Best of LinkedIn: Artificial Intelligence CW 15/ 16

Johnson & Johnson narrowed GenAI to a curated set of high-value use cases, pairing them with bespoke guardrails to accelerate scale

Multiple CIO studies show < 5 % of firms have moved GenAI beyond pilots; winners link initiatives directly to P&L and invest in data quality at source

HR leaders re-designed operating models for the AI era, emphasising reskilling and federated Centres of Excellence

Failure-rate analyses (88% of pilots) highlight the need for design-thinking, change-management and measurable KPIs from day one

Google launched Gemini 2.5 Pro; native “thinking-model” capabilities now free to test in AI Studio and via API, upping the bar for reasoning-heavy agents

Adobe unveiled experimental Project SlideWow; turns text outlines into full-fidelity presentation decks, compressing hours of design into seconds

Automation vendor Make released a no-code AI Agent feature; users hand natural-language goals to run end-to-end workflows while they sleep

Zapier rolled out Mission Control Panel; orchestrates 70 k+ app actions with Claude, OpenAI and others, enabling complex multi-agent automations from one console

Starkey introduced next-gen hearing aids with on-device AI models that adapt sound profiles in real time to individual hearing loss

TerraMind (DLR + KP Labs) announced a multimodal foundation model that predicts extreme environmental conditions for Earth-observation missions

Commerzbank embedded a generative copilot into its retail banking app, providing conversational product advice and personalised insights

An “AI Super Agent” startup hit $10 m ARR in nine days, underscoring market appetite for autonomous video- and voice-generation services

European vendors at GenAI Partnerday doubled down on joint go-to-market programmes around foundation models and accelerators

Commerzbank credited a multi-partner delivery team in its successful generative-AI rollout, signalling tighter bank-vendor collaboration ahead

The EU AI Act hits its 100-day countdown on 2 August 2025; organisations must label high-risk systems, stand-up post-market monitoring and budget for tiered penalties

Practitioners warned of common Act misconceptions and pushed for harmonised model registers, transparency reports and privacy-by-design defaults

Public demand for ethical AI continues to outpace regulation, increasing reputational risk for laggards

Anthropic released a hands-on engineering guide for building real-world AI agents, emphasising step-wise reasoning, tool abstraction and safety tests

OpenAI published an enterprise adoption playbook distilling architecture, data-pipeline and operating-model lessons from early deployments

Engineers debated fine-tuning versus retrieval-augmented generation; consensus formed around hybrid stacks for cost-to-quality optimisation

Google’s latest agent demo showcased multimodal perception and long-context planning, signalling rapid progress toward general-purpose digital workers

Velocity records reset as the $10 m-ARR milestone was reached in under two weeks, fuelling investor optimism for autonomy-first SaaS

Snowflake cost-management surfaced as a C-suite concern as generative workloads scale, prompting renewed focus on data-platform efficiency

Boards now demand business-case clarity before green-lighting additional model-training budgets, marking a shift from experimentation to ROI discipline