Is whiteboarding dead? What ecommerce technical interviews actually look like now that AI writes the code.

In an April 2026 company post, Sundar Pichai mentioned a number that should reframe how you think about your next interview: 75 percent of all new code at Google is now AI-generated and approved by an engineer, up from 50 percent the previous fall. A company where three out of four lines are drafted by a model cannot keep asking candidates to reverse a linked list on a whiteboard with the AI switched off and pretend that measures anything real. So it started to stop. Google is now piloting an interview where you get Gemini in the room with you, and the thing being graded is how you use it.

That is the headline, and it is tempting to read it as whiteboarding's obituary. The reality is more interesting, and for ecommerce engineers specifically, more reassuring. The whiteboard is not dead. The thing people hated about it is dying, and what is replacing it rewards exactly the skills a good platform engineer already has.

What actually changed at the top

The two companies setting the pattern are Google and Meta, and both moved in the same direction within months of each other.

Meta moved first. In July 2025, 404 Media reported from internal Meta communications that the company was building a coding interview where candidates could use an AI assistant, with one internal memo noting it would be more representative of the real developer environment and would make LLM-based cheating less effective. The format reached actual candidates that October, replacing one of the two onsite coding interviews. It runs sixty minutes in a CoderPad environment with an AI assistant built in, and candidates can switch between models mid-interview, including GPT, Claude, Gemini, and Llama. Instead of two algorithm puzzles, you get a multi-file project to iterate on. Meta evaluates four things: problem solving, code quality, verification, and communication. The official line is that AI use is optional, which is technically true and practically not, because the candidates who refuse to touch it are competing against candidates who use it well.

Google's pilot, reported in 2026, follows the same shape with a different emphasis. It is worth being precise about scope, because the coverage tends to inflate it: this is a limited pilot for junior and mid-level roles on select US teams, with a full transition expected to take twelve to eighteen months, so for a while candidates will hit either the old or the new format depending on the team. Within the pilot, candidates work in a three-panel CoderPad layout, a file explorer, a code editor, and a Gemini chat window that can suggest but cannot directly edit files. The new round is "code comprehension": you are handed an existing multi-file codebase and asked to read it, find bugs, implement a feature, and optimize it. Google describes the format internally as "human-led, AI-assisted," and the interviewers are explicitly scoring something they call AI fluency, which breaks down into prompt engineering, output validation, and the ability to debug a suggestion that is close but wrong.

Read those criteria again, because they are the whole story. Prompt engineering, output validation, debugging AI suggestions. None of that is "can you produce a sorting algorithm under pressure." All of it is "can you tell when the machine is wrong and fix it." That is a senior skill, and it is the one the old format never tested.

The part everyone gets wrong: the whiteboard didn't disappear

The clickbait version of this story is that algorithms are over and you never have to think about big-O again. That is not what happened, and believing it will cost you offers.

What actually happened is a split. The rehearsed part of whiteboarding, the part where you reproduce a memorized algorithm to prove you studied, is what AI made worthless, because the model does that instantly and so can anyone. But the conversation around the whiteboard, the part where you scope a vague problem, define edge cases, and reason about tradeoffs out loud, got more important, not less. In one recent Google loop, the round that produced the most hiring signal was not the coding exercise. It was ten minutes spent scoping the requirements for a trivial function, snake_case to camelCase, where the interviewer cared only about how the candidate defined the edge cases and input formats. The recruiter's verdict afterward was that it produced the most signal precisely because it could not be rehearsed.

That is the tell. Interviewers are deliberately steering toward the parts of the job that cannot be pattern-matched from a prep course, because the prep courses and the AI assistants have made everything else cheap. System design interviews, which are whiteboarding in everything but name, carry more weight than ever at senior levels, conducted on virtual canvases like Excalidraw with no code written at all. The whiteboard survived. The trivia died.

What this means for ecommerce engineers

Here is the part that should change how you prepare, because the ecommerce world was never really a leetcode world to begin with, and the industry is now moving toward where you already are.

Walk through how the platforms actually interview and the pattern is obvious.

