Miles & Machines | Training with an AI Running Coach

Category: Cologne Half Marathon 2016

  • The Cleanest Tempo Run So Far

    After several weeks dominated by heat adjustments, this tempo run finally took place in pleasant morning conditions. The plan called for a two kilometre warm-up, a sustained tempo block of approximately six kilometres and an easy cool-down.

    The result was the cleanest quality session of the project so far.

    The data

    MetricResult
    Distance10.0 km
    Time59:21
    Average pace5:56/km
    Average heart rate137 bpm
    Maximum heart rate153 bpm
    Average power256 W
    Maximum power298 W
    Elevation gain12 m
    Perceived effort6 to 7/10
    Faster blockApproximately 5.5 to 6.0 km
    Food before the runOne banana

    The overall pace includes the warm-up and cool-down, so it should not be confused with the pace of the tempo block. The training record confirms a steady faster section but does not preserve a complete kilometre-by-kilometre split table.

    Average heart rate was unusually low for a quality day, while maximum heart rate reached 153 bpm. Together with an effort rating of 6 to 7, that suggests a controlled sustained session rather than a threshold test.

    What the coach saw

    ChatGPT described the run as very clean. The flat route and cooler morning reduced the environmental noise that had complicated earlier comparisons.

    There was no need for a final acceleration. The value came from holding the faster block evenly and finishing with reserve. Average power of 256 W was higher than on the recent easy runs but remained stable enough to support the impression of control.

    The result also supported a more ambitious current form estimate. Training was beginning to suggest that a half marathon below 1:55 was realistic, with a possible range around 1:52 to 1:54 if endurance continued to develop.

    That was still a working estimate, not a race prediction. The tempo block was shorter than the race, and several specific sessions remained before October.

    The next-day decision

    Sunday’s long run had to stay easy. A good tempo session is useful only if the following training absorbs it. There would be no fast finish and no attempt to turn the weekend into two consecutive tests.

  • Easy Really Means Easy at Thirty-Three Degrees

    The hottest easy run of the project took place at 33°C. The instruction was simple: forget pace and remain in the aerobic range.

    That sounds obvious, but hot weather makes easy runs vulnerable to two bad decisions. I could try to preserve a normal pace and let the effort climb, or slow down and mistake the slower number for lost fitness.

    This time I chose the second pace and the correct effort.

    The data

    MetricResult
    DistanceApproximately 8.0 km
    Time55:22
    Average pace6:54/km
    Average heart rate130 bpm
    Maximum heart rate149 bpm
    Average power226 W
    Elevation gain88 m
    Temperature33°C
    Perceived effort2 to 3/10
    RouteMostly or partly shaded

    The pace was the slowest of the recent easy runs, but the route included 88 metres of elevation and the temperature was extreme. Average heart rate remained within the intended range.

    What the coach saw

    ChatGPT considered the session exactly what an easy run should be. The low average power, controlled heart rate and low perceived effort all pointed in the same direction.

    The maximum heart rate of 149 bpm did not change that assessment. Short hills and hotter exposed sections can produce brief rises without turning the entire run into a moderate workout.

    Most importantly, I did not use the pace display to negotiate against the purpose of the day. Running at 6:54/km in these conditions was not a regression. It was the pace required to keep the effort easy.

    Why this mattered for the weekend

    Saturday included a sustained tempo block. Preserving energy on Thursday meant that the quality session could be judged on its own rather than on fatigue created by an unnecessarily quick recovery run.

    The experiment was beginning to produce a simple hierarchy. On interval and tempo days, pace can help define the work. On easy days, heart rate and perceived effort have priority. In extreme heat, that distinction matters even more.

  • Controlled Intervals on Another Hot Evening

    The next interval session arrived on another hot evening. The planned structure was five 1,000 metre repetitions with jogging recoveries, but the main instruction was more important than the target pace: keep the efforts controlled and let the weather influence the speed.

