and we would distort the comp band for the rest of the team. But the marginal economics are extraordinary.

The honest version of the story is that we are taking advantage of a market mispricing while it lasts. We expect that, within two or three years, the going rate for a graduate of a properly-run AI-native operator course will be in the £80,000-£120,000 range, and the firms that are still trying to pay £40,000 will find they cannot hire anyone useful.

How to spot the graduates who are worth this

The hardest thing for an employer who is not yet calibrated on the new market is the signal problem. The CV looks like a junior CV. The age is twenty-two. The instinct is to pay the junior rate. How do you tell, in the interview process, which graduates have actually been trained for the new work and which have been trained for the 2022 job?

Three signals, in our experience, separate the cohorts cleanly.

Signal one: can they explain the code they did not write themselves. Sit them in front of a piece of code they shipped recently and ask them to walk you through it line by line. The graduates who have been trained well will do this fluently. The ones who shipped through an agent and did not internalise it will stumble within three minutes. This is a sharper test than asking them to write code in the interview, because what you are actually selecting for is reading and judging AI output, which is the modern skill.

Signal two: can they price. Ask them what the company they built in their course would charge for, and how they would justify the price to a buyer. The 2022-trained graduate will say "I don’t know" or quote whatever the closest competitor charges. The 2026-trained graduate will give you a reasoned answer that connects the buyer’s alternative cost of the pain to the price of the product. The depth of that answer is a near-perfect predictor of their ability to operate commercially inside your company.

Signal three: can they take feedback in public without freezing. Pitch them the work, in front of two of your team. Then push back on it in the room. The graduate who can update their view in real time, defend the things worth defending, and concede the things worth conceding, is the graduate who will be a credible operator inside your company. The one who freezes or argues defensively will be a 2022 graduate however they were trained.

These three signals can be tested in a single 90-minute structured interview. They are dramatically better predictors than the traditional algorithmic coding test, which selects for skills that the model now performs reliably.

Why we are writing this publicly

Two reasons.

The first is genuine evangelism. We meet a lot of CTOs at events and conferences who tell us, quietly, that they cannot find good juniors. They can. They are looking for them with 2022 filters. Adjust the filters and the candidates are there, in our cohort and others.

The second is selfish in a useful way. The graduates of the Academy are, for the next year or two, dramatically underpriced in the market. We hire some of them ourselves and the rest we recommend to the companies we know. The faster the broader market figures this out, the easier our subsequent cohorts will be to recruit, because the parents of prospective students will see that the salary outcomes are real. It is in our interest to make the market efficient.

If you are an employer in the UK and you are thinking about how to recruit graduate operators in the new shape, please do come to demo day. We hold it at the end of each cohort. Watch the students pitch. Talk to them at the drinks. Pay them what they are worth. The wider market will catch up; you do not have to wait for it.

The Moonlabs Academy is a twelve-week operator-led course in Derby. Demo day is held in front of an invited audience of investors, operators, and prospective employers. If you would like to attend the next one as a hiring observer, the team can be reached through the apply form.