AI Fluency, Assessments, and Rena

The assessment question for 2026 is no longer "did the candidate use AI?" It is "did they use AI well?" This product cycle moved AlgoArena closer to answering that question directly.

What changed

We shaped the assessment workflow around explicit AI assistance modes. A candidate can work without assistance, use guided help from Rena, or work in a more agentic mode where the product captures more of the conversation around planning, editing, and verification.

Rena also became a clearer product character: not a generic chatbot, but a coach and reviewer that can help expose how someone thinks. The important part is not that Rena gives an answer. It is that the candidate's choices around asking, accepting, revising, and validating become part of the evidence.

Why it matters

Blanket AI bans create a strange kind of theater. They make assessments look controlled while hiding the actual skill modern teams need: judgment with powerful tools in the loop.

The better path is to measure the interaction. Did the candidate ask precise questions? Did they test the suggestion? Did they catch the edge case? Did they understand the code they shipped? That is the signal we want [Assessments](/product/assessments) to surface.

Where it points

This work connects directly to the [Vibecoding product lane](/product/vibecoding) and the broader argument in [The Problem Was Never Vibecoding](/blog/the-problem-was-never-vibecoding). Vibecoding is not a shortcut around skill. Done well, it is a new surface where skill shows up.