How One Amateur Marathoner Used AI to Break Through a Training Plateau

Sonia had run two marathons. Her first finish time was 4:22. Her second, eighteen months later after a more structured build, was 4:19.

Three minutes of improvement over eighteen months wasn’t the trajectory she wanted. She was training consistently — 35 to 40 miles per week during builds — but the fitness gains weren’t showing up. She’d read enough about marathon training to understand periodization in theory. What she lacked was the practical system to apply it, week after week, around a demanding job and two young children.

She started using AI-assisted planning for her third marathon build. The result: a 4:04 finish. Fifteen minutes of improvement in one training cycle.

This is the story of what actually changed.

Baseline: What Was Working and What Wasn’t

Before the new training cycle began, Sonia ran a retrospective with AI. She provided her training log from the previous 18-week build: total mileage per week, session types, and a rough account of her intensity distribution.

The conversation surfaced two structural problems.

First, her long runs were too slow relative to her goal pace. She’d been running 11:30–12:00 per mile on long runs while targeting a 9:45/mile marathon pace. Research on marathon-specific adaptation suggests the long run should be run no more than 60–90 seconds per mile slower than goal marathon pace — she was running nearly two minutes slower, which built endurance but limited race-pace neuromuscular adaptation.

Second, her easy runs weren’t easy enough, and her tempo runs weren’t hard enough. When AI calculated her intensity distribution from the training log, roughly 55% of her weekly mileage was in the moderate zone — harder than easy but softer than genuine threshold work. She was spending most of her training in the zone that Stephen Seiler’s polarized training research identifies as the least productive for endurance adaptation.

These weren’t failures of effort. They were failures of structure. She was training hard enough to feel tired but not in a way that was generating the adaptation she needed.

The Macrocycle: Setting the Arc

Sonia’s target marathon was 18 weeks away. The macrocycle conversation produced the following structure:

Weeks 1–5 (Base block): Aerobic volume focus. Maximum 40 miles per week. No structured speedwork. All runs genuinely easy — heart rate under 140 bpm throughout. The goal was to rebuild aerobic infrastructure properly before adding intensity.

Weeks 6–10 (Threshold block): Two tempo or threshold sessions per week. One weekly long run at a pace no more than 90 seconds slower than goal marathon pace. Weekly mileage 40–46 miles.

Weeks 11–15 (Race-specific block): Marathon-pace work introduced. One weekly run containing sustained miles at goal pace. Long run with the final 4–6 miles at goal marathon pace. Weekly mileage maintained at 40–45 miles.

Week 16 (Cutback week): Planned recovery — drop to 28 miles, easy running only.

Weeks 17–18 (Taper): Volume reduces by 40% in week 17, 60% in race week. Intensity maintained with short, sharp sessions.

One important addition Sonia made: she booked two sessions with a running coach during the build — one in week 3 to review her form, one in week 10 to check her threshold mechanics before the race-specific block began. AI designed the schedule architecture. A human set of eyes confirmed her movement was sound.

Version 1: The Plan That Hit Real Life

By week 3, the base block was producing steady adaptation. Sonia was keeping her easy runs genuinely easy — harder than she expected, psychologically — and the polarized distribution was closer to 80% low intensity.

Then week 4 arrived. A work project required six consecutive 10-hour days. Her planned four training days shrank to two. Her long run — 14 miles — didn’t happen. She completed 18 miles for the week instead of 32.

In the old approach, this would have been the moment she went off-plan. She’d have tried to compress the missed volume into the next two weeks, created a mileage spike, run tired, and lost the adaptation she’d built.

Instead, she ran the Sunday microcycle check-in:

Week 4 was a work disaster. I planned 32 miles (Mon 8 easy, Wed 8 easy, Fri 6 easy, Sun 14 long). I got: Monday 10 easy, Wednesday 8 easy. That's it. Work absorbed everything else.

This week (week 5, the last week of base) I'm back to normal schedule. I'm in the base mesocycle — no speedwork, aerobic volume only.

Should I try to make up last week's lost mileage, or just run week 5 normally and accept the dip? The base block ends after this week.

AI’s response recommended against makeup volume. Making up 14 miles of long run in week 5 would have created an artificial mileage spike right before a phase transition — the wrong time to stress the body. Instead, it suggested running week 5 as planned and extending the base block by one week to allow the aerobic adaptation to consolidate before adding threshold work.

