Beyond the Prompt

At its "The Briefing: AI for Science" event, Anthropic announced Claude Science, an AI workbench that unifies fragmented scientific tools and datasets, and said it will pursue drug discovery for neglected diseases. The move puts a major frontier AI lab in direct competition with the pharma and biotech customers it also sells to — joining a crowded field that includes Insilico, Isomorphic Labs, and Big Pharma's in-house AI teams. University of Cambridge, UCL, and Oxford experts quoted in coverage say AI already accelerates hypothesis generation and molecule search, but no AI-designed drug has completed FDA approval, high-quality experimental data remains scarce, and human trials still take the better part of a decade.

The Junnar forest division in Pune district has deployed AI-based animal detection across 20 high-risk villages, extending an earlier network of 55 systems across 30 villages. Each unit uses a 180-degree camera to detect movement within 100 metres, automatically triggering a loud siren and sending an image alert to a mobile app used by forest officials and local volunteers. Officials report faster response times and fewer surprise encounters; the division has trapped 155 leopards in 14 months amid one of India's most prominent human-leopard conflict zones. Forest officials acknowledge a key limitation: leopards sheltering in sugarcane fields are harder for roadside cameras to detect. Early warning helps, but technology alone cannot eliminate conflict.

An AI model developed at the Mayo Clinic and published in the journal Gut can identify early signs of pancreatic cancer on CT scans up to three years before a diagnosis — and outperformed radiologists by a factor of three. The model detected subtle cellular abnormalities that protect the disease from immune defenses: signals so faint that even specialist radiologists missed them. With pancreatic cancer carrying a 13% five-year survival rate and projected to kill more than 52,700 people this year, early AI-driven detection could be transformative. The tool is now in clinical trials targeting high-risk patients with a family history but no visible symptoms. "We knew that the signal was there. We just needed to find a way to be able to detect it," said Dr. Ajit Goenka, radiologist and study co-author.

Gartner's 2026 Hype Cycle for Agentic AI places the technology at the Peak of Inflated Expectations — the fastest adoption intent of any technology in this year's CIO survey. The support layer (governance, agent security, cost management) is still maturing. Gartner predicts over 40% of agentic AI projects will be cancelled by 2027 without proper governance frameworks and clear ROI criteria. Most current deployments remain narrowly scoped assistants, not true goal-setting agents.

In the world's largest annual survey of data and analytics professionals — 1,579 respondents across industries and regions — data quality management returned to the #1 priority for 2026, beating AI, generative AI, and every emerging tech trend. Rounding out the top five: data security, data-driven culture, AI governance, and data literacy. BARC analyst Robert Tischler's framing captures the gap: "AI is a multiplier. It multiplies value if the data is good. It multiplies risk if the data is bad."

McKinsey's 2025 State of AI survey found 88% of organisations now use AI in at least one business function, but only about one-third are deploying it at enterprise scale. Larger companies (over $5B revenue) are roughly twice as likely to be scaling. McKinsey also found that 51% of organisations have already experienced a negative impact from an AI deployment — mostly around quality, trust, and workflow disruption. The conclusion: the challenge in 2025 wasn't access to AI. It was whether teams actually changed how they work around it.

Gartner predicts 40% of enterprise applications will include task-specific AI agents by end of 2026 — up from fewer than 5% in 2025. Gartner also flags "agentwashing": AI assistants being relabelled as agents. The distinction matters. Assistants respond to a prompt. Agents set goals, take actions, and adapt. By 2035, Gartner estimates agentic AI could represent 30% of enterprise software revenue — over $450 billion — up from 2% in 2025.

Anthropic's Project Vend tested Claude in a semi-autonomous role managing a vending machine business end-to-end — choosing products, pricing items, and responding to demand signals without constant human input. The system demonstrated basic multi-step execution and real-time adaptation, but revealed critical gaps: irrational pricing decisions, inconsistent strategy under changing conditions, and misplaced confidence where assumptions were weak. The experiment marks a directional shift as AI moves from output generation into partial operational ownership, exposing a new class of control, monitoring, and correction challenges that current architectures don't yet reliably solve.