    At 29°C, this was not the evening to chase the fastest split of the training block.

    The data

    MetricResult
    DistanceApproximately 10.2 km
    Time1:02:38
    Average pace6:09/km
    Average heart rate142 bpm
    Maximum heart rate166 bpm
    Average power248 W
    Elevation gain53 m
    Temperature29°C, evening
    Perceived effort6 to 7/10

    The faster sections felt controlled, and heart rate fell clearly during the recoveries. That recovery pattern mattered more than whether every repetition matched an exact pace.

    The summary record does not preserve a separately confirmed list of interval splits, so this post cannot make precise claims about the consistency of all five repetitions. The reliable conclusion is that the workout structure was completed without the final effort becoming maximal.

    What the coach saw

    ChatGPT rated the session as a successful progression from the previous week. The total distance returned above ten kilometres, maximum heart rate stayed below the planned ceiling and perceived effort remained at 6 to 7 rather than drifting to 9 or 10.

    The average values are not especially useful on their own because warm-up, recoveries and cool-down are included. The clearer signals were the repeated rises and falls in heart rate and the absence of a late loss of control.

    The heat adjustment also worked. A fixed pace target could have turned the final repetitions into a race. Using effort and heart rate kept the training stimulus without adding avoidable exhaustion.

    What changed next

    Thursday stayed completely easy, with pace treated as irrelevant. The weekend still contained a tempo run followed by a long run, so there was no reason to make the recovery run faster.

    This session did not produce a dramatic breakthrough. It showed something more useful for half marathon training: a hard workout can finish with the feeling that another repetition might have been possible.

  • A Fast Long Run Without Fuel

    The coaching plan deliberately reduced this long run to 16 or 17 kilometres. The previous Sunday had reached approximately 19 kilometres, and this week was intended to absorb the accumulated load before training increased again.

    The proposed Friday tempo session was not recorded as completed, so Sunday followed two days without running. The goal remained a controlled long run with reserve, not an attempt to compensate for the missing quality session.

    The run became the quickest long session of the project so far. It also repeated a mistake that should already have been familiar: I forgot the carbohydrates.

    The data

    MetricResult
    DistanceApproximately 17.0 km
    Time1:46:46
    Average pace6:16/km
    Average heart rate134 bpm
    Maximum heart rate159 bpm
    Average power247 W
    Elevation gain217 m
    Perceived effort6/10
    FluidApproximately 0.7 litres
    Carbohydrate during the runNone; forgotten

    The first half was cooler and more shaded. I kept some reserve and finished with the feeling that another two or three kilometres would have been possible.

    The maximum heart rate was higher than the average suggests, most likely during harder sections of the hilly route. The overall average remained controlled.

    What the coach saw

    ChatGPT saw a strong endurance result. The pace was ten seconds per kilometre faster than the 19 kilometre run one week earlier, while average heart rate was identical at 134 bpm.

    That comparison is encouraging, but it needs context. This run was shorter, the early conditions were more favourable and I had skipped Saturday’s tempo session. It is evidence of good current fitness, not proof of a sudden transformation.

    The additional recovery probably helped the long run. More importantly, I finished with reserve instead of forcing an end acceleration.

    Why 17 kilometres were enough

    Stopping at approximately 17 kilometres was part of the plan. The progression was not supposed to continue mechanically from 17 to 18 to 19 to 20 kilometres every weekend.

    The coaching logic was a simple cycle of load, load, recovery and then the next training step. Ending this run with another two or three kilometres available showed that the reduced week had worked. There was nothing to gain from turning the final reserve into extra distance.

    The repeated mistake

    For a run of almost one hour and 47 minutes, forgetting carbohydrate was unnecessary. I had already learned after the July interval session that demanding training and poor fuelling do not combine well.

    Nothing dramatic happened this time, but the absence of a crisis is not the same as good preparation. Future long runs were supposed to include carbohydrates from roughly 45 to 55 minutes onward so that the routine would be familiar before race day.

