Technology has spent decades cast as obesity’s accomplice — the glowing screens that keep people in their chairs and the delivery apps that summon fast food to the doorstep. A sweeping new review argues that the same machinery is quietly becoming obesity’s most versatile weapon. Writing in the open-access journal Advances in Therapy, endocrinologists Shinjan Patra, Akhila Bhandarkar, Nitin Kapoor and Sanjay Kalra chart a year of progress in which artificial intelligence flags weight gain before it becomes visible, deep-learning networks map body fat compartment by compartment, ever more potent medications reset the benchmarks of drug therapy, robots and endoscopes reshape stomachs through incisions no wider than a keyhole, and head-mounted displays coach patients to stare down a virtual doughnut without eating it. Drawing together landmark trials, machine-learning studies and randomized controlled data, the review portrays a discipline breaking decisively out of the consultation room.

The authors’ premise is disarmingly simple: obesity care has always been a logistics problem as much as a metabolic one. Conventional face-to-face programs work, but their benefits decay quickly, and commercial weight-loss schemes have never convincingly delivered long-term results. Trials that reduced the frequency of patient contact made outcomes worse, not better, because accountability and personalized feedback — the most fragile elements of any weight-management program — are precisely what thin out when clinic time is scarce. Smart technology, the review argues, can keep those components intact while cutting cost and provider time, and it can compensate for a well-documented gap: many primary care providers report inadequate training in psychological and behavioral counseling, the very skills obesity care demands. For any tool to qualify, the authors propose a seven-part test — it must be adaptable, affordable, accessible, accurate, intuitive, sustainable and scalable.

The review begins by dismantling the number most patients live by. Body mass index, long the gatekeeper of obesity diagnosis, is now formally considered insufficient: the Lancet Diabetes and Endocrinology Commission and the European Association for the Study of Obesity have both concluded that weight and height cannot distinguish fat from muscle, or visceral from subcutaneous fat, and that body-fat percentage varies with age, sex and ethnicity at any given BMI. Visceral adiposity — the metabolically active fat around the internal organs — can rise dangerously in people whose BMI barely moves, degrading insulin sensitivity and seeding diabetes. The Endocrine Society of India now classifies a BMI of 23 to 24.9 kg/m² as overweight and anything above 25 kg/m² as obese. Filling the diagnostic gap are convolutional neural networks: deep-learning architectures trained on magnetic resonance imaging, computed tomography and dual-energy X-ray absorptiometry scans that automatically segment and quantify visceral fat, subcutaneous fat and ectopic fat deposited inside organs. These AI-derived body-composition fingerprints predict cardiovascular events more effectively than BMI alone.

The pharmacological section reads like a dossier of record-breaking trials. In the phase 3b STEP UP trial, 1407 adults with obesity but without diabetes were randomized five-to-one-to-one to once-weekly injectable semaglutide at 7.2 mg, the established 2.4 mg dose, or placebo, alongside lifestyle intervention, and followed for 72 weeks. The higher dose removed 18.7 percent of body weight, versus 15.6 percent for 2.4 mg and 3.9 percent for placebo. Patients on 7.2 mg were 1.8 times more likely to lose a fifth of their weight and 2.4 times more likely to lose a quarter, and although gastrointestinal side effects were more frequent, serious adverse events were less common than at the lower dose, affecting 6.8 percent versus 10.9 percent of participants. For needle-averse patients, the OASIS 4 study tested oral semaglutide at 25 mg daily across 22 sites in four countries: after 71 weeks, participants had lost 13.6 percent of their body weight against 2.2 percent with placebo, with measurable gains in physical-function quality of life.

Tirzepatide, a dual GIP and GLP-1 receptor agonist, delivered the year’s most consequential head-to-head result. In the open-label SURMOUNT 5 trial, adults with obesity received either the maximum tolerated dose of tirzepatide, 10 or 15 mg weekly, or the maximum tolerated dose of semaglutide, 1.7 or 2.4 mg weekly, for 72 weeks. Tirzepatide produced 20.2 percent mean weight loss against 13.7 percent for semaglutide. Its reach extended beyond the scale: in the SUMMIT trial, 731 patients with heart failure with preserved ejection fraction and a BMI of at least 30 were followed for up to 104 weeks, and cardiovascular death or a worsening heart-failure event occurred in 9.9 percent of the tirzepatide group versus 15.3 percent on placebo, a hazard ratio of 0.62. In the phase 2 SYNERGY NASH trial, biopsy-confirmed fatty liver disease with moderate-to-severe fibrosis resolved in 44, 56 and 62 percent of patients receiving 5, 10 and 15 mg of tirzepatide respectively, versus 10 percent on placebo — hinting at a single weekly injection that treats the liver as well as the waistline.

