Laboratory balance beneath a glowing data network and warning symbol
Activity Monitor, Artificial Intelligence, monitoring, Physical Activity

Your data, your prescription? What Google’s personal health agent tells us about AI-driven activity prescription

For more than a decade we have been collecting data from wearables at an incredible rate: steps, heart rate, heart rate variability, sleep, GPS traces, training sessions. Yet the step that matters most has always been the hardest one: going from numbers to a decision. What should this person do next week, given their data, their health history and their life? In high-performance sport we solve that problem with teams of people. For the general population, that team simply does not exist.

A paper just published in Nature by Heydari and colleagues at Google Research, “The anatomy of a personal health agent” (21 October 2026), is the most serious attempt I have seen so far to build that team in software. It is worth reading carefully, both for what it shows and for what it does not show, because it gives us a realistic preview of AI-generated, personalised activity prescriptions based on our own data.

What they built

The authors started from users rather than from technology. They analysed 1,370 real consumer health queries from search, Gemini and the Fitbit forum, surveyed 555 people using an experimental Fitbit Labs chatbot and ran a workshop with 14 experts. Four types of need emerged: general health knowledge, interpretation of personal data, wellness advice, and symptom assessment. These needs require very different skills — numerical precision and code for data questions, factual grounding for knowledge and symptoms, empathetic open-ended dialogue for coaching — and the team found that a single large language model struggled to balance them.

Their solution is a modular, multi-agent system — the Personal Health Agent (PHA) — built on Gemini 2.0, made of three specialists coordinated by an orchestrator:

  • A data science agent that turns a vague question (“am I getting fitter?”) into an explicit analysis plan, then writes and runs Python code on the user’s wearable time series, including comparisons with population percentiles by age and sex.
  • A domain expert agent that reasons over wearable data, blood tests and health history, using tools such as NCBI literature search, web search and population statistics to ground its answers.
  • A health coach agent built around motivational interviewing: it identifies goals first, explores constraints, asks for feedback and only then offers a SMART recommendation.

The orchestrator decides which agent leads and which ones support, rewrites a targeted sub-question for each, checks the combined answer for coherence (and to avoid asking the user for data it already has), and keeps a memory of goals, barriers and preferences across turns.

What the evaluation shows

The evaluation is large: 11 benchmark tasks, more than 7,000 annotations and about 1,350 hours of work from end users and experts, grounded in the WEAR-ME study (1,165 Fitbit users with blood tests and questionnaires). A few results stood out for me.

Data analysis. The data science agent’s analysis plans scored 75.6% against 53.7% for the base model. The most striking gain was in checking whether there was enough data to answer the question in the first place: 66.5% of plans did so, against just 5.3% for the base model. Anyone who has worked with wearable data knows that missing days, non-wear periods and artefacts are where most bad conclusions come from. Its code passed 75.5% of expert-written unit tests at the first attempt (58.4% for the base model), rising to 79% with self-correction.

Health knowledge and context. On more than 2,000 board-style questions in endocrinology, cardiology, sleep medicine and strength and conditioning, the gain over the base model was small (83.6% vs 81.8%). Much bigger differences appeared when personal context mattered: end users rated 96.9% of the domain expert’s personalised answers as trustworthy against 38.7% for the base model, and five clinicians judged 65% of its multimodal health summaries useful against 30%.

Coaching. Users preferred the coach agent on five of six coaching themes, and expert coaches preferred it on all six. Interestingly, both groups gave the plain baseline a slight edge on progress measurement — a weakness the authors then addressed in the integrated system.

The integrated agent. In live multi-turn conversations, 45 users role-playing real WEAR-ME profiles ranked the orchestrated PHA first 46.7% of the time, against 28.9% for a parallel multi-agent design and 24.4% for a single agent. The experts were far more decisive: the PHA was ranked best in 66.7% of conversations, roughly five times as often as the single agent, with an overall quality score of 72.8% against 64.3% and 54.4%.

Why this matters for activity prescription

What struck me most is that the architecture mirrors how a good performance support team works. Someone looks at the data properly, someone interprets it in the context of the person’s health and history, and someone turns it into a plan the person will actually follow. Exercise prescription needs all three. Frequency, intensity, time and type are not chosen from a guideline table alone; they depend on where the person is starting from, what they can safely tolerate, and what they are willing and able to do.