    The run strengthened the performance trend. The planned reduction showed the value of recovery, while the forgotten fuel showed that execution outside the running itself still needed work.

  • The Easy Run That Did Its Job

    The easy run after Tuesday’s heat intervals looked exactly as restrained as it was supposed to look. Heart rate stayed low, power remained well below the quality sessions and the pace matched the warm conditions.

    It was relaxed. It also did not feel especially enjoyable.

    That is not a contradiction. Some recovery runs are memorable because the legs feel light. Others are useful precisely because nothing much happens.

    The data

    MetricResult
    DistanceApproximately 8.8 km
    Time59:03
    Average pace6:42/km
    Average heart rate127 bpm
    Maximum heart rate146 bpm
    Average power229 W
    Maximum power371 W
    Elevation gain47 m
    Temperature27°C
    Perceived effort3/10
    How it feltRelaxed, but not much fun

    The heart rate stayed stable through the run. There was no sign in the summary that Tuesday’s interval session had turned this into a hidden moderate workout.

    What the coach saw

    ChatGPT considered the run exactly right for a recovery day. An average heart rate of 127 bpm kept it clearly in the aerobic range, while 6:42/km was entirely appropriate at 27°C.

    The lack of excitement was not treated as a warning sign. Easy running is not required to feel entertaining. The purpose was to absorb the interval session and remain ready for the next useful training stimulus.

    The coach also noted a positive trend: average heart rate on comparable easy runs had moved from approximately 134 to 128 and then 127 bpm. Weather and daily variation make that an imperfect comparison, but the direction was encouraging.

    A scheduling decision

    I could not run on Saturday and asked whether the planned tempo session could move to Friday. Based on the low effort and stable heart rate, ChatGPT initially approved the change with a shorter structure:

    • 2 km easy warm-up
    • 5 to 6 km at tempo effort
    • 2 km easy cool-down

    The pace would depend on temperature, with the harder end of the range avoided if it remained above 25°C.

    No completed tempo activity was later recorded in the coaching chat. The next documented session was Sunday’s long run. This journal should not invent a reason that the source record does not contain.

    The easy run still did its job. It supplied low-cost aerobic volume and left the scheduling choice open without pretending that every planned workout was ultimately completed.

  • Four Repeats in Thirty-Degree Heat

    Two days after the long run I was not supposed to do, the plan returned to interval work. The original session called for six 1,000 metre repetitions, but the conditions and the accumulated load made that target too ambitious.

    It was 30°C and humid. The sensible version of the workout was four repetitions, with the emphasis on control rather than completing the number written in the first draft.

    The data

    MetricResult
    DistanceApproximately 8.5 km
    Time52:08
    Average pace6:08/km
    Average heart rate140 bpm
    Maximum heart rate168 bpm
    Average power249 W
    Maximum power399 W
    Elevation gain43 m
    Temperature30°C, humid
    Perceived effort7/10

    The fastest recorded section was approximately 5:14/km. The complete split list is not available in the training record, so there is no value in inventing a neat progression that may not have happened.

    What is clear is that the four faster blocks were completed and the overall session stayed below an all-out effort.

    What the coach saw

    ChatGPT treated the shorter session as the correct adjustment, not as a failed attempt at six repetitions. Heat raises the cardiovascular cost of the same pace, and the 19 kilometre run on Sunday was still close enough to matter.

    The maximum heart rate of 168 bpm showed that the workout still reached the intended intensity. Adding two more repetitions simply to satisfy the original plan would have increased fatigue without proving anything useful.

    The important decision was to preserve the purpose of the session: several controlled faster efforts, recoveries that allowed the heart rate to fall and enough restraint to train again later in the week.

    What changed next

    Thursday remained an easy run with no strides. The goal was to see whether the legs recovered normally after the heat session.

    The Saturday tempo run was no longer automatic. It would depend on how the easy run felt, not just whether its average heart rate looked good.

