Your engineers write code with AI. Your interviews need to test for that reality, not pretend it doesn’t exist.

That’s the thinking behind HackerRank’s July release. It’s not a feature dump. It’s a shift in what “technical signal” means when every candidate has a copilot open in a second window.

Whiteboard problems and isolated coding challenges answer one question: can this person write an algorithm from memory. That question matters less every year. What matters now: can this person plan a solution, direct an AI assistant, evaluate what it produces, and ship something that works.

The July release builds toward that standard across three fronts: how candidates work with AI during assessments, how thoroughly you can verify what actually happened, and how directly your tools connect to the data.

Plan Mode is the clearest example. It’s now available in the AI Assistant across Screen and Interview as an AI add-on, and it changes what an assessment measures.

Instead of jumping straight to code, candidates use Plan Mode to explore the problem and draft an implementation approach first, with AI, before writing a line. You get visibility into the reasoning: how they broke down the problem, what tradeoffs they considered, whether their plan matches their eventual solution.

For engineering leaders, this is the difference between grading an output and grading a process. A candidate who plans well and executes with AI assistance looks a lot more like a senior engineer on your team than one who happens to remember a sorting algorithm.

Previous versions of AI Fluency scoring leaned on chat interactions alone. That missed most of the story. Engineers don’t just ask an AI assistant questions; they read suggested diffs, accept some, reject others, and course-correct constantly.

The updated evaluation pulls signal from IDE activity directly, capturing how candidates interact with AI and how they apply what it suggests as they work. Candidates with minimal AI interaction no longer get scored on fluency at all, since there isn’t enough data to make the grade meaningful. That’s a good default: a fluency score should reflect judgment under real usage, not a proxy built on too little information.

Paired with the new Diff View for front-end, back-end, and full-stack questions, you can now compare a candidate’s final project against its starting state, including every change the AI Assistant contributed. It’s the closest thing to reviewing a real pull request during an interview.

The Library expansion this release leans hard into realistic engineering work. New Plan-Build-Review repository assessments drop candidates into prebuilt, production-style codebases (Workflow, LinkUp, Shipway, Flagship) to resolve support-ticket-style issues: investigate, plan a fix with AI, implement it, then explain the approach on review.

That’s 159 new repository tasks spanning MERN, Spring Boot, Django, Go, and .NET, plus 159 project-based questions across .NET, Angular, React Native, React, Selenium, Spring Boot, and PySpark. Add 100 new coding questions, 11 framework-agnostic project questions, and 40 new AI Engineering questions covering Prompt Engineering and RAG, and the message is consistent: assessment content is catching up to what engineering work actually looks like in 2026.

Rust is also now supported in the execution environment, for teams evaluating systems-level or performance-critical roles.

Chakra, HackerRank’s AI interviewer, picked up an in-line code editor for coding questions this release. Candidates can now write, edit, and run code inside the interview itself while Chakra provides directional guidance, then probes their reasoning with follow-up questions once they submit, much like a human interviewer would.

That’s a meaningful step beyond conversational-only interviews. It means Chakra can now assess both what a candidate builds and how they explain their thinking, in a single session.

Operationally, Chakra also added full session video playback (screen, webcam, and transcript together), CSV export of scores and summaries, must-have requirement weighting for interview setup, and a Workday integration alongside programmatic API access to reports and candidate data. For teams running Chakra at scale, that’s less manual review and more direct pipeline into your ATS and dashboards.

More AI in the workflow means more surface area for gaming the system. This release tightens that up meaningfully:

• Desktop App enforcement in Chakra: locks the screen, restricts unauthorized applications, and flags anomalies in webcam feeds and code-writing patterns during interviews.
• Integrity enforcement: Chakra now pauses automatically if a candidate connects a second monitor, shares their screen, or exits full-screen, resuming only once conditions are restored.
• Improved proctoring accuracy: better object detection for phones and tablets (including partially visible devices), broader detection of suspicious applications including newer AI tools, and more reliable multi-face and absence detection across different lighting setups.
• Language-aware plagiarism thresholds: detection now accounts for natural code-length differences across programming languages, improving accuracy for concise submissions.
• Gaze Detection (limited availability): flags repeated look-aways followed by resumed typing, surfaced as a medium-severity signal so hiring teams can review rather than auto-flag.

None of this is about distrust by default. It’s about making sure the signal you’re grading is the candidate’s own judgment, not something happening off-screen.

Maybe the most forward-looking piece of this release: a read-only Model Context Protocol server, currently in limited availability. It lets AI models securely query supported HackerRank resources directly, so your team can build custom dashboards or generate hiring insights without exporting data into another tool first.

This is the same architecture powering agentic workflows across engineering orgs, now applied to your hiring pipeline.

Every one of these changes points the same direction: stop measuring whether a candidate can work without AI, and start measuring whether they can work well with it. Plan Mode shows their thinking. AI Fluency evaluation and Diff View show their judgment. Plan-Build-Review puts them in a codebase that looks like your codebase. Integrity controls make sure the signal is real.

That’s what hiring for the agentic era actually requires: assessments built for how engineers work today, not how they worked a decade ago.

The AI Add-on package includes HackerRank’s most advanced features for assessing next-gen developers and maintaining test integrity in an AI-native world. Contact your account manager or email support@hackerrank.com to enable it.