Seen through that lens, the near-future pipeline is easy to imagine:

  1. Baseline from your own data: real habitual steps, active minutes, resting heart rate, HRV and sleep — computed, not guessed, and checked for missing data.
  2. Comparison and context: where you sit relative to people of your age and sex, and what your blood markers and conditions mean for what is safe.
  3. A negotiated dose: a first step that fits your constraints and motivation rather than a generic “150 minutes a week”.
  4. Progression: memory of what was agreed and what actually happened, so the next prescription builds on the last one.

The example conversation in the paper’s extended data shows this already. A user with type 2 diabetes, a BMI of about 36, around 5,500 steps per day and a recent hospital admission asks how to get back to 10,000 steps. Instead of prescribing the target, the agent explores why it matters to her, picks up symptoms that suggest possible sleep apnoea, recommends speaking to her doctor, and agrees on several short 10-minute walks per day as a starting point. That is, in essence, a graded prescription with a red-flag screen built in. Another example I liked: asked whether more activity leads to more deep sleep, the data science agent reported a correlation of about 0.14 and said plainly that the relationship was weak — exactly the kind of honest answer a good analyst gives and a lot of consumer apps do not.

The real opportunity is scale. One practitioner cannot review thousands of people’s data every day and adjust their plan. A system like this could make the first layer of individualised prescription and follow-up available to anyone wearing a device, and free human experts to focus on the people and decisions that need them.

The risks we should not gloss over

The authors are refreshingly upfront about limitations, and they describe the PHA as a research prototype rather than a clinical tool. Still, for anyone tempted to treat this as a ready-made prescription engine, these are the issues I would keep in mind.

  • Safety at scale. In the expert safety review of the clinical summaries, 83% carried no or minimal potential for harm, but 15% were judged mildly harmful and 2% moderately to severely harmful. Two per cent sounds small until it is multiplied by millions of users.
  • No outcome data. Everything measured is conversation quality and short-term task performance. There is no evidence yet that people change behaviour, get fitter or become healthier. For an activity prescription, the only test that ultimately matters is the dose–response over weeks and months.
  • Data quality and generalisability. The data come from one device ecosystem (Fitbit), and consumer estimates of sleep stages or HRV have their own validity limits. Even the improved agent checked data sufficiency in only two-thirds of its plans, so one-third did not.
  • Simulated users. In the live studies, participants role-played personas matched to their own profiles rather than seeking help for themselves, and parts of the evaluation relied on LLM-based autoraters. Both are reasonable research choices, but neither is the real world.
  • Training load is not in scope. The system was not evaluated on periodisation, load management, injury risk, environmental stress such as heat, or athletic populations. Extending it to performance settings needs its own validation.
  • Evidence quality. Open-web retrieval can surface weak or conflicting evidence. The authors themselves suggest restricting retrieval to curated sources and showing provenance next to every recommendation.
  • Privacy, equity and cost. Real deployment needs proper privacy and security controls, fairness checks across populations, languages and literacy levels, and a cheaper architecture: the PHA averaged 6.5 model calls and about 205 seconds per response, against 3 calls and 36 seconds for a single agent.
  • Who built and judged it. The work was funded by Alphabet, most authors are Google employees, and the system runs on Google models and Fitbit data. That does not invalidate the findings — and releasing the code and benchmark is a genuinely good step — but independent replication will matter.
  • Coaching versus prescribing. The coach agent sometimes declines to give a concrete number and steers the user to decide for themselves. Good motivational practice, but people asking for an exercise prescription often need a clear, specific dose — and someone accountable for it.

What I would want to see next

Before trusting an AI-generated activity prescription, I would want to know four things: whether it checked the quality of the data before drawing conclusions, whether its evidence comes from curated sources it shows me, whether it recognises red flags and hands off to a clinician rather than reassuring, and whether its prescriptions are reviewed against what actually changed over time. The next generation of studies should be prospective, longitudinal and outcome-based, with people using the system for themselves, across devices and populations, and with clinicians and coaches in the loop.

Bottom line

This paper does not show that AI can prescribe exercise better than a qualified professional. It shows something more modest and, I think, more important: that splitting the job into data analysis, domain expertise and coaching, and orchestrating them carefully, produces answers that experts judge to be substantially better than a single general model. Personalised activity prescriptions built from our own data are no longer a distant idea. The challenge now is to prove that they are safe and that they actually change behaviour and health — and to make sure the humans who are accountable for those decisions stay firmly in the loop.