    This was a small but important correction to the coaching process. A plan should define the training stimulus. It should not turn the original repetition count into a test of obedience.

  • The Run I Was Not Supposed to Do

    This post needs to begin with the plan I was given, because I did not follow it.

    I had received a vaccination on Friday. My doctor advised 48 hours without training. ChatGPT agreed and recommended cancelling Saturday’s tempo session. If I felt completely well after the full 48 hours, Sunday was supposed to be a very easy run of 45 to 60 minutes, roughly 7 to 9 kilometres.

    Instead, I ran approximately 19 kilometres.

    The data

    MetricResult
    DistanceApproximately 19.0 km
    Time2:02:27
    Average pace6:26/km
    Average heart rate134 bpm
    Maximum heart rate148 bpm
    Average power242 W
    Elevation gain215 m
    Temperature27°C
    Perceived effort5 to 6/10
    How it feltEnjoyable

    Much of the route was shaded and ran through the woods. I tried to keep the effort steady rather than the pace. I slowed down on climbs and exposed sections, then ran faster downhill or in the shade.

    The run felt good and the numbers were strong.

    What the coach saw

    Compared with the previous long run, I ran around nine seconds per kilometre faster at slightly higher power. Average heart rate rose from 128 to 134 bpm, which was plausible given the quicker pace, the temperature and the recent vaccination.

    MetricPrevious long runThis run
    Time2:03:522:02:27
    Average pace6:35/km6:26/km
    Average heart rate128 bpm134 bpm
    Average power236 W242 W
    Elevation gain250 m215 m

    There was no obvious late collapse. My heart rate remained controlled, and I finished feeling positive.

    A good run does not make the decision good

    It would be easy to use the result as proof that running long was fine. That would be the wrong lesson.

    The recommendation had been deliberately conservative because vaccine reactions can appear later, and because unusual symptoms during exercise should not be ignored. I asked for advice, received a clear answer and then exceeded the suggested distance by more than double.

    The fact that nothing went wrong does not make that a sensible application of the coaching process.

    This is exactly why I promised to document deviations rather than rewriting the story afterwards.

    What came next

    The following day was set aside for complete recovery. The next quality session would depend on normal resting values, no vaccine symptoms and fresh legs. If any of those conditions were missing, the intervals would be replaced by an easy run.

    From a running perspective, this was the strongest long session so far. From a decision making perspective, it was the clearest failure of the experiment.

  • An Easy Run at 30 Degrees

    Two days after the interval session, I went out for an easy run. It was 30°C, although the sky was cloudy, which made the conditions slightly more manageable than direct sun.

    The aim was recovery, not another test of fitness.

    The data

    MetricResult
    DistanceApproximately 8.0 km
    Time52:24
    Average pace6:32/km
    Average heart rate128 bpm
    Maximum heart rate139 bpm
    Average power230 W
    Elevation gain46 m
    Temperature30°C, cloudy
    Perceived effort4/10

    Despite the temperature, average heart rate stayed low. The short drops towards 90 bpm in the recording were most likely sensor errors or brief interruptions rather than real physiological changes.

    What the coach saw

    ChatGPT considered this a successful aerobic run. The power was clearly below the harder sessions, and the heart rate remained in the intended easy range.

    My effort rating of 4 out of 10 was slightly higher than ideal for a recovery run, but the heat offered a reasonable explanation. I had not forced the pace, and there was no sign that Tuesday’s interval session had left me unable to run easily.

    The initial weekend plan was a tempo session on Saturday followed by a long run on Sunday. The coach reduced the tempo target because of the heat and my earlier nausea. It also placed more emphasis on eating before the session and carrying enough fluid during the long run.

    Then the plan changed

    On Friday, I received a vaccination. My doctor advised me to take 48 hours off training.

    ChatGPT’s recommendation was clear. The Saturday tempo run should be cancelled. After the full 48 hours, Sunday should be limited to 45 to 60 minutes of very easy running, and only if I felt completely well.