Shopify has run a collaborative pair-programming format for years, where you write code with an interviewer who acts as a teammate rather than a silent examiner, and the company has long prized clean, pragmatic code over algorithmic puzzles. A large chunk of the loop is the technical deep dive, a sixty-minute conversation about a real project where you made meaningful decisions. And Shopify has already crossed into the new world directly: candidates now report that AI use in the pair-programming round is optional and even encouraged, with interviewers explicitly assessing how well you evaluate AI-generated code. That is human-led, judgment-first interviewing, and the bones of it predate the AI panic by half a decade.

WooCommerce and agency work lean even harder on the take-home and the project walkthrough, because the actual job is rarely "invent an algorithm." It is reading an unfamiliar codebase, understanding a plugin's hooks, debugging someone else's integration, and extending it without breaking checkout. The hiring-assessment platforms built for this ecosystem have made the parallel explicit. CodeSubmit, which supplies WordPress take-homes, describes a review flow that maps the git tree, reads the README and the most-modified files to understand how a submission actually works, then carries shortlisted candidates into a live "AI-enabled" follow-up that scores prompt quality, critical evaluation, and how they troubleshoot with the same repo on screen. That is Google's code-comprehension round and Meta's AI-fluency rubric, described in a WordPress hiring product, shipped before either big-tech pilot made headlines. The standard practitioner advice for vetting a WordPress developer says the same thing in plainer terms: skip the algorithm trivia, hand them a real task that mirrors the work, and walk through two or three past projects probing architecture, the specific hooks used, and the tradeoffs they made.

Magento and Adobe Commerce interview the same way, with the platform's own complexity standing in for the puzzle. Serious Magento screens are built around architecture thinking and production readiness rather than raw PHP, and the questions that carry the most signal are exactly the code-comprehension ones: how do you safely override core functionality without breaking upgrades, how do you debug a slow checkout and find the bottleneck, how do you resolve conflicts between two extensions that rewrite the same class. None of that is rehearsable from a leetcode list. All of it is "here is a large, opinionated codebase, prove you can reason inside it without breaking production," which is the same thing Google now hands candidates a multi-file repo to test.

The pattern across all three is hard to miss. The big platforms are converging on the interview that ecommerce shops have quietly run for years.

That convergence is also why the take-home itself is not going away even as live whiteboarding mutates. Take-home assignments correlate better with long-term performance than live coding (0.62 versus 0.57 in one analysis), and most developers prefer them. The catch in the AI era is obvious: if you can use AI on a take-home, so can everyone, and the assignment stops discriminating. The smart shops have already adjusted, pairing a take-home with a live walkthrough where you defend your architectural choices. You cannot defend code you do not understand, which means the walkthrough quietly tests the one thing AI cannot fake for you: whether you actually know why your own solution works.

The skills that survive

Strip away the platform-by-platform detail and the same short list keeps surfacing, across Google's rubric, Meta's four criteria, and every ecommerce loop worth taking.

Reading code matters more than writing it. Engineers at every level now read and modify existing code far more often than they write from scratch, and the interview has followed the job. Practice reading unfamiliar code fast and saying what is wrong with it.

Judgment about AI output is the new core competency. The failure mode interviewers are screening against has a name now, "vibe coding," typing a vague prompt, pasting the result, and hoping the tests pass. The skill they reward is the opposite: using the assistant for the well-defined subtask while keeping a firm grip on whether its output is actually correct.

And communication is no longer the soft skill it was filed under. When the code is cheap, the explanation is the product. Scoping a problem, naming your assumptions, and walking someone through a tradeoff are now the highest-signal things you can do in a room, because they are the things neither a prep course nor a language model can hand you.

So, is whiteboarding dead?

The marker-and-glass ritual where you silently reconstruct merge sort to prove you grinded the prep is dead, or close enough that you can stop optimizing for it. What replaced it is an interview about reading code, working with a tool that is now part of the job, and explaining your reasoning to another human who is trying to figure out whether they would trust you with production.

For ecommerce engineers, that is not a threat. It is the format your corner of the industry has been using all along, finally going mainstream. The companies are no longer asking whether you can write the code. They are asking whether you can tell when it is wrong. If you have spent years debugging someone else's WooCommerce plugin or untangling a Liquid template at 2am, you have been training for this interview the whole time.