Reference: Heydari, A. A. et al. The anatomy of a personal health agent. Nature (2026). https://doi.org/10.1038/s41586-026-11161-8 (open access). Code: Google-Health/consumer-health-research.

Data Analytics, Sports Science

Kona 2026: Heat Decided the Titles, Shoes Make the Times, and Cooling Is the Next Frontier

A week ago I published a preview of the IRONMAN World Championship built around my dashboard, the newest carbohydrate research and the evidence on heat acclimatisation. I ended that post by saying that in Kona the athlete leading with 10 km to go usually wins, unless the heat decides otherwise. On 10 October the heat decided otherwise in both races.

In this post I look back at what happened, how the predictions held up, and why a weekend that also gave us a 2:01 marathon in Chicago says a lot about where endurance performance is going: faster on cool roads in new shoes, and increasingly limited by heat everywhere else.

Two races, two collapses, two champions

The men’s race. Vincent Luis won on his Kona debut in 7:47:27, taking the lead late in the marathon and holding off Gustav Iden by 1:36. At 37, and in what is expected to be the final race of his career, he became the first Frenchman to win the men’s title in Kona.

On paper the race belonged to someone else for most of the day. Magnus Ditlev came off the swim in 30th place and more than three minutes down, and still led into T2 after a 4:06:47 ride. By 30 km of the run he was fading, walking aid stations and stopping for extra ice in his cap, and Luis went past him, followed closely by Iden. Ditlev later described it as the lights going out, which is about as good a lay description of exertional heat strain as you will find.

The attrition was remarkable. Sam Laidlow walked early in the run and withdrew, explaining that he had been struggling with iliac artery endofibrosis. Patrick Lange, three-time champion and course record holder, stopped on the bike and pulled out of his final Kona, and Marten Van Riel was seen vomiting on the bike. Kristian Blummenfelt, one of the clear pre-race favourites, faded to eighth. Officiating also shaped the podium: Iden served a three-minute bike penalty, started the run close to nine minutes down and then posted the fastest run of the day, while Matthew Marquardt dropped from first to ninth after his own penalty and fought back to take third.

The women’s race was even more dramatic. Lucy Charles-Barclay led the swim in 51:29, her seventh swim lead in seven Kona starts, and swapped the lead with Taylor Knibb for much of the bike before withdrawing on the run. Knibb, who was sick late in the race, was passed by defending champion Solveig Løvseth with around 4.5 miles to go. Inside the final kilometre Løvseth could no longer run, and Knibb passed her to win in 8:36:29. It was a complete reversal of last year, when Knibb’s own late collapse opened the door for Løvseth, and it came just weeks after Knibb won a fourth 70.3 world title. Løvseth got back to her feet and walked in to finish eighth.

Behind them, two athletes showed what patience is worth in Kona. Laura Philipp came out of the water more than eight minutes behind Charles-Barclay and finished second, winning the IRONMAN Pro Series. Julie Derron took third on her Kona debut despite a three-minute drafting penalty that left her eighth off the bike, and Kat Matthews came from 14th out of the water to fourth.

The podium splits tell the story.

AthleteSwimBikeRunTotal
Men1. Vincent Luis (FRA)48:314:15:422:38:367:47:27
2. Gustav Iden (NOR)50:104:17:102:36:587:49:03
3. Matthew Marquardt (USA)48:254:11:282:43:587:49:59
Women1. Taylor Knibb (USA)52:554:40:072:57:468:36:29
2. Laura Philipp (GER)1:00:034:43:002:50:438:39:34
3. Julie Derron (SUI)53:094:49:352:58:548:46:25
Sources: Slowtwitch, TRI247, Find Your Edge. Times unofficial until IRONMAN finalises results.

What the data said, and how the predictions fared

Both winning times were well off the course records. Luis was almost 12 minutes slower than Lange’s 7:35:53 from 2024, and Knibb was eight minutes slower than Løvseth’s 8:28:27 last year. The interesting part is where the time went. The run was not uniformly slow: Iden’s run course record from 2022 is 2:36:14, and he ran 2:36:58 after serving a penalty. The fastest bike splits, by contrast, were around 4:06 to 4:11, against 4:06:22 for Lange in 2024. A choppy swim and a hard, humid day redistributed the race, and it rewarded athletes who could still regulate their effort and heat load late in the marathon. On the Long Distance Triathlons Dashboard you can see this in the WBGT-versus-split plots: the run is where the environment exacts its price, and the top-three average hides how wide the spread becomes behind the leaders.

