The Business AI Briefing

LG AI Research released K-EXAONE 2.0, 750 billion parameters, Apache 2.0, on Hugging Face. If your model policy still lists only US vendors, your continuity plan is stale.

CEOs keep funding AI experiments. ROI stalls because nobody owns the outcome, the baseline, or the kill criteria.

July 31 public beta of V4-Flash-0731: vendor agent benches beat Pro preview at roughly one-third the output price—while Nikkei flags a heating price war and peak-hour pricing plans.

July 30 updates add production Gemini 3.5 Flash and preview feedback services—plus API breaking changes around sampling parameters and turn-structure validation for agent deployments.

A solo operator documents a production OpenClaw pipeline with isolated researcher, writer, and publisher agents—scoped tools and parallel execution after shared-context failures.

Smarsh × FTI Consulting 2026 findings: 55% of enterprises deploying AI, only 26% say governance keeps pace, and just 30% can reliably detect unauthorized shadow AI.

Final Article 50 guidelines landed in July 2026; chatbot disclosure, deepfakes, and public-interest AI text face enforcement from 2 August—with fines up to €15M or 3% global turnover.

Independent benches put Inkling-Small within a point of its larger sibling at under a third the parameters—routing beats paying frontier rates for every ticket.

GPT-5.6 Luna drops 80% and Terra 20%. Cheaper tiers only help if you map work to the right model.

Missing outcome owners and nine-month hardening cycles kill AI pilots, not weak models. Stop funding experiments nobody can be held accountable for.

1,000+ frontier-lab employees asked Washington for pacing tools. For SMEs, national coordination still does not install a local stop button on your agents.

Snowflake's Cortex AI Gateway puts agent activity logs and token attribution on the same control plane—because if finance can't see which workflow burned the tokens, nobody is governing AI.

BenchLM July Intelligence Index mirror puts Claude Opus 5 at 60.7 percent, but the top three sit within about 1.8 points across 167 models.

A hidden prompt caught students who never read the chatbot's answer. The same habit shows up in client emails and proposals.

Perplexity Personal Computer for Windows can work across local files, Office 365, and the web from $200 a month. Convenience is easy. Ownership is the real question.

A Kirkland playbook architect is launching Irving with proprietary deal software and a tiny lawyer roster. Owner-led firms should ask which workflows they can productize-with governance-before a competitor does.

Faster model cadence is fine. An approved-model list that goes stale before legal finishes the memo is not. Continuity is now an operating job.

Domino's fifth Enterprise AI Report shows production up, profit stuck. The last-mile gap between deployed models and business users is the bill.

When agents cannot predict their own token burn, the fix is routing, caps, and attribution - not another unlimited pilot.

Another non-U.S. lab parked a 3T-class model on the open-weight calendar. Continuity is now a quarterly vendor map, not a one-time API pick.