What Happened
Latent Space published "AI Engineer Will Be the LAST Job" on March 7, 2026, arguing that software engineers will be the final profession standing as AI automates everything else. The piece uses labor market data and platform statistics to back up a provocative claim.
The core data points: Citadel Securities data shows software engineer job postings are rebounding higher even as overall job postings decline and AI models get better at coding. Anthropic reports that software engineering now accounts for over 50% of all Claude model use cases. And the trend is accelerating, not plateauing.
The argument hinges on a specific application of Jevons Paradox. When AI gets better at coding, demand for software engineers does not drop. It increases, because engineers are the ones using AI to automate other professions. Every new AI agent is fundamentally a coding agent with extra skills. Code mode, filesystems, and sandboxes are becoming universal agent capabilities.
The article's final prediction: between AI researchers and AI engineers, the researchers will hang up their hats first. Engineers will be the last ones standing because someone needs to deploy the "last mile" of automation for every industry and workflow.
Why It Matters
This reframes the "will AI take my job" conversation in a useful way. Instead of the vague anxiety that AI will replace everyone equally, it identifies a specific mechanism: software engineers are the operators of AI automation, not just its targets.
The Anthropic statistic is particularly telling. Software engineering taking over 50% of Claude usage means developers are not just early adopters. They are the primary users, and the gap with other professions is widening, not closing. The article warns against assuming other fields will catch up, calling that "the classic egocentric error."
For anyone choosing a career direction or deciding where to invest in skills, this data suggests doubling down on engineering rather than hedging away from it. The tools are getting better, but the people who wield those tools are becoming more valuable, not less.
Our Take
The Jevons Paradox argument is compelling but incomplete. Yes, better AI coding tools increase demand for engineers who can use them. But this assumes a continuous supply of new things to automate. At some point, the automation backlog gets worked through.
What we find more interesting is the practical observation that "all agents are just coding agents with extra skills." This matches what we see in the tool landscape. Claude Code, Cursor, Amazon Q Developer, Aider - these are not niche developer tools anymore. They are becoming the primary interface for building any kind of AI automation, whether it is for marketing, finance, HR, or operations.
The 50% Claude usage figure for software engineering also suggests something about where AI tool development will focus. If half your revenue comes from developers, you build for developers. This creates a flywheel: better dev tools attract more devs, which generates more revenue, which funds better dev tools.
The career advice implied here is sound: learn to build with AI, not just use AI. The gap between someone who prompts ChatGPT and someone who builds agent workflows with Claude Code is growing wider every month. The latter group is not worried about job security.