The "I don’t do interviews" approach replaces traditional interviews with real work — paid trial projects, actual tasks, or work samples — to evaluate candidates. It exists because interviews primarily measure interviewing ability (composure, polish, rehearsed answers), which correlates weakly with job performance, while watching someone do representative work measures the thing you actually care about.
The source challenges interview theatre because polished answers can be a weak proxy for job performance. A representative, paid, and bounded work sample gives the hiring team something closer to the work it actually needs to evaluate. Zero-to-one recruiting adds the founder context: early hiring sets the standard for how evidence and judgment will be handled later.
Why can trial work beat an interview?
An interview is a performance in a particular social setting. It can reveal communication, curiosity, and how someone thinks aloud, but it also rewards preparation, confidence, familiarity with the format, and the ability to tell a compelling story. Those are not always the capabilities the role needs most.
A real work sample changes the signal. The person has to interpret a brief, choose a method, manage ambiguity, produce an artifact, and explain the trade-offs. The task should be paid, bounded, representative, and accessible. The ABC pitch framework is a useful scoring discipline because the reviewer can separate the claim, the evidence in the work, and the next action rather than grading polish.
| Signal | Interview | Work sample |
|---|---|---|
| What is observed | Conversation under pressure | Representative task and process |
| Main confound | Polish and familiarity | Task design and unequal access |
| Useful for | Communication and questions | Capability and judgment |
| Fairness need | Consistent questions | Paid, scoped, accessible task |
| Decision quality | One perspective | Work plus human context |
What makes a work sample fair?
Fairness begins before the candidate sees the task. The employer should define the capability being tested, the time boundary, the tools permitted, the data or scenario provided, how the work will be evaluated, and whether the output will be used commercially. A task that secretly extracts free client work is not an assessment; it is unpaid production.
The reviewer should also allow reasonable accommodations and recognize that a sample is evidence, not a full identity. A person can have a poor day, misunderstand an ambiguous brief, or demonstrate strength the task never asked for. Behavioral pod synergy is relevant because the hire will work in a team, not in an isolated test chamber.
| Element | Define before the task | Human check |
|---|---|---|
| Capability | What role behavior is being tested? | Does the task represent the role? |
| Scope | How long, with what tools and inputs? | Is the burden reasonable and paid? |
| Rubric | What does good work show? | Is the rubric consistent and accessible? |
| Use | Will the output be used or retained? | Was consent and compensation clear? |
| Context | What can the conversation add? | Who reviews the whole person fairly? |
How can AI help design and review a trial?
AI can turn a role outcome into a candidate-facing brief, check that the rubric matches the task, generate equivalent variants, identify ambiguous instructions, and summarize the evidence against the approved criteria. It can also help compare reviewer notes so the panel discusses the work rather than the candidate’s writing style or similarity to a favored person.
The risks are substantial. A model may reproduce bias in the rubric, over-weight surface style, expose candidate data, or create a false impression of objectivity. A human hiring owner validates the task, speaks with the candidate, checks accessibility and permission, and makes the decision. Agentic workflow automation can coordinate the review; it cannot make the hiring decision.
| Stage | AI contribution | Human control |
|---|---|---|
| Observe | Collect the approved role outcomes, task brief, work sample, candidate context, and reviewer notes. | Confirm consent, compensation, accessibility, scope, and who may see the material. |
| Interpret | Map the work to the rubric and flag missing evidence or reviewer inconsistency. | Review the artifact and context; challenge the model’s framing. |
| Act | Prepare interview follow-ups, a decision brief, or a request for another bounded sample. | Choose the candidate or close the process with human accountability. |
| Review | Compare the hire’s later work and feedback with the original role signal. | Improve the task without claiming that one sample predicts everything. |
PPC Snobs in practice: evidence needs a human owner
Our own Landers and technical workflows use structured briefs, acceptance criteria, source paths, and review gates because evidence should survive handoff. A hiring sample can use the same discipline: clear scope, explicit rubric, visible uncertainty, and an owner who can explain why the evidence mattered.
AI can help create the brief, compare the output with the rubric, and route the open questions. It should not be allowed to silently decide that one candidate is more “culture fit” or to turn a writing pattern into a character conclusion. The library model is a useful reminder to preserve only the approved context, not every private note a system can collect.
- Make the trial paid, bounded, representative, accessible, and explicit about use.
- Define the capability and rubric before seeing the candidate’s output.
- Use AI to structure evidence and questions; require a human hiring decision.
- Review the later job signal without pretending one sample is a guarantee.
Where AI stops
AI may draft a task, check rubric consistency, summarize work evidence, and prepare follow-ups. It must not infer protected traits, rank candidates as a final decision, use unpaid production work, reject a person automatically, or replace the accountable hiring owner.
Should interviews disappear completely?
No. Conversation remains useful for communication, expectations, motivation, collaboration, and questions the work sample cannot answer. The better change is to stop treating interview polish as the main proof of competence. A short conversation around a real artifact can reveal more about reasoning than a long sequence of hypothetical questions.
The process should also be proportionate. A small role may need one representative exercise; a high-risk role may need multiple evidence sources and a deeper conversation. Evidence-led hiring is not a demand for more hoops. It is a demand that each step earn its place and respect the candidate’s time.
| Step | Best signal | Keep in mind |
|---|---|---|
| Conversation | Communication and mutual expectations | Do not confuse polish with capability |
| Work sample | Role-relevant execution and judgment | Pay, scope, and accessibility matter |
| Reference or history | Prior context with permission | Check relevance and recency |
| Decision review | Evidence and risk discussion | A human remains accountable |
Design a work-sample hiring system
Connect role outcomes, paid trials, rubrics, candidate consent, AI assistance, and human judgment without pretending one test predicts a whole person.
Questions the operator should be able to answer
What does “I don’t do interviews” actually mean?
Replacing traditional interviews with real work — paid trial projects, representative tasks, or work samples — as the primary way to evaluate candidates. You watch someone do the job rather than hear them describe it.
Why are interviews unreliable?
Because they mainly measure interviewing skill — composure, rehearsed answers, charisma — which correlates weakly with job performance. Great interviewers can underperform and excellent workers can interview poorly, so the signal barely overlaps with the role.
How do I evaluate with real work?
Give candidates a paid, scoped, representative task they’d actually do on the job, then judge the output and process against what the role requires. Paying for their time is respectful and elicits serious effort, and one real task reveals more than an hour of conversation.
Should I stop talking to candidates entirely?
No — conversation still matters for fit, communication, and values. The point is not to let interview performance stand in for job performance. Use conversation for collaboration and alignment, and real work to assess actual competence.
Editorial source: the PPC Snobs resource library and editorial review of September 8, 2026. Evidence and proposed workflows are identified below.
Editorial method: source-grounded answers, clear authorship, visible evidence qualifications, contextual resources, and structured data that matches the article.
Evidence lane: observed / source-grounded hiring principle; proposed AI-assisted work-sample design and review. PPC Snobs favors evidence-led operating decisions and uses structured review artifacts in our own work. This article proposes a hiring workflow; it does not claim a completed trial program, predictive validity result, or legal sufficiency.
Route the decision to the capability that owns the evidence.