OpenAI releases GPT-5 with agentic workflows; pricing drops 40%

What's actually new

GPT-5 extends the context window to 2M tokens, doubles the reasoning ceiling on agentic tasks, and drops the per-token price by 40%. The price cut is the headline number; the context window is the quieter but bigger shift.

A 2M-token window means a model can now hold a whole codebase, a full case file, or a year of customer support transcripts in working memory at once. The implication: tasks that used to require careful chunking, retrieval, or summarisation — the kind of engineering scaffolding that defined "AI engineering" for the last 18 months — simplify to "just put it all in the prompt."

Where the job market moves

• Content and copywriting — a third of junior copy roles get squeezed harder; senior strategic roles get the leverage
• Paralegal and research — tasks most exposed to automation in mid-2025 are now commodified
• Individual contributor engineering — skill ceiling goes up; floor goes up too
• Prompt engineering as a discrete career — mostly gone; becomes an expected competency, not a job title

The profile implication

A resume that says "Proficient in GPT-4" is now useless. The relevant skill is evidence of ship velocity — what you built with the tool, how fast, and what outcomes. A portfolio that shows work dated after each major model release becomes a much more compelling signal than one that just lists the tools.

Write your profile like a changelog of wins, not an inventory of software.