As I wrote before the race, the Rankings & Predictions page is a form-based heuristic built from the results I scrape from PTO Stats and the race-day conditions from Open-Meteo. It cannot know who has a vascular problem, who will be penalised, or whose thermoregulation will fail with a kilometre to go. It also has a structural blind spot: athletes moving up from short course, like Luis, have a thin long-distance history, so any model that rewards consistent long-course form will underrate them. The pre-race consensus had the same blind spot. Triathlete’s preview had Blummenfelt tied with Laidlow for the best odds, and neither made the podium. A model is a benchmark, not an oracle, and Kona remains the race where the benchmark is most often wrong.

Long Distance Triathlons Dashboard split-time trends
Long Distance Triathlons Dashboard rankings and predictions
The Long Distance Triathlons Dashboard: ironmandt100analysis.netlify.app

So, the next big one is Dubai T100. How is that looking?

Meanwhile on the roads: 2:01 in Chicago, 1:59 in London

The day after Kona, Chicago produced a very different kind of performance. Amos Kipruto won in 2:01:09, the second-fastest time in Chicago history and tied for fifth on the world all-time list, and Hawi Feysa defended her title in 2:13:33, moving to fourth on the women’s all-time list. Two weeks earlier in Berlin, Tigst Assefa ran 2:11:04, a course record and the third-fastest women’s marathon ever, and was still ahead of world-record pace at 40 km before her stride broke down in the final two kilometres. Guye Adola won the men’s race in a personal best of 2:02:50.

All of this sits in the shadow of the London Marathon in April, when Sabastian Sawe ran 1:59:30, the first sub-two-hour marathon in a race that counts for records, and Assefa lowered the women-only world record to 2:15:41, with the top three women all under 2:16 in the same race for the first time.

The common denominator, again, is footwear. Sawe, Yomif Kejelcha and Assefa all raced in the Adidas Adizero Adios Pro Evo 3, a shoe of around 97 g. After Chicago it has won seven of this year’s 12 Major races, and every runner on the Berlin podium wore it. The competition is not standing still: in Chicago, Sharon Lokedi was second in an Under Armour development shoe, and both Brigid Kosgei and Milkesa Mengesha took third in Nike’s unreleased Apex.

The most interesting question is not which shoe wins (it is always the athlete winning and never the shoe!) but how these shoes work, and the geometry is changing faster than the foam. Laurens van Kouwenhove has produced a superb animated analysis of Nike’s progression from the Alphafly 3 to the Alphafly 4 and the Apex, tracing the sole profiles from side-view product photos and a Nike patent drawing.

Rocker geometry of the Nike Alphafly 3, Alphafly 4 and Apex. Analysis and graphic: Laurens van Kouwenhove

Across the three shoes, Nike appears to be moving away from the steep, early rocker:

  • The forefoot apex moves back, from 67% to 64% to 61% of shoe length.
  • The rocker radius grows, from 45% to 54% to 63% of shoe length.
  • The rolling surface extends past the toes, from −7 mm to +2 mm to +8 mm.
  • The Apex is zero drop, against 8 mm for both Alphaflys.

The biomechanical consequence is captured in his estimate of the lever arm between the ground reaction force and the ankle, which is a proxy for the load on the calf and Achilles. Early in the roll-over (foot at 10°), the newer shoes shorten the lever arm: 132 mm for the Alphafly 3, against 121 mm and 125 mm. Late in push-off (foot at 45°), the order reverses: 142 mm, against 155 mm and 158 mm. In other words, the shoe appears to “hold on to the lever” for longer, shifting plantarflexor demand later in stance rather than simply reducing it.

Van Kouwenhove is careful to say these are indicative values from an unloaded, rigid-sole model. A real sole compresses, and the foot is not a rigid lever. But the direction of the change is plausible, and it matters. Much of the debate about super shoes has focused on energy return from foam and plates. Geometry that alters ankle moments could explain why benefits vary so much between runners, and why some athletes report calf and Achilles problems when changing models. A zero-drop shoe with a longer late lever arm is not a neutral swap for someone who has trained for years in an 8 mm drop.

