Healthcare and Artificial Intelligence: Where the Field Stands in 2026 and Where It Could Be by 2035

Purpose. Healthcare is the industry where the promises of artificial intelligence are easiest to state and hardest to check. This article places the field as of October 2026 and describes, with the limits that any look ahead has, where it could be by about 2035. Design and methods. A review that combines three kinds of evidence: the share of the healthcare research literature that involves AI (OpenAlex, 2010–2025), the public list of AI-enabled medical devices authorised by the U.S. Food and Drug Administration, and the randomised trials and external audits that have tested AI tools in practice. Claims taken from the literature were checked against the cited sources. Scenarios for the near, medium and long term are built on this evidence and are judgements, not model forecasts. Findings. The share of healthcare works that mention AI rose from 1.2% in 2018 to 8.5% in 2025, and language-model works went from 319 to 15,209 in three years. The FDA list held 1,614 devices by 29 June 2026, with 335 authorised in 2025; 76% were in radiology and 96% went through the 510(k) route. Trial evidence is concentrated in a few tasks: AI-supported mammography reading was non-inferior to double reading on interval cancers in a trial of 105,934 women and raised sensitivity from 73.8% to 80.5%, whereas ambient documentation scribes cut note time by 9.5% for one product and not significantly for another, and access to a language model did not improve physicians' diagnostic reasoning in a randomised vignette study. We expect documentation and screening tools to become routine by 2028, outcome-based evidence and reimbursement to decide the pace in 2029–2032, and the picture beyond that to remain open. Originality. The article ties research activity, regulatory records and trial evidence in one view of a single industry, states what each source can and cannot show, and sets out scenarios with signposts that readers can check as they unfold.