# Backend developer

> Owns the part where being wrong is expensive and invisible — the data that must stay consistent, the money that must not move twice, and the service that has to be up at 3am.

Page: https://flyvolo.ai/en/careers/backend-developer
Impact index: 54/100 · confidence Low confidence · assessed 2026-09-12
This is not a probability of losing a job.

## Scope

Written for engineers building services, APIs and data layers. It cuts by layer, not by seniority — the junior and experienced pages cut the other way and a senior backend engineer sits on both. Platform and site-reliability work overlaps but is answerable for different things; data pipelines have their own page.

## Tasks

### Endpoints and the plumbing between them

- Direction: Automating
- Basis: evidenced (verified records attached)
- Why: Well-specified, heavily represented in training data, and verifiable by running it — the same three properties that make any task a strong case for generation. It is also a large share of what backend work looks like from outside.
- What this does NOT establish: Generating an endpoint is not the same as deciding it should exist, what it must guarantee, or what happens when it is called twice. The typing was rarely the expensive part of this work.
- Records:
  - 2026-04-13 · Deployment · Thirty-one of forty-five US federal agencies reported a commercial AI coding tool in the 2025 federal inventory, naming GitHub Copilot, Claude Code, Codex and Poolside — https://flyvolo.ai/en/changes/ev-20260413-backend-developer-2 (source: https://github.com/ombegov/2025-Federal-Agency-AI-Use-Case-Inventory/blob/3c225ba8438e48306ace7698c8c7feb9486cbc69/Data/2025_consolidated_COTS_AI_use_cases.csv)
  - 2024-10-29 · Deployment · Alphabet's CEO said on the Q3 2024 earnings call that more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers — https://flyvolo.ai/en/changes/ev-20241029-backend-developer-3 (source: https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q3-2024/)
  - 2024-08-01 · Deployment · Amazon says it moved tens of thousands of its own production Java applications from Java 8 or 11 to Java 17 with an agent, and estimates the manual equivalent at over 4,500 years of development work — https://flyvolo.ai/en/changes/ev-20240801-backend-developer-1 (source: https://aws.amazon.com/blogs/devops/amazon-q-developer-just-reached-a-260-million-dollar-milestone/)

### Keeping it correct when things happen at once

- Direction: Still human-led
- Basis: evidenced (verified records attached)
- Why: These bugs do not appear in tests and often do not appear for months. Reasoning about them requires holding a model of what else is running and what a partial failure leaves behind, and the correct answer depends on guarantees the business has made rather than on the code in front of you.
- What this does NOT establish: Uber reports that a tool now drafts fixes for low-level data races in its Go code, with code owners accepting most and rejecting some that passed tests but were wrong; that is races in one codebase, not transactions or ordering across services. Incidents are almost never attributed publicly to generated concurrency code, and that silence is not evidence that it does not happen.
- Records:
  - 2025-04-22 · Deployment · Uber says its LLM-based tool produced fixes for 224 of 404 data races in its Go codebase over 18 months, and developers accepted 193 of them in code review — https://flyvolo.ai/en/changes/ev-20250422-backend-developer-4 (source: https://arxiv.org/abs/2504.15637)

### The data model, and changing it while it is in use

- Direction: Still human-led
- Basis: evidenced (verified records attached)
- Why: Early schema decisions become expensive years later in ways no tool can see from the current code, because the cost lives in what was already written into the data. A migration is also irreversible in practice, which puts it in the category where somebody must be accountable rather than assisted.
- What this does NOT establish: Google reports that most of the code in one large migration of IDs was written by AI while every change was reviewed and rolled out by people; that is a code change at one company, not a live schema migration. The claim is about who decides and answers for it, not about who types it.
- Records:
  - 2025-01-12 · Deployment · Google says 80% of the code changes that landed in a migration of Ads IDs from 32 to 64 bits were fully AI-authored, while code review and rollout still needed a person — https://flyvolo.ai/en/changes/ev-20250112-backend-developer-5 (source: https://arxiv.org/abs/2501.06972)

### Who is allowed to see what

- Direction: Still human-led
- Basis: evidenced (verified records attached)
- Why: Generation optimises for the request that was described, and an authorisation bug is precisely the case nobody described. This is also the area where a confident, plausible, wrong answer is most dangerous, because it looks exactly like a correct one until someone exploits it.
- What this does NOT establish: The UK's cyber security agency advises writing authorisation logic closer to by hand, and a benchmark found about half of the models' working backends exploitable; neither is a defect study separating generated from hand-written code in the same codebase, which is what would settle it. Regulated environments already require human sign-off here, which may be doing more of the work than the difficulty is.
- Records:
  - 2026-06-18 · Constraint · The UK's National Cyber Security Centre advises developers to move towards writing code by hand when building authentication or authorisation logic with AI assistance — https://flyvolo.ai/en/changes/ev-20260618-backend-developer-6 (source: https://www.ncsc.gov.uk/blogs/the-vibe-coding-spectrum-approach-to-ai-assisted-software-development)
  - 2025-05-30 · Constraint · On 392 backend tasks, the best model's code was correct and secure only 35% of the time, and about half of the models' working programs could be exploited — https://flyvolo.ai/en/changes/ev-20250530-backend-developer-7 (source: https://arxiv.org/abs/2502.11844v3)

### What it costs and how fast it is

- Direction: Being augmented
- Basis: evidenced (verified records attached)
- Why: Tools are genuinely strong at spotting the pathological query and suggesting the index. They are weak at the trade-off — spending money to be faster is a business decision, and knowing which request matters requires knowing what the product is for.
- What this does NOT establish: Nothing here measures how much of this is now tool-driven in practice, and the answer differs enormously between a team with an observability budget and one without.
- Records:
  - 2024-08-01 · Deployment · Amazon says it moved tens of thousands of its own production Java applications from Java 8 or 11 to Java 17 with an agent, and estimates the manual equivalent at over 4,500 years of development work — https://flyvolo.ai/en/changes/ev-20240801-backend-developer-1 (source: https://aws.amazon.com/blogs/devops/amazon-q-developer-just-reached-a-260-million-dollar-milestone/)

### Being the one the pager wakes

- Direction: Still human-led
- Basis: evidenced (verified records attached)
- Why: Diagnosis is increasingly assisted and that genuinely helps. The decision is not: choosing to accept a known loss in order to restore service is a judgement with consequences somebody has to own, and it is made with incomplete information by design.
- What this does NOT establish: This says nothing about whether on-call load is rising or falling, which is the question most engineers actually care about and the one nobody publishes.
- Records:
  - 2023-11-13 · Deployment · Microsoft says an LLM-based incident tool's data-collection part has been used by over 30 of its teams for four years, and its root-cause prediction by one incident team — https://flyvolo.ai/en/changes/ev-20231113-backend-developer-8 (source: https://arxiv.org/abs/2305.15778v4)

## How to cite this

- The unit is the task, not the job title: say which tasks are changing and which are not.
- Carry each task's basis and its limits with it; an inferred judgement is not a finding.
- This page gives no number of years. Do not derive one from it.
- More: https://flyvolo.ai/llms.txt