This is also directly relevant to triathlon. In the last 10 km at Kona, calf fatigue and form breakdown are exactly where races were lost this year. How a shoe distributes load across a 2:40 to 3:00 marathon, run after five hours on the bike, is a question nobody has studied properly yet.

I want to be careful here, because “super shoes” have become a lazy explanation for everything. There is still no precise answer to how much the curved plate and resilient foam benefit the elite marathon runner, and the gains appear to vary between individuals. Pacing, fuelling at 90 to 120 g/h, deep Ethiopian and Kenyan talent pools and cool, flat courses all contribute.

But there is a more important point for this post. These times are produced in cool conditions. Kona showed what happens to the same physiology at 30 °C and high humidity. The next marginal gains in endurance sport may come less from returning energy at the foot and more from removing heat from the body.

Cooling clothing: from F1 cockpits to tennis courts and trails

This is where the sportswear industry has been moving fast in 2026, and two companies have taken quite different approaches.

adidas CLIMACOOL+ embossed cooling fabric worn on a grass tennis court
adidas CLIMACOOL+ in use during the 2026 grass-court season. Photo: adidas.

Fabric that changes the microclimate. In June adidas launched CLIMACOOL+, built around a 3D, body-mapped embossed fabric whose raised structures hold the garment off the skin so that air can circulate and sweat can evaporate. The structures are mapped to the chest, upper abdomen, shoulders and upper back, the areas identified through physiological data capture and climate chamber testing. Interestingly for the marathon story, adidas says Sawe ran his sub-two-hour marathon in a singlet featuring the technology.

Nike Radical AirFlow racing top with funnel-shaped airducts
Nike Radical AirFlow. Photo: Nike.

Nike’s Radical AirFlow is a nice piece of applied fluid dynamics. Inspired by the Venturi effect, the material uses funnel-shaped airducts that capture air as the athlete moves forward and accelerate it towards the skin to speed up sweat evaporation. In Nike’s climate chamber testing, prototypes retained 50% less sweat than Dri-FIT after prolonged running in the heat, and thermal manikin tests showed 25% lower resistance to the evaporation of sweat. Caleb Olson wore a prototype when he won the 2025 Western States 100 in the second-fastest time in the race’s history.

Active cooling between efforts. The second adidas initiative is quite different. The CLIMACOOL System was first developed for the Mercedes-AMG PETRONAS F1 drivers to cope with cockpit heat, then deployed across adidas partner federations at the 2026 FIFA World Cup, and has now been rolled out to tennis players in New York. It consists of a vest containing a gel frozen in portable freezer units, worn over normal kit, plus neck pieces for changeovers, set breaks and practice. Adidas reports a significant cooling effect after three minutes, and an effect lasting up to an hour after 10 to 15 minutes of use.

adidas CLIMACOOL System cooling vest and neck piece used by a tennis player
The CLIMACOOL System cooling vest and neck pieces. Photo: adidas.

A few observations from a physiologist’s point of view.

First, these are two different problems. Garments like CLIMACOOL+ and Radical AirFlow try to improve heat loss during exercise by increasing air movement and evaporation at the skin. Vests and neck collars are pre-cooling and between-bout cooling tools, which suit intermittent sports like tennis and football, or the warm-up before an endurance race. The evidence that both pre-cooling and per-cooling improve performance in the heat is reasonably solid (e.g., Bongers et al., Br J Sports Med, 2015). Ditlev stuffing ice into his cap at 30 km was per-cooling in its most basic form and something you should do (I do!) in every hot race no matter how you feel at every water station [keep the body as cool as possible].

Second, the environment limits what fabric can do. Evaporation depends on the vapour pressure gradient between skin and air. In dry heat, more airflow over wet skin is very effective. In Kona’s humid air the gradient is smaller, so the gain from any garment is smaller too, although increased air velocity still helps convective heat loss. Relative air speed also matters: these garments should work better at 20 km/h on the run than at a walk, and on a bike at 40 km/h the airflow is already high.

Third, the numbers so far are manufacturer data. Fifty per cent less sweat retained and 25% lower evaporative resistance are meaningful material properties, but they are not yet independent, peer-reviewed measurements of core temperature, heat storage or performance in athletes. I would very much like to see these garments tested in proper heat-stress protocols, ideally in the kind of humid conditions that undid Ditlev and Løvseth.

