91% of Engineers Now Use AI Coding Agents. Is Your Technical Hiring Still Testing Them the Old Way?
91% of Engineers Now Use AI Coding Agents. Is Your Technical Hiring Still Testing Them the Old Way?
CodeSignal's agentic coding assessments evaluate how candidates work WITH AI — and the data says your best engineers already do.
Three out of four engineers shipped production code that was partially or primarily AI-generated in the last six months. Yet most technical hiring processes still put candidates in a locked-down IDE, hand them an algorithm puzzle, and tell them to solve it alone.
That disconnect is exactly what CodeSignal set out to fix.
On April 2, 2026, the assessment platform — used by Netflix, Capital One, Meta, and Dropbox — launched what it calls the industry's first agentic coding assessments: a test format that lets candidates use AI tools like Claude Code, Cursor, and Codex during the evaluation, mirroring how software actually gets built today.
The Numbers That Forced the Shift
CodeSignal's internal survey of 450 U.S. engineers, conducted in March 2026, painted a stark picture of the AI-native engineering workforce:
- 91% of respondents use agentic AI coding tools at work.
- 75% shipped production code that was partially or primarily AI-generated in the past six months (as of March 2026).
- 73% believe engineers who don't adopt agentic AI tools risk becoming less competitive.
- 56% would hesitate to hire or work with engineers who are unfamiliar with agentic AI tools.
The message for talent acquisition leaders is blunt: if your technical screen bans the tools engineers use every day, you're measuring the wrong skills — and you're likely losing top candidates who see it as a red flag.
Note: This survey was conducted by CodeSignal among its user base, which may skew toward technically progressive organizations. The figures are directionally informative but should not be treated as representative of all engineering populations.
How Agentic Assessments Work
CodeSignal's agentic format follows a three-step structure designed to reveal not just coding ability, but engineering judgment in an AI-augmented environment:
Step 1 — Interpret requirements. The candidate receives a real-world problem and must decide how to approach it, what to prioritize, and what tradeoffs matter.
Step 2 — Build with AI. The candidate works alongside agentic AI tools — Claude Code, Cursor, or Codex — to develop the solution. This is not autocomplete. Candidates must prompt effectively, evaluate AI-generated code, and iterate on results.
Step 3 — Explain decisions. The candidate walks a human reviewer through their choices: why they accepted certain AI suggestions, rejected others, and how they validated the final output.
Hiring teams receive full transcripts of all AI interactions during the assessment, giving evaluators a window into the candidate's reasoning process — not just the final artifact.
"Engineers are no longer coding alone; they're working with AI agents," said Tigran Sloyan, Co-Founder and CEO of CodeSignal. "Our assessments now reflect that reality."
Validation at Scale
CodeSignal didn't launch agentic assessments cold. The company reports that tens of thousands of candidates completed AI-assisted assessments during 2025 — six months before the formal April 2026 launch — and roughly one-third of CodeSignal's customers adopted an AI-assisted format during that period.
That volume of real-world testing matters for HR leaders evaluating the format: it suggests the scoring models and assessment structures have been refined against actual candidate behavior, not theoretical frameworks alone.
Cosmo: AI-Powered Assessment Creation
On April 28, 2026, CodeSignal expanded its agentic capabilities with Cosmo, an AI agent that creates custom skills assessments on demand. Cosmo draws from a library of over 1,000 certified assessments and 4,000-plus questions — all validated by industrial-organizational psychologists — to generate role-specific evaluations in minutes.
For HR teams without dedicated I/O psychologists on staff, Cosmo removes a critical bottleneck. Instead of spending weeks designing and validating a custom assessment, hiring managers can describe the role and receive a psychometrically grounded evaluation immediately.
What This Means for TA Leaders
The shift to agentic assessments has three practical implications for talent acquisition strategy:
1. Rethink what you're measuring. Traditional algorithm tests measure memorization and isolated problem-solving. Agentic assessments measure prompt engineering, code review, systems thinking, and the ability to collaborate with AI — skills that directly predict on-the-job performance in modern engineering environments.
2. Expect candidates to demand it. With 56% of engineers hesitant to work with colleagues who can't use AI tools, candidates will increasingly view AI-free assessments as a signal that your engineering culture is behind. The assessment experience is now a talent brand touchpoint.
3. Move before the market does. CodeSignal's customer adoption data shows that a third of their client base already shifted to AI-assisted formats during 2025. Early movers gain an advantage in both signal quality (better hires) and candidate experience (lower drop-off rates). The window for competitive differentiation is closing.
The Bottom Line
The engineering workforce has already made its choice: AI agents are standard tools, not optional extras. Technical hiring assessments that ignore this reality are measuring a skill set that no longer reflects the job.
CodeSignal's agentic assessments offer one concrete path forward — and with named customers like Netflix, Capital One, Meta, and Dropbox already on the platform, TA leaders have relevant proof points to bring to their next hiring process review.
The question isn't whether to adopt AI-inclusive assessments. It's whether you can afford to wait.
FAQ
Q: What are agentic coding assessments?
A: Agentic coding assessments are technical evaluations that allow candidates to use AI coding tools (such as Claude Code, Cursor, and Codex) during the test, mirroring how engineers actually work in modern development environments. CodeSignal's format includes three steps: interpreting requirements, building with AI tools, and explaining decisions to a human reviewer.
Q: Why should HR teams move away from traditional coding tests?
A: A March 2026 survey of 450 U.S. engineers found that 91% use agentic AI coding tools at work and 75% have shipped AI-generated production code. Traditional algorithm-based tests that ban AI tools fail to measure the skills engineers actually use on the job, potentially screening out strong candidates.
Q: What is CodeSignal's Cosmo AI agent?
A: Cosmo is an AI agent launched on April 28, 2026 that creates custom skills assessments on demand, drawing from over 1,000 certified assessments and 4,000-plus questions validated by I/O psychologists. It allows hiring teams to generate role-specific evaluations in minutes without needing in-house assessment design expertise.
Sources
- PR Newswire, "CodeSignal Launches Industry-First Agentic Coding Assessments for AI-Era Engineering Hiring," April 2, 2026 — https://www.prnewswire.com/news-releases/codesignal-launches-industry-first-agentic-coding-assessments-for-ai-era-engineering-hiring-302732265.html
- PR Newswire, "CodeSignal Expands Agentic Capabilities Enabling Hiring Teams to Create Expert-Level Skills Assessments in Minutes," April 28, 2026 — https://www.prnewswire.com/news-releases/codesignal-expands-agentic-capabilities-enabling-hiring-teams-to-create-expert-level-skills-assessments-in-minutes-302755219.html
- CodeSignal Newsroom, April 28, 2026 — https://codesignal.com/newsroom/press-releases/codesignal-expands-agentic-capabilities-enabling-hiring-teams-to-create-expert-level-skills-assessments-in-minutes