    That meant the original four run week was no longer relevant. Missing one quality session would not harm the preparation. Training hard during a possible vaccine reaction could.

    The next decision was no longer about pace or fitness. It was about whether I would actually follow the safety advice I had asked for.

  • Five Repeats, 28 Degrees and a Lesson About Fuel

    The next quality session consisted of five 1,000 metre repetitions with jogging recoveries.

    The temperature was 28°C. I completed all five repetitions and rated the session 7.5 out of 10. It was hard, but not an all out effort.

    The data

    MetricResult
    DistanceApproximately 10.3 km
    Time1:01:39
    Average pace5:59/km
    Average heart rate142 bpm
    Maximum heart rate167 bpm
    Average power253 W
    Maximum power346 W
    Elevation gain9 m
    Temperature28°C
    Perceived effort7.5/10

    The five faster blocks were clearly visible in the heart rate and power data. Heart rate climbed into the 150 to 165 bpm range during the repetitions and dropped during the recoveries.

    There was no dramatic cardiovascular collapse. The final repetition was difficult, but the overall structure remained intact.

    What the coach saw

    ChatGPT considered the workout a successful quality session. Completing all five repetitions with jogging recoveries created a higher training density than walking or stopping between efforts.

    The heat was again treated as an additional load. The coach did not interpret the average pace as a direct measure of what I could do in cooler conditions.

    The next run was reduced to a genuinely easy 8 kilometres. The planned tempo run later in the week was also capped rather than extended.

    The part the workout data did not explain

    About an hour after finishing, I began to feel nauseous. I had eaten very little before the session.

    The likely combination was heat, fluid and salt loss, and insufficient energy intake. I cooled down indoors, drank slowly and ate a small amount of easily digestible food.

    This became a practical lesson for future quality sessions. Hard running in hot weather is not the right time to test how little food I can get away with.

    The coach recommended a small carbohydrate based meal or snack before demanding workouts, followed by fluids, carbohydrate and protein afterwards.

    The interval session itself was good. The preparation around it was not.

  • A Long Run on Tired Legs

    Less than a day after the tempo run, I started the longest session of the project so far.

    The instruction was simple. Keep it easy, ignore pace on the hills and do not add a faster finish. I walked the steeper climbs and tried to keep the effort steady.

    The data

    MetricResult
    DistanceApproximately 18.8 km
    Time2:03:52
    Average pace6:35/km
    Average heart rate128 bpm
    Maximum heart rate143 bpm
    Average power236 W
    Elevation gain250 m
    Temperature26°C
    Perceived effort7/10
    Fuel and fluid500 ml water, one gel at km 10

    The average heart rate was remarkably low for a run of more than two hours on a hilly route. It rose gradually from around 120 bpm early in the run to somewhere near 135 to 140 late in the session, but there was no sharp spike.

    What the coach saw

    ChatGPT rated the effort management very highly. Walking the climbs kept the session aerobic, and the stable power suggested that I did not suffer a major late collapse.

    The contrast between the objective and subjective data was interesting. My heart rate looked very comfortable, but I rated the run 7 out of 10. The previous day’s tempo session probably contributed to that difference.

    This is exactly why perceived effort belongs in the log. Heart rate alone might have made the run look almost effortless. It was not.

    Fuel practice

    I drank 500 ml of water and took a gel at kilometre 10. The gel contained around 21 grams of carbohydrate.

    That was enough to finish, but the coach recommended taking more during future runs lasting over two hours. The proposed range was roughly 40 to 50 grams of carbohydrate per hour.

    That recommendation sounded high compared with what I was used to, so it would need to be tested gradually rather than introduced all at once.

    The week in context

    This completed a four run week with intervals, an easy run, a tempo session and a long run. It was the strongest week of the preparation so far.

    The coach planned another interval session for Tuesday, but with an important condition. A fifth repetition should only be added if I felt properly recovered.