Leadership1 publisher2 min readPublished
Tighten's CEO read all 703 applications to see which ones a person wrote
Matt Stauffer put fun questions about the Oxford comma and favorite films into a project manager application, and the fully automated answers gave themselves away on tone. The reading took all 703.
The Board Room · Leadership desk
What happened
- Tighten, a remote programming company of around 15 full-time staff, received 703 applications for one project manager role, and chief executive Matt Stauffer read every one of them.
- The application posted on August 29 asked how candidates pronounce GIF, what their favorite movie is, what they think of the Oxford comma, and what shaped them as a project manager.
- Stauffer said the model appears to have settled on a short list of favorite films it recommends, naming The Pursuit of Happiness, Inception, The Secret Life of Pets and The Lion King.
- Oxford comma answers kept arriving with the same phrase, "reduces ambiguity in lists," and some applicants pasted the chatbot's framing text along with the answer itself.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- cost Tighten's screening ran at about 47 applications per full-time employee. An employer with a larger funnel has to pay for that depth of reading in senior hours, or automate the screening it built to catch automated answers.
- constraint The giveaways are a snapshot of what current models habitually produce. Once applicants learn which phrases flag them, the question set stops sorting anything and has to be rewritten each hiring round.
- exposure Stauffer's own standard was that a phrase gave a "really good chance" of an AI answer, not proof, so a candidate whose unaided writing matches the pattern bears the cost of a wrong call.
The favorite-movie question caught the automated answers because they got the register wrong. Stauffer said the model "has decided there are a few movies that everybody should list as their favorite" [7], and one application opened with "I really love Inception" before running 10 sentences on why project managers like films like Inception [8]. "It doesn't feel like the appropriate tone or communication style for a question about your favorite movie," he said [9].
Knowing the pattern did not spare him the reading. About 30 applications in, Stauffer recognized fully AI answers [6], which left 673 still to read after the pattern was familiar [3]. He said the company could not interview at that volume: "We can't talk to over 700 people" [15]. He also described the current market as one where "the hiring world is miserable, and a lot of people are filling out thousands of applications" [20].
He draws the line at output. "Everybody has a different relationship with AI, and we're open to people at varying levels of engagement, as long as your brain is still involved and the final output is good," Stauffer said [12]. He described applications carrying signs of AI where "clearly, there's a human involved, who prompted, reviewed, and edited their own humanity into it" [18]. His stated worry is commercial: "I just get concerned because if I can tell you used AI, clients can too" [13]. He uses the tools himself, mostly in writing and coding, as a reviewer and conversation partner [17].
Pronouncing GIF has nothing to do with running a project [5]. The reason to ask anyway is structural. Tighten is fully remote, most of its interaction happens over Zoom or in writing, and Stauffer wanted to find who could do those at a professional level [14]. On that spec, written communication is part of the work being bought, and an applicant who hands the writing to a model has not written the sample being judged.
The evidence here is narrow. It covers one hiring round, for one role, at a company of around 15 full-time people [3], read by a CEO who studied English, writes blogs and wrote a book [4]. The essay does not say how many of the 703 he rejected or whom Tighten hired [19]. It does establish that redesigned questions surfaced the fully automated applications early and that signs of AI were not, by themselves, disqualifying [16].
What to watch
- Whether Tighten reports what the personal questions predicted about the person it hired.
- Whether other employers turn published giveaways into screening rules, and how applicants misread as automated appeal them.
- Whether applicant tracking vendors start scoring for these patterns.