An Anthropic hiring manager dropped the coding interview. What junior engineers should do

What happened this week

On October 27, Felix Rieseberg, who builds Claude Cowork and Claude Code Desktop at Anthropic, published advice for junior engineers that opens with the end of the coding interview. “As a hiring manager at Anthropic, I stopped asking candidates to do them,” he wrote. Small coding challenges used to tell him something about a candidate. Now “there’s no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both.”

The post is called “Hunger beats tenure,” and he means it: “I’m seeing engineers who are comparatively new fly past people who are probably better coders and have been around much longer.” He says “junior” is the wrong word for who it’s for. “It’s really for engineers without social capital, status, or power.”

The same day, Amazon published a memo from its senior engineering leaders on the revised role guideline for its software development engineers. It tells its engineers to “learn and apply AI into your daily work as soon as possible,” but it doesn’t define good AI practice, because “any definition we write now will likely look antiquated by this time next year.” The guideline rests instead on what isn’t changing: “impact on our customers, our businesses, and our teams.” The memo calls measures like token usage and code changes “deeply flawed,” warns against making them targets, and adds that “tokenmaxxing is at odds with frugality.” It still finds that they “correlate well with the outputs that matter,” and that people not adopting AI “are falling behind.”

The numbers for people starting out

Both land among hard numbers for new engineers, three of them published this month:

Source	Published	What it found
Anthropic	October 16, 2026	Juniors fell from 27% to 9% of its technical hires since early 2025, and from 21% to 14% of its technical staff
Stanford Digital Economy Lab	October 21, 2026	Employment of 22- to 25-year-olds in software development and customer service has fallen over 3% a year since November 2022, with the decline sharpening to 4.4% in recent months
SignalFire	October 26, 2026	New-grad hiring is down about 65% at 12 large tech companies and about 76% at early-stage startups, compared with 2019
Indeed Hiring Lab	July 23, 2026	4.5% of US software development postings were entry-level in the first quarter of 2026, the lowest share of any sector
New York Fed	Figures for 2024	Recent graduates in computer engineering had 7.8% unemployment and in computer science 7.0%, against 4.2% across all majors

Where software hiring is growing, it’s for experience. US software development postings on Indeed are up almost 15% since Claude Code launched in February 2025, but 71% of the rise from May 2025 to May 2026 came from senior roles. Those who find work tend to do well: in the New York Fed’s 2024 figures, 19.1% of recent computer science graduates worked in jobs that don’t need a degree, against 39.4% of all graduates, and their median early-career pay was $87,000, against $58,000.

AI isn’t the only cause. Indeed’s economists note that New York Fed researchers tie recent graduates’ troubles more closely to remote work, and that others point to interest rate hikes and companies shrinking after the pandemic hiring boom. Anthropic is specific about its own case, though. Talk of needing fewer juniors began in its technical teams in January 2026, its report says: “The teams that relied most heavily on Claude in their own work found that writing code and finding bugs could increasingly be offloaded to Claude.” SignalFire says boilerplate code, unit tests and routine debugging are what graduates used to spend their first 12 to 18 months on, and that “junior engineers can slow a team down because the operational bottleneck has moved from writing code to reviewing it.”

What hiring managers look for now

Anthropic’s economists ran an automated coder over the interview scorecards of thousands of people hired since 2024. Two changes stand out:

• A record of building with AI. Hiring managers mentioned a candidate’s track record with AI tools in around 5% of technical assessments in 2024. By 2026 that had tripled.
• Autonomy and judgment. These drew the biggest increases in comments on technical candidates, while discussion of domain expertise shrank. “I expect models to continue to be jagged, so for my team I want them to be fungible across domains,” one manager said.

Rieseberg is still working out how to interview for people who build well with AI. For now, “I’m scanning resumes and portfolios for little prototypes that exhibit taste, empathy, and principled thinking.” Amazon’s leaders found that engineers “closer to their products and businesses are able to take the most advantage of AI,” and recommend that “ICs of all levels find ways to become deeply connected to your customers.” SignalFire tells companies to hire new graduates “who can manage autonomous workflows, run rapid prototyping experiments, and audit automated code outputs.”

