Everyone passes the screen now

Ask a talent lead how their first technical round is performing and you will usually hear that pass rates look healthy. Ask what happens after that round and the story changes. More candidates clear the screen than a year ago. More candidates also get rejected at the onsite than a year ago.

Both numbers moved in the same direction. That is the part worth sitting with, because it means the screen is not finding better people. It is finding better prepared ones.

The screen was always a filter on preparation

A standard first round is a fixed list of questions. It has to be. Consistency is what makes it fair across dozens of candidates and what makes it defensible afterwards.

That fixed list was fine for a long time, because preparing for it took real effort. A candidate had to find the likely questions, work out good answers, and rehearse them until they were smooth. The effort of preparing correlated loosely with the ability to do the job, and loosely was good enough.

That correlation is gone. The questions for any given role are published, aggregated, and discussed in public. A candidate can now produce a well structured, technically accurate, confidently delivered answer to any known question in the time it takes to read it.

Preparation used to cost days. It now costs seconds.

So the round still works exactly as designed. It measures preparation. What changed is that preparation stopped being scarce, and a filter everyone clears is not a filter.

The cost does not disappear, it moves

A screening round exists to be the cheap stage. Its whole purpose is to make sure expensive people spend their time on a smaller and better set of candidates.

When the screen stops discriminating, that cost does not vanish. It moves downstream onto your engineers.

This is the part that rarely shows up on a hiring dashboard. The screen looks like it is working, because pass rates are strong and time to onsite is fast. Meanwhile a senior engineer spends the first fifteen minutes of every onsite discovering something the screen was built to catch. Multiply that across every open role and a broken filter quietly becomes the most expensive line in the process, paid in the time of the people you can least afford to interrupt.

What preparation still cannot fake

Rehearsed answers share a signature. They are complete and they are shallow.

A prepared candidate can describe a service architecture cleanly, name the right components, and use the right vocabulary in the right places. Ask what broke and the answer thins out. Ask what they tried that did not work, who disagreed with the approach, or which part they would build differently now, and there is nothing underneath, because none of that was in the material they prepared from.

Depth is specific to what a person actually did. It cannot be prepared in general, only recalled in particular. That property is the one worth building a hiring process around.

Two things still separate people

The first is a question that did not exist before the candidate answered.

If the follow up is generated from the candidate's own last answer, it cannot appear in any question bank, because it did not exist until that moment. Do that three times in a row and you are no longer testing recall. You are descending into one specific claim, and the person who lived it keeps going while the person who prepared it runs out of material. The competencies stay the same for everyone. The rubric stays the same for everyone. Only the surface questions change.

The second is work product. Not a description of what someone would build, but the thing itself, running. Give a candidate a live environment and a real brief, break something in it on purpose, and score what survives. A failing health check does not care how well the answer was worded.

Assume the AI is already in the room

Most responses to this problem try to remove AI from the room, through lockdown browsers, proctoring software, and monitored screens. It is an arms race, it treats every candidate as a suspect before they have said anything, and it aims at the wrong target.

The better target is a round where having AI open does not help you pass. Those rounds already exist. A generated answer does not survive three layers of follow up on one specific incident, because the specifics were never available to generate from. A generated plan does not survive an environment that actually runs it.

There is a fairness argument here too, and it is the one we find more convincing than the efficiency argument. A process that rewards preparation rewards whoever had the most time to prepare, the best coaching, and the strongest network telling them what to expect. That is not the same population as the people who can do the work. Testing what someone can actually do is not only a sharper filter. It is a more open one.

The script is what expired

The screening round is not obsolete. The script is. A first round that asks every candidate the same forty questions is measuring something that used to be a proxy for ability and is not one any more.

Recio is built on that premise. Every follow up comes from the candidate's last answer, every candidate builds in a live environment, and every score links back to the moment in the transcript that earned it. The mechanics are on the interview engine page.

The shorter version is this. Stop asking candidates what they know. Start finding out what they have done.