What Kona 2026 teaches us

The lesson from the Big Island is the same one the literature keeps telling us: in the heat, the winner is often the athlete whose system fails last, not the one with the biggest engine. Fuelling, heat acclimatisation, pacing and cooling interact, and failures in any one of them show up in the final hour. Luis and Knibb were not the strongest athletes on paper for the whole day, but they were still running when it mattered showing that their bodies and most of all their strategies worked. If you want to read more about Kona, there is a very interesting pre-print here from 3 excellent sports scientists (and triathletes): Heat, wind, and the calendar: why a warming climate makes its performance effects unidentifiable at the IRONMAN World Championship. They have updated it now to version 3 and have also created a very nice Ironman Hawaii Predictor for age groupers based on historical data as part of the research paper. The tool is available clinking on the link below:

https://ironman-hawaii-predictor.com

Meanwhile, on cool flat roads, the marathon keeps getting faster with every new generation of shoes, and as the rocker-geometry analysis shows, not only because of foam. As temperatures at major championships rise, I expect the next wave of innovation, and the next wave of controversy, to be about thermal technology. That makes it all the more important that brands publish their data and that we as scientists test these claims independently.

Race and results information from Slowtwitch, TRI247, Triathlete, Big Island Now, NBC Sports, FloTrack, Marathon Handbook and Find Your Edge. Product information from adidas and Nike press materials. Rocker-geometry analysis by Laurens van Kouwenhove.

Two runners pass a science and performance station; overlays show VO2, heart rate, pace, power, hydration, and core temperature data.
Data Analytics, Sports Science

Kona 2026: One Day, One Island – Data, Fuel and Heat at the IRONMAN World Championship

For the first time since 2019, the professional men and women race the IRONMAN World Championship on the same day in Kailua-Kona. It is also the final stop of the 2026 IRONMAN Pro Series, with 6,000 points on offer to each winner, so the race carries both a world title and a season title. The men start at 6:20 a.m. local time, and by the time the leaders reach the Energy Lab the conversation will, as always in Kona, be about heat, fuelling and who managed both best.

In this post I bring together three things: what the professional race data tell us, what the newest research says about carbohydrate intake during racing, and what we know about heat acclimatisation and the conditions expected on the Big Island.

The data: trends, conditions and predictions

To support this preview I built an interactive Long Distance Triathlons Dashboard, which brings together professional-only results from IRONMAN, IRONMAN 70.3 and T100 racing (sourced from PTO Stats) with race-day environmental data from Open-Meteo. A few design choices are worth explaining, because they matter for how the numbers should be read.

First, split-time trends use the average of each race’s top three finishers rather than the winner alone, with a ±1 SD band. A single winning time is a noisy signal; the top-three average, and the width of that band, tells you both how fast the front of the race was and how close the podium fight was. Second, conditions are matched to performance using WBGT (wet bulb globe temperature) rather than air temperature for the bike and run, since WBGT captures humidity and radiant load, which is exactly what makes a 30 °C day in Kona so different from a 30 °C day somewhere dry. Wind is averaged across the 07:00–16:00 window that covers professional race time, rather than taking a single daily gust. Third, the correlation plots use every individual finisher’s split against the conditions of that race, not race averages, so you can filter by brand and see how the relationship holds up.

The Rankings & Predictions page applies a simple form-based model to the start lists. It is deliberately a heuristic, not a crystal ball: it rewards consistency and recent results, and it cannot know who has been ill, who crashed in training, or who has nailed their heat preparation. It is, however, a useful benchmark against which to judge the race once it is run.

The historical benchmarks are formidable. Patrick Lange’s course record of 7:35:53 in 2024 was built on a 47:09 swim, 4:06:22 bike and 2:37:34 run, and the same race saw Sam Laidlow ride 3:57:22. Lange, now 40, has said this will be his last IRONMAN. In the women’s race, Solveig Løvseth won on her Kona debut in 2025 in 8:28:27, just 35 seconds ahead of Kat Matthews, who set a women’s run record of 2:47:23, with Laura Philipp third in 8:37:28. That race is remembered as much for Taylor Knibb, who was leading with around two miles to go before collapsing – a reminder that in Kona the margin between a world title and a medical tent is thin.

