framework · strategy
PM interview frameworks in 2026: when CIRCLES helps and when it signals No Hire
Best for: Deciding when to use a framework vs. when structured recitation kills your candidacy
CIRCLES is not a red flag because it is wrong. It is a red flag because it signals that you prepared a template rather than developed a judgment. Every strong answer has structure. The question is whether your structure is visible scaffolding narrated out loud, or internal discipline that produces a clear, reasoned answer. In 2026, AI pre-screening tools flag CIRCLES-pattern responses as low-signal, and a new class of AI-round interviews makes step-walking structurally impossible. The cost of template-thinking has gone up.
Why CIRCLES breaks for AI products
CIRCLES was designed by Lewis C. Lin around 2013 for deterministic product design questions: “design a fridge feature,” “improve Gmail.” Those assume a fixed system where the PM’s job is to specify what engineers build. AI products are probabilistic: model behavior varies, failure modes are non-obvious, and the PM’s job includes specifying what “good” even means before proposing a feature.
Paweł Huryn’s argument (his piece is titled “How to Ace the AI Product Sense Interview Without Using CIRCLES”) is structural: CIRCLES steps assume you can enumerate solutions and evaluate them against known tradeoffs. In an AI product question, you cannot evaluate a solution without first specifying an eval. Aakash Gupta made the same point more bluntly: “You can’t CIRCLES your way through ‘how would you increase Claude Code WAU 10x?’” That question is not a design question. It is a strategy question. CIRCLES burns interview time on user comprehension that does not resolve the strategic judgment being probed.
The specific mismatch: CIRCLES S-step (List solutions) incentivizes a brainstorm of five ideas. In a serious AI product sense question, the interviewer is watching for whether you can name which cognitive task the AI is performing, what failure looks like for real users, and what you would measure before shipping. A list of ideas without that grounding reads as surface-level even if the ideas are individually reasonable.
There is also a structural shift underneath all of this. When feasibility is effectively free (AI collapses the cost of building almost anything), the bottleneck moves to viable: is this a pain people will pay to solve, in a market large enough to sustain the business? And to lovable: does it meet people where they actually are, anticipating their needs without being obnoxious? CIRCLES C-I-R steps spend most of their time on the question that is now cheap to answer, and skip past the two that matter.
The decision tree
Use structured frameworks openly when the interviewer is a recruiter (not a PM), the screen is a 20-minute call where covering the bases matters, or the question is an estimation where MECE decomposition is the point.
Stop narrating the framework when the question is strategic, involves AI products or agents, you are past the recruiter screen, the interviewer has asked a follow-up that should redirect you, or you are at Meta, Anthropic, or OpenAI (where the process is specifically designed to break formulaic answering).
Meta introduced a live “Product Sense with AI” round where candidates prototype with AI tools in real time. There is no format for CIRCLES step-walking there. Exponent’s 2026 product sense guide flags a specific No Hire signal in this context: saying “we could use AI to personalize it” without specifying what the AI is doing, what data it needs, and what failure looks like. That is exactly the output CIRCLES S-step produces when applied reflexively.
What a strong answer does instead
strong
"Before I suggest directions: is the constraint here engagement, retention, or monetization? That changes everything about what I'd propose. [Confirmed: existing user engagement.] I want to focus on a specific moment: the user who set up the product, used it twice, and then quietly stopped. That person did not churn visibly, they just drifted. The job they hired the product for probably did not get done in the first two sessions. The viability question is whether solving that re-engagement is worth building: is this pain acute enough to pull them back, or have they already moved to a competitor? If they are still in the product's orbit, I would propose one thing: a contextual moment that meets them where the job actually surfaces, in their calendar or email, rather than asking them to return to the product unprompted. The lovable version anticipates that moment. Success metric: reactivation rate for this cohort in 30 days, against a holdout."
weak
"Let me walk through CIRCLES. First: comprehend the question. You are asking how to improve the product. Then I'll identify the customers: I will pick two segments, casual users and power users. For casual users the need is ease of use; for power users, speed and customization. Now I'll list five solutions: simplified onboarding, keyboard shortcuts, AI-powered suggestions, a dashboard, better notifications. Evaluating: AI-powered suggestions would be the highest impact because personalization drives retention..." The problem: the segmentation is generic and connects to no tradeoff decision. The five ideas produce no reasoned bet. The AI step names no cognitive task, no failure mode, no eval. And in a live prototyping round, there is no time for this walk at all.
The meta-irony
AI pre-screening tools used by some companies now flag CIRCLES-pattern responses as low-signal because the pattern is over-indexed in candidate prep materials. The same structure designed to help you look prepared now marks you as someone who memorized a script. Anthropic’s process includes questions like “who do you respect but disagree with on values?” specifically because those have no CIRCLES equivalent. The prep that works there is real opinions, developed through real engagement with the problem space.
The practical implication: use frameworks to organize your thinking before you speak. Stop using them as the thing you say.
For the full CIRCLES reference, see CIRCLES framework. For what changes when the product involves AI, see how AI changed PM interviews and feasibility is free.