The Debrief 001: What We Fixed, and Who Broke It First
Three real product updates at Careersy AI, and the exact moment each one was built to fix. This is not a changelog.
5 min read

A machine scores your CV before a human reads a word. Then a headline tells you that using AI yourself is cheating. Pick one.
Here is the straight answer, from someone who spent 13 years on the hiring side of the ANZ tech market. Using AI in your job search isn't cheating. Using it to manufacture a version of you that falls apart the moment someone asks a follow-up question is. The real line isn't honest versus dishonest. It's translation versus invention. Use AI to make your actual experience legible to the systems and people deciding your career. Don't use it to invent experience you can't defend in the room.
That distinction is the whole game. Here is the difference at a glance.
| Translation (the right way) | Invention (backfires) | |
|---|---|---|
| What AI does | Turns your real experience into the language the system reads | Manufactures skills, results, or answers you don't have |
| What it looks like | A clearer, job-aligned resume; sharper true stories | Keyword-stuffed claims; AI-fed live interview answers |
| In the room | You defend every line, because it's true | The script runs out and the gap shows |
| The outcome | Read correctly, gets you shortlisted | Caught by a human, even when the software missed it |
No. And the people most likely to tell you it is are usually screening you with AI of their own.
Look at the asymmetry. Most large employers now run your application through software that ranks and filters before a recruiter spends a minute on it. That screening AI isn't neutral, either. A University of Washington audit of three million resume comparisons in 2024 found the model favoured white-associated names 85% of the time, and never once preferred a Black man's name over a white man's. According to a 2025 Greenhouse survey, 70% of hiring managers trust AI to make better hiring decisions, while only 8% of job seekers think AI makes hiring fair.
So the field is already thick with machines. One side automates the judging and calls it efficiency. The other side automates the applying and gets called a cheat. That isn't a moral rule. It's a double standard with a marketing budget.
Using AI isn't the problem. What you use it for is.
Not with software, and not reliably.
The detection tools barely work. OpenAI quietly retired its own AI-text detector in 2023 because it was inaccurate. A Stanford study that same year found seven popular detectors falsely flagged 61% of essays written by non-native English speakers as AI-generated. Vanderbilt University switched off Turnitin's AI detector rather than risk wrongly accusing its own students. If you write English as a second language, which describes a large share of the ANZ tech market, these tools are likelier to finger your genuine writing than to catch anyone gaming the system.
Here is the part most "beat the AI detector" advice gets wrong. The fact that software can't catch you doesn't mean nobody can. A recruiter can't prove you used AI. But they can spot two things every time: writing that is polished and says nothing, and a candidate who can't back up a single line of their own resume when you ask the second question.
The detector can't catch you. The conversation can.
Because the room came back, specifically to stop this.
Through 2025, Google, McKinsey, Amazon, Cisco and Deloitte's UK arm all reinstated in-person interview rounds, openly, to counter AI-assisted candidates. Google's CEO said the point was to confirm people actually know the fundamentals when there's no second screen to read from. Gartner found 72.4% of recruiting leaders now run interviews in person to fight fraud, and that candidates were more likely to apply when they did, not less.
Here is the part that stayed with me from years on the hiring side. The candidates who came undone were almost never the ones who used AI to prepare. They were the ones who used it to sound like someone else, and then had to be that someone, out loud, for an hour, with a straight face. You can hear the moment the script runs out. One recruiter put the tell plainly: they're parroting answers without real understanding.
Ask the student who built an AI tool to feed himself answers in live coding interviews. He landed offers from Amazon, Meta and TikTok, posted about it, got reported, and watched all of them get pulled. The fabricated version of you does not survive contact with a human who is paying attention.
Use it to translate what's real, not to invent what isn't. In practice, that's four moves.
Point it at the job description, not a blank page. Read the ad the way a recruiter does, and find where your real experience already matches the words the system scans for. "Incident response" and "being on call" can be the same work wearing different clothes. Use the employer's actual language for the skills you genuinely have. That isn't gaming the filter. It's refusing to be misread by it.
Make it sharpen your real stories, not write fake ones. Hand AI what you actually did and ask it to find the so-what: the result, the number, the thing that made the work matter. Most resumes die on vague. AI is good at asking "so what" until a bullet earns its place.
Pressure-test every line until you'd be glad to be asked about it twice. If a claim only survives as long as nobody checks, it was never yours to use. Run your resume and your answers back through AI playing the interviewer, and find where a sharp follow-up would expose a gap. Then close the gap with the truth, not with better wording.
Stop mass-applying. AI made it trivial to fire off a hundred applications, which is exactly why it stopped working. LinkedIn was seeing roughly 11,000 applications a minute in early 2025, up about 45% in a year. Nobody reads to the bottom of that pile. Six applications you could argue for out loud beat eighty you sent on hope.
There's evidence this is the stronger bet, not just the cleaner one. An NBER study of roughly half a million jobseekers found that people given algorithmic help to write clearer resumes were hired about 8% more often, with no drop in how employers rated their work. Clarity surfaces real ability. It doesn't manufacture it.
Know what you're actually owed here, because it's less than the internet implies.
Most of the candidate protections you read about, bias audits, opt-outs, a right to a human reviewer, come from New York City's hiring-AI law and the EU's AI Act. They don't apply in Australia or New Zealand. Australia's new rules requiring employers to disclose automated decisions don't even start until December 2026, and they cover disclosure, not a veto. You can't opt out of AI screening here. New Zealand's protections are softer again.
What is true locally is the flood. Robert Half found 65% of Australian hiring managers say the surge of AI-enhanced applications has made it harder to verify who can actually do the job. That cuts in your favour if you're the legible, defensible one in the pile. It cuts against you if you're just more of the noise.
It's the tool we built for exactly what this post is about, and yes, we have a stake in telling you it exists. CV Canvas is an interactive CV editor inside Careersy AI that sharpens how your experience reads without inventing a word of it. It's live now, and free to start.
It works one line at a time. You select a bullet, ask for a stronger version, and it comes back leading with your real result, with the few terms a recruiter scans for in bold. Every rewrite is grounded in three things and nothing else: your actual CV, the evidence you've given it, and the job description you're targeting. It will not add a number, a tool, a title, a team size, or a certification you didn't give it.
Here is the part that matters, and the part most tools get wrong. When a sharper line would need a fact it doesn't have, it doesn't fill the gap with something that sounds good. It stops and asks you.
Select a line, type "say I led a team of 8," and a typical AI resume tool writes "Led a team of 8 support specialists." Done, and false, and quietly waiting to fall apart in the interview. CV Canvas does the opposite. It asks:
Before I write that, I have to ask. I can see you built the escalation process the team uses and wrote the onboarding guide for new hires, but nothing in your CV mentions a title with "Lead" or a number of people. Did anyone actually report to you, and how many? Tell me the real number and I'll write it in. If no one did, your strongest true line is leading the escalation process the whole team relied on, and I can sharpen that instead.
Three real product updates at Careersy AI, and the exact moment each one was built to fix. This is not a changelog.
5 min read
Using AI more is not the same as being good at it. From the hiring side, here is the four-level ladder that separates the people who just use AI from the ones who own the outcome.
9 min read