Canva Says "Insist," Google Says "In Person." Rewrite These 4 Bullets.

The interview market split in half this year, and a resume cannot straddle it. Canva, Meta, and Datadog have published explicit "AI-allowed" or "AI-required" interview policies in 2025; Google, McKinsey, Cisco, and Deloitte have pulled rounds back in-person to stop cheating. The bullets that win at Canva are the exact bullets that read as red flags at Google, which means the question is no longer "should I mention AI tooling," it's "which camp am I applying to this week."

The split that actually matters now

Two camps have formed, and the dividing line is employer culture, not role or seniority. Canva, Meta, Datadog, and Rippling have told candidates to bring Copilot, Cursor, or Claude into the room. Google, McKinsey, Deloitte UK's graduate program, Cisco, and Anthropic have reinstated in-person rounds or explicitly banned assistants.

The dates are tight:

• June 11, 2025: Canva's Simon Newton publishes "Yes, You Can Use AI in Our Interviews. In fact, we insist," covering backend, ML, and frontend candidates.
• July to August 2025: 404 Media and WIRED report Meta is piloting candidate AI use in coding tests. Meta's own line: "we're testing how to provide these tools to applicants during interviews."
• October 2025: Meta's roughly 60-minute AI-enabled coding round goes live for SWE and M1 (E4/E5) candidates, replacing one of two traditional onsite coding rounds.
• October 20, 2025: Canva's Llew Vallis publishes "AI Interview Success: An Interviewer's Inside Guide," describing what stronger candidates actually do when AI is in the room.
• February and June 2025: Sundar Pichai signals at a town hall and later on Lex Fridman's podcast that Google is bringing back in-person rounds, after the CNBC story about Roy Lee's Interview Coder tool went public.

Datadog threaded the needle. Its official "Interviewing at Datadog: AI Guidelines" page splits AI use by stage: permitted for resumes and cover letters, allowed with disclosure on take-homes, restricted in live rounds. For some roles, Datadog added an AI-assisted coding round in 2025 where evaluation shifts to "direct an AI tool, evaluate suggestions, debug errors."

That phrase is the key. The hiring bar did not drop. It moved.

Why "AI-allowed" is a stricter bar, not a looser one

Permitting AI compresses coding time, which pushes the evaluation signal onto code review under pressure, and that is a skill most resumes do not advertise. When Canva's interviewer watches a candidate accept a Copilot suggestion that quietly breaks a null check, they see it in real time. The ban-and-detect regime at Google punishes you for using the tool. The Canva regime punishes you for using it badly.

Fabric's "State of AI Interview Cheating in 2026" report, covering 19,368 interviews between July 2025 and January 2026, puts a number on how much the ban camp is losing:

Six in ten flagged cheaters passed anyway. Detection-only is losing, which is why Canva's "transparent use" camp and Google's "in person" camp are converging on the same selection pressure from opposite directions: both are trying to force the interview back onto a signal the AI cannot fake, which is judgment about AI output.

The escalation curve is what tipped it. Fabric flagged 9% of candidates in July 2025, 45% by September 2025, then rates stabilized. CodeSignal's own data shows technical-assessment cheating rose from 16% to 35% year-over-year. A Gartner survey found 72.4% of recruiting leaders now run at least some interviews in person to fight fraud. Two different responses, same underlying trigger.

The 4 bullets to rewrite, by camp

Rewrite each of these bullets twice: once for an AI-permissive employer, once for an AI-restrictive one. Keep the underlying work the same; change what the bullet emphasizes.

Bullet 1: The "shipped a feature" line

Old: "Built a React dashboard that cut customer support tickets 22%."

For Canva, Meta, Datadog: "Directed Cursor to scaffold a React dashboard; rejected 3 of 7 agent-generated components after review, shipped revision that cut support tickets 22%."

For Google, McKinsey, Anthropic: "Architected and shipped a React dashboard, authored the component library, cut support tickets 22%."

The permissive version signals the skill Canva's October 20 follow-up post actually tests: direct, verify, extend. The restrictive version signals authorship, which is what Google's in-person panel is now re-weighting.

Bullet 2: The "code review" line

Old: "Reviewed PRs for a team of 8 engineers."

For permissive employers: "Reviewed 40+ PRs/month including Copilot-generated changes; flagged 1 incident of hallucinated API usage before merge."

For restrictive employers: "Led code review for a team of 8; owned the merge-queue standards and reduced revert rate 30%."

The permissive version turns review into the central skill, because that is what AI-enabled rounds now measure. The restrictive version keeps review as leadership scope.

Bullet 3: The "refactor" line

Old: "Refactored legacy billing module from Python 2 to Python 3."

For permissive employers: "Ran a Claude Code agent across the billing module's 14k lines, hand-corrected 212 of its suggested edits, migrated Python 2 to 3 with zero prod incidents."

