RANT Why the UK Tech Job Market Is This Broken Right Now
Every job spec wants five years of experience with a framework that's two years old. Nobody's said the quiet part, which is that half of them aren't hiring at all.
I spent nineteen minutes talking an interviewer through Floyd’s cycle detection algorithm last month, out loud, on a shared screen, for a role that as far as I could tell from the job spec would never once require detecting a cycle in anything, and somewhere around minute six they stopped taking notes and started nodding instead, which is the tell that the pace wasn’t what they were after. Four companies, four near-identical versions of the same test, and none of the actual jobs on offer have needed it yet.
The UK job market genuinely shrank in early 2025
It would be easy to write this off as one bad run of interviews, except the numbers for early 2025 back up the bad mood. The Office for National Statistics put UK unemployment at 4.4% for the three months to February, above where it stood a year earlier, with the number of people on payrolls falling by 21,000 over the same quarter. That’s a national figure, not a tech-specific one, but it’s the same direction every recruiter I’ve spoken to is describing as “competitive,” in a tone that means something worse.
Meta cut roughly 3,600 people in January, about 5% of its workforce, specifically targeting people it classed as low performers, and several of those let go said afterwards they’d been rated at or above expectations in their last review. Whatever process rates someone “meets expectations” one quarter and cuts them as a low performer the next isn’t really measuring performance, and the dressing-up annoys me more than the layoffs themselves.
Every hiring freeze now comes with an AI excuse attached
Salesforce’s Marc Benioff announced in February that the company wouldn’t be hiring any engineers this year, because Agentforce, the company’s own AI product, was apparently doing the job instead. He’s entitled to make that claim. A company whose product depends on convincing the market its engineers are replaceable is not a neutral source on whether that product actually works, in much the same way a chip manufacturer’s own benchmark slides aren’t a neutral source on that chip’s performance, a point I’ve made before about the general vendor confidence around AI I’ve already lost patience with, and nobody’s published the productivity numbers behind this particular decision for anyone else to check.
The interview format everyone hates hasn’t gone anywhere
This isn’t a new complaint about the whiteboard-algorithm interview. Ten years hasn’t fixed it regardless. Max Howell, who wrote Homebrew, the package manager a large share of Google’s own engineers have installed on their laptops, was turned down by Google in 2015 for failing to invert a binary tree on a whiteboard, and said so publicly in terms considerably blunter than I’m using here. A decade on, the test he failed is still the industry default, and the tool he wrote is still on more machines than most of the people administering that test could tell you.
What I’d actually want tested, if anyone asked me, is whether a candidate can read someone else’s four-hundred-line pull request and explain in plain terms why it’s slow, because that’s the job most days, and no interviewer has asked for that instead of a reversed linked list yet. I’ve started keeping a running note of which companies favour which flavour of tree traversal, purely so I stop relearning the same three techniques from scratch every time a recruiter emails back.
I’ve got three more of these lined up before the end of the month, one of which has already sent over a take-home exercise on top of the two live rounds. That either means they genuinely can’t tell a good engineer from a bad one without four separate tests, or nobody there has worked out how much unpaid time they’re asking a candidate to hand over. Either explanation says more about the process than about whoever’s sitting the test.