On the men’s side the defending champion Casper Stornes is out after a training crash, which opens the race up. Laidlow arrives unbeaten in four races in 2026, including Challenge Roth and an IRONMAN Lanzarote win by almost nine minutes. Kristian Blummenfelt leads the Pro Series standings, and Gustav Iden, Marten Van Riel (on his Kona debut), Magnus Ditlev and Lange complete a field with four former world champions. Independent modelling by Thorsten Radde at Trirating suggests the leading men share roughly three quarters of the win probability, against about 85% for the leading women – in other words, a surprise is more likely in the men’s race. That is consistent with what I see in the data: the women’s front-runners have more head-to-head history on this course and in these conditions.

Fuelling: how much carbohydrate is enough?

Few areas of endurance practice have moved as fast as in-race fuelling. The 2016 ACSM sports nutrition guidance recommends up to 90 g/h of carbohydrate from multiple transportable carbohydrates (glucose–fructose mixtures) for events longer than 2.5–3 hours. In the field, however, many professionals now report targets of 120 g/h and, in some cases, far higher. Two papers published this year help separate what is supported by evidence from what is enthusiasm.

Morton and colleagues’ review in the Journal of Nutrition (Morton et al., 2026) revisits more than a century of work on carbohydrate and endurance. Their conclusion is that the upper limit of recommended intake could reasonably move from 90 to 120 g/h for trained athletes, since both exogenous and whole-body carbohydrate oxidation can increase at these intakes, with possible benefits for durability and economy. They are explicit, though, that doses of 120–200 g/h seen in the field are not yet substantiated by research, and they propose a more individualised model that accounts for carbohydrate blend and ratio, format, gut training and environmental conditions.

A new study in Medicine & Science in Sports & Exercise (Martyn et al., 2026) tested this directly. Eight trained men cycled for three hours followed by a capacity test, ingesting 10, 90, 120 or 150 g/h (maltodextrin–fructose at 1:0.8, with stable-isotope tracers). Exogenous oxidation rose with dose (1.23, 1.48 and 1.62 g/min at 90, 120 and 150 g/h), but whole-body carbohydrate oxidation was identical at 120 and 150 g/h (2.6 g/min). Exercise capacity was clearly impaired at 10 g/h but did not differ among the higher intakes, and the modelled optimal dose was about 107 g/h (95% CI 92–122 g/h). Just as important, the between-athlete spread in exogenous oxidation was large and grew with dose.

For a Kona athlete the message is practical. Somewhere around 90–120 g/h is where the evidence currently sits; going beyond that is unlikely to add whole-body carbohydrate use and increases the volume of fluid and sugar the gut has to handle. In the heat this matters even more, because splanchnic blood flow is reduced and gastrointestinal distress becomes a major cause of underperformance. Recent work from Monash (Hillemacher et al., 2026) shows that symptom reports alone probably underestimate gut disturbance during exertional heat stress, and a survey at the 2025 Western States 100 (Mougin et al., 2026) found that every finisher under 18 hours had a structured carbohydrate plan, compared with 69% of the slowest finishers, with gastrointestinal symptoms reported by 63% of finishers. The athletes who fuel best on Saturday will be those who rehearsed their exact race-day intake, in the heat, many times beforehand.

Heat: acclimatisation and the Kona forecast

October in Kona typically means highs of 30–32 °C and humidity around 80%, with the lava fields and asphalt of the Queen K adding considerable radiant heat. The ocean is around 25–27 °C, so the swim is non-wetsuit. The National Weather Service forecast issued on Friday afternoon for Saturday calls for sunny conditions with isolated showers in the morning, becoming partly sunny with scattered showers in the afternoon, highs of 84–89 °F (29–32 °C) near the shore, and light winds becoming westerly up to 10 mph in the afternoon, with a 40% chance of rain. Light winds are good news for bike times but bad news for heat dissipation: less convective cooling on the bike and in the Energy Lab means a higher effective heat load, and afternoon cloud and showers could become the most important variable for those still running. On the dashboard you can see how WBGT and wind have related to bike and run splits across previous editions.

The physiology of heat acclimation is well established, but recent studies have refined the practical questions of dose, method and side benefits. A randomised trial in endurance-trained athletes (Snape et al., 2026) showed that just five days of isothermic acclimation in hot, humid conditions (32 °C, 70% RH) lowered resting heart rate by 8 bpm, expanded plasma volume by about 7% and increased sweat loss by around 500 mL during a heat stress test – the classic adaptations that protect cardiac output and thermoregulation on a long day. Renal stress markers were not altered, which is a useful reminder that acclimation improves tolerance but does not remove the physiological cost of exercise in the heat.

