Engineering budgets for hiring are up 40% year-over-year. Recruiting tools proliferate. Yet most CTOs I talk to say the same thing: "We're seeing more candidates than ever, but fewer we'd actually hire." The math isn't mathing. More spend should mean better results. Instead, we're in a paradox where investment and quality move in opposite directions.

The AI flood broke the signal

In 2023, a Senior Backend Engineer role at a 50-person SaaS company got maybe 80 applications. In 2025, that same role gets 300+. Not because the market got better. Because AI made applying frictionless.

Candidates use ChatGPT to rewrite resumes for every job. They generate cover letters in seconds. They pass ATS filters they'd have failed two years ago. The result: your funnel is 4x larger, but the signal-to-noise ratio collapsed.

You're not getting better candidates. You're getting better-looking applications from the same candidate pool. And now you're paying for tools to process this flood—more ATS seats, more video screening platforms, more coding test licenses. None of which solve the core problem: you still can't tell who's actually good.

The assessment arms race is pricing out accuracy

Here's what the typical tech hiring stack looks like in 2025:

• ATS with AI resume screening: $8K–$15K/year

• Coding challenge platform: $10K–$25K/year

• Video interview tool: $5K–$12K/year

• Recruiting agency fees: 20–25% of first-year salary per hire

For a company hiring 10 engineers a year at $120K average salary, you're spending $50K on tools plus $240K in agency fees. That's $290K before you count internal time.

And yet: bad hires still happen constantly. I've seen teams spend six figures on this stack and still hire someone who can't deploy code in their first month. Why? Because every layer of this stack is optimized for filtering volume, not for showing you how someone actually works.

The tools measure the wrong things

LeetCode-style platforms test algorithmic recall. Video screeners test how well someone interviews. Take-home projects test who has 8 free hours on a weekend. None of these replicate the actual job.

When you hire a backend engineer, you need someone who can:

• Debug a production issue at 2pm with logs, metrics, and incomplete information

• Make tradeoff decisions between query optimization and code readability

• Explain why they chose Redis over Postgres for a specific use case

• Refactor legacy code without breaking everything

Standard assessments don't test any of this. They test whether someone can invert a binary tree under time pressure. That's not useless, but it's not the job either.

The time trap: you're spending more hours to hire worse

Let's be honest about the real cost. It's not the tool subscriptions. It's your engineering time.

Here's the actual hourly breakdown for a typical hire in 2025:

At a loaded cost of $100/hour (conservative for senior eng time), that's $2,700–$3,700 in internal cost per hire. For 10 hires, you're burning 300+ engineering hours per year just on interviewing.

And this assumes you're efficient. Most teams aren't. Most teams do 3–5 rounds per candidate, interview 15–20 people to make one hire, and still get it wrong 30% of the time.

The false optimization: faster funnels, same broken filter

The industry's answer to this has been: automate the top of the funnel harder. AI resume screeners. Async video interviews. Auto-graded coding tests.

This doesn't work. It just moves the problem.

You filter 300 applications down to 50 with automation. Great. Now you have 50 people who passed keyword matching and answered trivia questions. You still don't know:

• Can they debug a gnarly concurrency issue?

• Will they over-engineer a simple feature?

• Do they ask the right questions before writing code?

• How do they use AI tools—thoughtfully or as a crutch?

So you still interview 15 of them. You still burn 30 hours. You still hire someone who looked good on paper but ships slowly in practice.

The funnel got faster. The hiring decision didn't get better.

What actually changed in 2025

Two things made this worse:

1. AI made everyone look mid-level on paper

Junior devs can now generate code that looks senior. Resumes are polished to perfection. Interview answers are coached by AI. The traditional signals—resume quality, how someone talks about past work, even coding style—are all noisier than they were 24 months ago.

2. The talent market bifurcated

The best engineers aren't applying to 50 jobs. They're getting hired through networks or they're staying put. The people in your 300-application funnel are disproportionately:

• Junior devs trying to look senior

• Decent engineers who are bad at interviewing

• People mass-applying with AI tools

You're spending more to process a pool that got worse on average, while the top 10% of the market never enters your funnel.

The real solution nobody wants to hear

Stop trying to filter faster. Start watching people work.

The only way to know if someone is good is to watch them do the actual job. Not a proxy. Not a quiz. Not a theoretical system design. The job.

• Give them a bug in a real codebase and watch how they debug it

• Give them a slow API and watch how they optimize it

• Give them a deployment that's failing and watch how they troubleshoot

This takes 30 minutes. You see everything: how they think, how they use tools, how they explain tradeoffs, whether they ask about constraints, how they handle ambiguity.

It's faster than a 4-round interview loop. It's more accurate than a resume screen plus a LeetCode test. And it actually shows you the candidate doing the work you hired them to do.

The uncomfortable takeaway

You're not spending more because hiring got harder. You're spending more because you're using the wrong tools to solve the wrong problem. The problem isn't "how do I process 300 applications faster." The problem is "how do I know who's actually good before I waste 30 hours interviewing them."

More spend won't fix that. Better tools that measure the wrong thing won't fix that. The only thing that fixes it is changing what you measure. Stop filtering by keywords and coding trivia. Start filtering by watching people work.

Until you do, you'll keep spending more and hiring worse. The tools will keep getting more expensive. The candidate pool will keep getting noisier. And you'll keep burning engineering time on a process that doesn't work.