What to do

Rieseberg’s steps are written for engineers already inside a company. With the rest of this week’s advice, they come to six:

1. Learn what the business is trying to do. Rieseberg starts with Patrick McKenzie’s 2011 line that “engineers are hired to create business value, not to program things.” Find out what your company, or at a large one your group, is trying to achieve. The goals of your manager’s manager are “often a useful proxy,” he writes.
2. Pick one ambitious idea, and build it. Come up with a few ideas that would dramatically improve the odds of reaching those goals, then build one as a working prototype over a weekend, instead of writing requirements documents and lobbying for time with designers and product managers. Keep it to one: “junior people often make the mistake of cramming multiple ideas into a single prototype, hiding three excellent ideas among 20 bad ones.” The prototype’s job is “to communicate an idea so well that leadership quickly understands both the opportunity and that you’re the one who can run with it.” His example: two relatively new Anthropic engineers built a working graphical interface for Claude Code over one weekend, and the first public version shipped about eight days later.
3. Use what you notice because you’re new. Newcomers see friction that everyone else has stopped seeing. At Notion, an offline mode was thought impossible, in part because its CEO was assumed to reject it. That part turned out wrong, and “nobody had checked.” If you work on a product, you also know better than your colleagues what it’s like to use from the outside.
4. Show that you check the agent’s work. If reviewing is now the bottleneck, being good at it is the job. Read every change an agent makes, test it, and be ready to explain it. Canva, which expects candidates to use AI in interviews, also expects them “to take full ownership of any code they produce, whether they wrote it themselves or with AI assistance.”
5. Measure yourself by outcomes, not tokens. Amazon’s memo warns that measures like token usage “could, if used as a target instead of a measure, drive negative outcomes.” Show what changed for a user or the business, not how much code an agent wrote.
6. Say where AI doesn’t help. “If your work does not benefit from current AI tooling, that is a finding, not a failing,” Amazon’s leaders write. “An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for.”

Before you build on your employer’s code, check which AI tools it allows.

Is the coding interview dead?

Not everywhere yet. Rieseberg’s change covers his own hiring. Anthropic’s published guidance for candidates, last updated July 10, 2025, still says live interviews are “all you–no AI assistance unless we indicate otherwise,” and that take-home assessments are done without Claude unless it says so.

Others went the other way earlier. Since June 2025, Canva has expected backend, machine learning and frontend candidates to use tools like Copilot, Cursor and Claude in its technical interviews. For backend and frontend roles, an “AI-Assisted Coding” interview replaced its computer science fundamentals screen, though Canva says “code fluency and technical depth were still absolute requirements, just evaluated in a different context.”

So ask each recruiter which rounds allow AI, and practice both ways. Fundamentals matter in both: they’re how you catch an agent’s mistakes.

If you don’t have the job yet

The advice carries over to the job search: the prototype is the portfolio piece Rieseberg says he looks for. SignalFire’s advice to new graduates is to “bypass the permission economy” with “a verifiable portfolio of shipped applications, active open-source contributions, custom agent workflows, and documented customer discovery.” For internships, it suggests teams “actively implementing agentic AI, custom LLM fine-tuning, or data pipeline scaling.”

Some graduates aren’t waiting. SignalFire found that 2025 graduates of the top 20 US computer science programs were twice as likely to call themselves founders as the class of 2022, and 45% less likely to land a job at one of the 12 big tech companies it tracks.

A coding agent is what makes a weekend prototype possible, and you can start one without paying:

Agent	To start free	For more (checked October 29, 2026)
GitHub Copilot	Copilot Free, or Copilot Student for verified students	Copilot Pro, $10 a month
Codex	ChatGPT Free and Go, with GPT-6 Luna in the desktop app, “subject to rollout”	ChatGPT Plus or Pro
Antigravity	“A meaningful quota, refreshed weekly” without a paid Google plan	Google AI Pro or Ultra
OpenCode	Its free models, or a model on your own computer	Any provider you connect
Claude Code	Not on Claude’s free plan	Claude Pro or Max

GitHub says every Copilot plan includes Copilot CLI and the Copilot app, though Copilot Student “excludes third-party agents.” The rest are from OpenAI, Google, OpenCode and Anthropic. A model on your own computer costs nothing to run but needs a capable machine; our memo on Flint covers one.

If you end up with more than one agent, Enjoy runs Claude Code, Codex, Antigravity and OpenCode side by side on the plans you already have, and lets you switch agents in the middle of a conversation. Enjoy is a monthly subscription, with the first seven days free (pricing).

What hasn’t been said

• Whether Anthropic as a whole has dropped coding interviews. Rieseberg writes as one hiring manager, and Anthropic’s candidate guidance hasn’t changed since July 2025.
• What replaces them. On interviewing for people who build well with AI, Rieseberg writes: “I haven’t quite gotten there yet.”
• How many juniors Anthropic hires. Its report gives shares, not counts, and says technical hiring “has continued growing in absolute terms.”
• What Amazon’s guideline itself says. Amazon published the memo about it, not the guideline, and the memo doesn’t mention interviews or junior hiring.
• How SignalFire counts. SignalFire is a venture capital firm, and its figures come from its own Beacon AI platform, which it says tracks more than 650 million people.