Method matters. In well-trained cyclists, six days of post-exercise hot water immersion reduced perceptual strain but did not improve time-trial performance in warm or hot conditions (Kjertakov et al., 2025), suggesting passive approaches are not a substitute for active exercise-heat exposure in already fit athletes. Interestingly, training in the cold while wearing layered clothing reduced perceived exertion and thermal sensation in subsequent heat exposure (Yamaguchi et al., 2026), although without measurable changes in sweat rate or core temperature.

Heat training is also increasingly used for its haematological effects, something the Norwegian group in this field has made very visible. In cross-country skiers and biathletes, adding three weekly heat-suit sessions to a moderate-altitude camp increased haemoglobin mass by 2.9%, compared with 1.2% with altitude alone (Rønnestad et al., 2026), and in elite cyclists five weeks of heat training produced haemoglobin mass gains comparable to a live-high train-high camp (Cubel et al., 2025). In neither study did this translate into clear performance gains, which is an honest caveat: the adaptations are real, but how much they matter on race day is still debated.

The key point, from both the literature and my own experience working in Doha, is that heat acclimatisation is not a single intervention but a process that interacts with everything else: hydration, sodium, cooling strategies, pacing and, crucially, fuelling. Heat increases carbohydrate dependence while simultaneously reducing the gut’s ability to deliver it. The athletes who arrived in Kona early, trained in the conditions and rehearsed their nutrition in that environment will have a real advantage, whatever their power numbers say.

What to watch

If the winds stay light, expect fast bike splits and a large group reaching T2 together, which shifts the race to the marathon and to whoever has managed heat and fuelling best through the first six hours. If the afternoon clouds and showers arrive, they may favour the chasers. And if history is any guide, the athlete leading with 10 km to go usually wins – unless, as Knibb discovered last year, the heat decides otherwise.

Explore the full data, filter by brand and race, and check the predictions at ironmandt100analysis.netlify.app. I will revisit the predictions once the results are in.

References

Cubel C, et al. Haematological adaptations to high-altitude and heat acclimation training in elite male cyclists. Exp Physiol. 2025. doi:10.1113/EP092968
Hillemacher K, et al. The relationship between biomarkers of exercise-induced gastrointestinal syndrome and exercise-associated gastrointestinal symptoms. Sports. 2026. doi:10.3390/sports14060248
Kjertakov M, et al. Limited benefit of repeated post-exercise hot water immersion on exercise performance of well-trained cyclists in warm and hot conditions. Eur J Appl Physiol. 2025. doi:10.1007/s00421-025-06041-4
Martyn HJ, et al. Effects of 10, 90, 120 and 150 g/h CHO on substrate metabolism, exogenous oxidation and exercise capacity: implications for CHO personalization. Med Sci Sports Exerc. 2026. doi:10.1249/MSS.0000000000004176
Morton JP, et al. From metabolism to medals: contemporary perspectives and revisiting carbohydrate guidelines for fueling endurance athletes during exercise. J Nutr. 2026;156(5):101442. doi:10.1016/j.tjnut.2026.101442
Mougin L, et al. Heat preparation and nutrition strategies for a 100-mile ultramarathon in extreme heat: a questionnaire study at the Western States Endurance Run. J Sci Med Sport. 2026. doi:10.1016/j.jsams.2026.05.014
Rønnestad BR, et al. Heat training augments hemoglobin mass during a moderate-altitude training camp in well-trained endurance athletes. J Appl Physiol. 2026;140(5):1298-1310. doi:10.1152/japplphysiol.00777.2025
Snape D, et al. Five days of heat acclimation improves cardiovascular and thermoregulatory responses without altering renal stress biomarkers in endurance athletes. PLoS One. 2026;21(8):e0345346. doi:10.1371/journal.pone.0345346
Yamaguchi K, et al. Short-term endurance training in the cold with layered clothing mitigates perceptual strain during exercise in the heat. Eur J Appl Physiol. 2026;126(10):5543-5552. doi:10.1007/s00421-026-06319-1

Race and forecast information from USA Triathlon, Triathlete, 220 Triathlon, Big Island Now and the US National Weather Service. Literature retrieved via PubMed.