Behind the headline drugs, an invisible layer of artificial intelligence is reshaping who gets treated and when. Supervised machine-learning models — logistic regression, random forests and gradient boosting machines — sift electronic medical records, combining demographics, medical history, laboratory values, medication lists and social determinants of health into obesity-risk scores. One model built on birth records, pediatric growth charts and family history predicted early childhood obesity accurately in more than 85 percent of cases, and newer versions fold in sleep patterns, screen-time exposure and neighborhood characteristics. Transformer-based natural language processing — the BERT and GPT family of models behind modern chatbots — now mines unstructured clinical notes for dietary patterns, activity levels and psychosocial red flags that structured billing codes never capture. Linked to fitness trackers and calorie-counting apps, such algorithms can catch the earliest lifestyle drifts that precede weight gain. The authors are candid about the caveats: diagnostic algorithms can inherit demographic bias, and informed consent, algorithmic fairness, safety and data privacy remain legally unsettled.

Surgery, too, has been shrinking. Endoscopic sleeve gastroplasty, the flagship of a growing family of endoluminal techniques, uses a suturing device threaded through the mouth to fold and shrink the stomach, altering gastric physiology with no external incisions. Multicenter series and randomized data show clinically meaningful total and excess weight loss at 6 to 24 months, an acceptable safety profile and faster recovery than laparoscopic sleeve gastrectomy, making it attractive for patients with class I to II obesity — a BMI between 30 and 40 — or as a bridge therapy for higher-risk candidates, though long-term durability remains under study. Robotic-assisted bariatric surgery brings tremor-filtered instruments, enhanced articulation and three-dimensional vision to the operating table. The evidence is more equivocal: some registries show comparable or better outcomes, including reduced bleeding, while systematic reviews consistently report longer operative times and higher costs without reliable reductions in complications. Robots, the review concludes, currently earn their keep mainly in technically demanding or revisional operations.

The review’s most striking material concerns virtual reality, which the authors treat not as a gadget but as a clinical instrument. A head-mounted display does not merely show an image; it replaces the user’s world, generating “presence” — the felt sense of being physically inside a simulation — and with it control over sensory exposures no clinic could stage. The theoretical core is the allocentric lock hypothesis: the brain stores the body in two reference frames, an egocentric one built from proprioceptive and interoceptive sensation and an allocentric one anchored to external space. In long-standing obesity, the brain can become locked into a rigid, negative allocentric memory of the body, so that a patient who has lost 20 kilograms still perceives themselves as heavy — a mismatch that breeds behavioral exhaustion and relapse. Virtual embodiment exercises are designed to fuse the two frames and unlock that memory. In a study of virtual reality cue exposure therapy for binge eating and bulimia, 100 percent of patients achieved abstinence from purging immediately after treatment and maintained it for seven months, against roughly 75 percent for standard cognitive behavioral therapy. Over one year, 48 percent of the VR-enhanced group maintained or improved their weight loss, versus 29 percent for cognitive behavioral therapy and 11 percent for standard inpatient care, an odds ratio of 7.03.

Virtual reality also extends to metabolism. So-called exergames such as Supernatural, Beat Sabre and FitXR push players through full-body workouts, and studies using indirect calorimetry — the gold standard for measuring energy expenditure — confirm genuinely vigorous effort, with many sessions exceeding six metabolic equivalents of task, the threshold for vigorous activity. Yet the numbers on patients’ wrists are unreliable: waist-worn accelerometers miss 45 to 65 percent of the energy burned in virtual reality because the workouts emphasize arm movements and static squats that sensors read as standing still, while wrist-worn devices overestimate expenditure by 108 to 112 percent. Clinicians should steer patients toward chest-strap heart-rate monitors or simple ratings of perceived exertion. The same sober accounting applies to the technology’s obstacles: cybersickness from conflict between the eyes and the vestibular system, the digital divide that puts room-scale systems — which demand an unobstructed two-by-two-meter area — beyond many households, missing billing codes and thin long-term evidence. Meanwhile, the unglamorous infrastructure of digital care keeps proving itself: interactive voice response systems improve follow-up rates among patients with limited digital literacy; app-based interventions succeed when frequent self-monitoring is paired with human coaching or evidence-based algorithms; continuous glucose monitors are being repurposed as biofeedback and dietary-adherence tools, with a scoping review of 31 studies finding 93 percent deployed them to track glycemic variability; and randomized trials show telehealth achieves short-term weight loss non-inferior to in-person programs. Regulatory change is afoot too, as the FDA’s prescription digital therapeutics category opens a path for software that can be prescribed and reimbursed like a drug.

The review closes with a warning about the data plumbing underneath it all. Many consumer apps and devices sit outside traditional health-care privacy regimes, and patients consistently voice concern about confidentiality, data sharing and commercial secondary use of their health information. The authors call for privacy-by-design architectures, transparent consent policies, secure cloud systems, federated analytics that keep data on the device, and explicit informed consent whenever third-party platforms touch clinical care. The authors are equally candid about their own analysis: the technologies surveyed are heterogeneous, real-world trial evidence is scarce, and a field moving this quickly will always outrun any snapshot of it. Still, the verdict is confident. Mobile health, eHealth and telemedicine are no longer experiments at the margins of obesity medicine but functioning parts of it, and the decisive task ahead is no longer invention but translation — carrying these results into practice, so that the technology which helped make obesity a global epidemic can be conscripted to help end it.