Build1 publisher2 min readPublished
Applied Intuition splits its engineering interview between a no-AI screen and a two-hour AI build
Applied Intuition screens engineers with a 45-minute no-AI coding task, then gives onsite candidates two hours to build a design with any AI model. Other teams can copy the split now, though its only published evidence is interviewer confidence from the company's own trial.
The Engineer · Build desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Applied Intuition says it kept the no-AI screen because engineers remain responsible for every line in the codebase and must understand code without an assistant.
- The company rebuilt the onsite after retooling its organization around AI coding agents, saying its old all-no-AI process no longer matched how it works.
- The onsite's 45-minute system design question is usually derived from a problem the company has actually had to solve.
- In a trial, interviewers rated their confidence in seven hiring signals from 1 to 5, and six of the seven scored between 4.0 and 4.7.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision A team copying the split has to give code comprehension its own stage with no assistant present, or it loses that signal once AI is allowed in every round.
- cost Each candidate puts in about 210 minutes of timed technical work before the demo even starts, so adopting the template means booking a long onsite day.
- capability Interviewers can be trained quickly when the exercise mirrors their daily work, but that advantage holds only where engineers already build with AI agents every day.
- decision Waymo forbids AI in every live interview, so candidates arrive with opposite expectations and a split loop has to state its AI rule for each stage in advance.
The detail I would copy first is the order of the onsite steps. The candidate settles a design in 45 minutes. Then they get two hours to build it with any model and any frameworks. Then they walk a panel of engineers through what they built, the decisions they made and what they would do differently [2]. A candidate who let a model make every choice in the build still has to explain those choices out loud to people who build systems for a living [2].
Software engineer Kesav Viswanadha wrote the post that set out the format, on the company's engineering blog in July 2026 [3]. The details here come through Four-Leaf, which publishes interview prep advice on dev.to [16]. On the first stage, Viswanadha wrote: "Keep our technical phone screen as is" [14]. His case for the rebuilt onsite is that coding is getting easier, so decision-making, technical excellence and cultural factors count for more. AI, in the post's words, is "just the medium to enable it" [9]. Both stages cover engineering hiring only [15].
I think calibration is the best argument for keeping the two stages apart. The company says the format keeps changing because every model release shifts what a candidate can do in two hours [7]. It is also still scaling the format across more than 1,000 engineers [7]. A no-AI coding task measures the same skill after a model release as before it [1]. The two-hour build moves with each release. A prototype scope that separated candidates in July can be routine after the next model ships, so the design questions and the passing bar have to be re-tuned whenever the tools change [7].
The trial results need the same scrutiny as any benchmark table. They record how confident interviewers felt about each signal [10]. The named signals include practical skills, tackling unfamiliar challenges, technical communication, logical thinking and technical mastery [11]. For those scores to carry over to another company, two things would have to be true. Interviewer confidence would have to predict how hires actually perform on the job. And the other team's design problems would have to come from its own real work. Applied Intuition does not say which part of the onsite measures which signal [12]. A team copying the format therefore has to decide for itself whether technical communication gets scored during the build or during the demo.
What to watch
- Whether Applied Intuition publishes data linking its trial confidence scores to how the people it hired actually performed.
- How the two-hour build scope and passing bar change as new models ship while the format rolls out across 1,000-plus engineers.
- Whether other employers publish per-stage AI rules like Applied Intuition's or follow Waymo in banning AI from every live interview.