For restrictive employers: "Migrated the 14k-line billing module from Python 2 to 3 with zero prod incidents; wrote the test harness that caught 212 edge cases."

Same work, same number. Different claim about who did it.

Bullet 4: The "mentored juniors" line

Old: "Mentored 3 junior engineers."

For permissive employers: "Coached 3 junior engineers on AI-assisted workflows; set the team's guidelines for when to accept vs. reject Copilot suggestions."

For restrictive employers: "Mentored 3 junior engineers; led the design-review ritual that shortened onboarding from 10 weeks to 6."

Pasting a posting into Refolk is the fastest way to see which angle an employer is actually asking for. The posting language from Canva and Datadog tilts hard toward verbs like "direct," "evaluate," and "disclose"; the language from Google and McKinsey tilts toward "author," "design," and "architect."

The index numbers nobody has published

Early movers get a brief window. In my index of professional profiles, only 1,279 US software engineers at the SWE, Senior, Staff, or MLE level currently list GitHub Copilot or Cursor as a skill, and 1,244 use "AI-assisted," Copilot, Cursor, or Claude Code in a headline. That is tiny compared to the overall US engineer population, which is the point.

Here is the dataset the two camps are fighting over:

Two things jump off that table. First, 64% of US engineers claiming AI-assisted fluency are already Senior-titled, which means a junior candidate writing the same bullets looks like they are reaching. Second, the top employers of engineers who advertise Copilot or Cursor fluency on their profile include Microsoft, JPMorgan Chase, HashiCorp, Google, FedEx, and Walgreens Boots Alliance. That mix is enterprise-heavy, not AI-startup heavy. The signal is more valuable at a bank than at an AI lab, because the AI lab assumes you have it.

The junior squeeze

Juniors should not use the same bullets. Fabric found junior candidates cheat at roughly twice the senior rate, which means interviewers at AI-permissive employers are watching junior AI usage harder, not more leniently. Combine that with the finding that 64% of claimed AI-fluency headlines are senior, and a junior who writes "directed a Claude agent to refactor X" reads as either dishonest or over-reaching.

Junior-safe versions of the four bullets:

• "Paired with my team lead using Copilot on a 2k-line migration; owned the test coverage."
• "Observed and documented our team's AI code-review heuristics during my first 90 days."
• "Shipped feature X using Cursor under supervisor review; 3 of my 11 PRs required rework."
• "Logged which AI suggestions I incorrectly accepted during mock coding rounds, and what the correct review step would have been."

That last one is unusual on a resume, but it signals the exact self-awareness Canva's October 20 interviewer guide calls out as a differentiator.

What interviewers are actually watching for

The named tool to know is Cluely, a $50 to $150 per month interview-assistant Fabric cited in 45% of its flagged sessions. The rest breaks down as 34% ChatGPT voice mode, 18% tab switching, 3% off-camera help. Interviewers at both camps have learned the tells, which is why open-permission policies at Canva and Meta are not an invitation to lean on Cluely. They are an invitation to show that you can out-perform someone leaning on Cluely.

The clearest A/B pairing is Rippling and Anthropic. Both are AI-first companies. Rippling publicly allows ChatGPT and GitHub Copilot in coding interviews. Anthropic explicitly bans AI during assessments. The role is the same; the resume bullet should not be.

FAQ

Should I list GitHub Copilot or Cursor as a skill on my resume?

Yes, if the target employer is on the permissive list (Canva, Meta, Datadog, Rippling) or is an enterprise where the tool is still differentiating (Microsoft, JPMorgan Chase, HashiCorp, FedEx, Walgreens Boots Alliance all employ engineers who list it). Only 1,279 US engineers at SWE through Staff level currently do, so the signal is still rare. Skip it for Anthropic or any employer that has moved rounds back in-person.

Is using AI in a Meta or Canva interview actually safe?

It is safe to use, but the evaluation shifts to how well you direct and verify the tool. Canva's follow-up by Llew Vallis describes stronger candidates rejecting bad suggestions, correcting hallucinations, and narrating their review process. Meta's roughly 60-minute AI-enabled round for E4/E5 candidates measures the same thing. Pasting and accepting gets filtered harder than before, not easier.

What should a junior engineer do differently?

Write pair-and-observe bullets, not direct-and-lead bullets. Fabric found juniors cheat at roughly twice the senior rate, so interviewers scrutinize junior AI claims harder. Phrases like "paired with," "under supervisor review," and "documented the team's heuristics" land better than "directed an agent."

How do I tell which camp an employer is in?

Read the posting verbs and check their careers page. Datadog publishes a dedicated AI guidelines page. Canva publishes two engineering blog posts. Google, McKinsey, Cisco, and Deloitte UK have each publicly discussed reinstating in-person rounds. If the posting uses "author," "design," and "architect," write for the restrictive camp; if it uses "direct," "evaluate," and "disclose," write for the permissive one.