The piece argues that large language model agents have already collapsed the costliest part of software creation - writing code - and are on track to absorb almost the entire development lifecycle. Two big assumptions underlie this forecast: agents will learn to handle review, testing, deployment, monitoring and scaling; and demand for software is effectively unbounded because most organizations still run on brittle, manual processes that could be automated if cost fell. The author breaks development tasks into categories: collapsed (actually writing code), going soon (code review, maintenance), next (shipping to production, scaling), and possibly safe (discovering what to build, defining “good”, and crafting delightful UX). Benchmarks show rapid improvement (fixing-bug scores rising from ~50% to ~95% on many tests), Claude Code’s 567 PRs merged at 84% (about half without human edits), and examples of agents finding complex bugs. At the same time platforms show explosive agent activity: GitHub metrics jumped (36M developers, ~25% more commits, merged PRs ~3.5x since 2023) and one estimate cites 17M agent PRs per month; some projects have closed contributions as AI noise increased.
Those shifts are already reshaping labor markets: Stanford data finds a 19% employment gap for 22-25 year olds in AI-exposed jobs, and entry-level hiring is down ~65% at big tech and ~75% at startups while engineering’s share of hires rose from 46% to 55%. The implication is that junior roles - whose work was producing code from specifications - are disappearing, and the durable human work left is product discovery, domain judgment and translating tacit customer needs into exact specifications. These product-engineering skills resist automation because they require context, conversation and bespoke decisions that don’t scale or live in training data, so future software professionals will be more like specialists in eliciting and defining human wants than typists of code.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
OpenAI's Decisions API can evaluate GitHub pull requests by analyzing diffs against specific review questions, such as compatibility or migration needs. This helps organize review processes by providing clear answers to targeted concerns. (vercel.com)
Strands Decider 2B is an open-source decision model designed for building AI agents. It supports various tools, integrations, and deployment options for creating and managing intelligent systems. (strandsagents.com)
UniEvo-VL is a framework that enables multimodal models to improve themselves by using their own feedback during testing. It enhances image generation performance without external supervision, showing significant gains on benchmark tests. (arxiv.org)
OpenAI has produced mathematical manuscripts and supporting proof artifacts available on GitHub. These resources aim to advance research and understanding in mathematical AI applications. (github.com)
OpenAI is sharing its progress in mathematics research and tools, including a GitHub repository. The initiative aims to engage the mathematical community with AI developments, despite some controversy over sharing permissions. (openai.com)
An AI agent attempts to perform ten web errands daily using only plain web requests, with all transcripts published publicly. The experiment scores 113 well-known sites on their usability for AI agents, highlighting the challenges and successes encountered. (agentability.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.