That one-week extension cost nothing. It maintained the base’s training effect. And it gave Sonia the mental reset of treating the disruption as a plan adjustment rather than a failure.

The Threshold Block: Discipline Under Pressure

The threshold mesocycle (weeks 6–10 in the revised plan, weeks 6–11 after the extension) introduced two threshold sessions per week.

Sonia had done tempo runs before, but not with precise intensity guidance. She’d been running what felt like “a good hard effort” — which, it turned out, was often moderate effort with a few genuinely hard miles mixed in. That’s grey-zone training dressed up as threshold work.

The threshold sessions in this block had a simple prescription: 20–25 minutes of continuous running at the pace she could sustain for approximately 60 minutes in a race (an accepted proxy for lactate threshold pace). No faster. No slower.

The weekly Intensity Ledger check-in became a useful enforcement mechanism:

This week's sessions:
- Tuesday: 8mi, 20-min tempo at 8:22/mi average (target was 8:15–8:25), rest easy Z1-2
- Thursday: 7mi easy, 90% Z1-2
- Saturday: 6mi easy, 85% Z1-2
- Sunday: 14mi long at 10:05/mi average

Please calculate my intensity distribution and flag if I'm drifting into grey zone.

Week over week, seeing the numbers kept her honest. When grey-zone drift appeared — usually on the long run when she pushed the pace out of enthusiasm — the Intensity Ledger caught it. AI’s framing was useful here: not a criticism, but a question. “Your long run pace dropped under 10:00 in miles 10–12 — was that intentional, or did pace drift under fatigue?”

That question, asked consistently, changed how she ran her long runs.

The Race-Specific Block: Sharpening the Ax

In weeks 12–16 (revised plan), the race-specific mesocycle introduced goal marathon-pace work. The structure:

• One weekly run with a sustained goal-pace segment (starting at 4 miles, building to 8 miles by week 15)
• One long run per week with the final 4–6 miles at goal marathon pace
• All other mileage genuinely easy

This was the block where AI helped most with session design. Sonia’s goal was 4:03 (9:18/mile). Designing sessions that built familiarity with that specific pace — not approximately that pace, not “good marathon effort,” but that pace — required specific prompting:

I'm in week 13 of my race-specific block. This Sunday I'm doing a 17-mile long run. I want to finish the last 5 miles at goal marathon pace (9:18/mile). Design the session structure: when to ease into goal pace, how to manage the miles before, what to do if I'm struggling at mile 14.

The conversation that followed included pacing strategy, decision-making protocols for the final miles, and a reminder that the purpose of the session is race-pace familiarity — arriving at mile 12 already tired and fighting to hold 9:18 is more valuable than a comfortable 10:00 long run.

Race Day

Sonia finished in 4:04:12.

The fifteen-minute improvement over her previous time wasn’t a product of training harder. She’d been training hard for two years. It was a product of training with better structure and more responsive adaptation — a plan that survived contact with real life rather than collapsing when work intervened.

Three things she credits:

1. The weekly microcycle check-in — 10 minutes every Sunday that kept the adaptation loop closed.
2. The Intensity Ledger — making grey-zone drift visible changed her behavior at easy efforts and long runs.
3. The two sessions with a human coach — her form had a subtle crossover gate that was contributing to hip flexor strain. That caught early prevented what could have been an injury during the race-specific block.

AI handled the first two. The coach handled the third.

Beyond Time was what Sonia used to block-protect her training sessions on her calendar — treating each run as a committed appointment rather than an item that could be bumped. That structural commitment to her schedule was the foundation on which the planning system ran.

The Lessons

A plan that adapts is better than a plan that’s perfect. The one-week base extension after the work disruption was a small change that maintained the training effect. A static plan would have demanded she press on with a threshold block on an undertrained base.

The Intensity Ledger changed behavior faster than any advice. Seeing the numbers weekly, watching grey-zone drift appear in her own data, produced the behavior change that knowing the theory had not.

AI and human expertise are complementary, not competing. The macro and micro planning was AI’s domain. The form assessment was a human’s domain. Neither replaced the other.

The plateau wasn’t a fitness problem. It was a planning problem — and a planning tool fixed it.

Related:

• The Complete Guide to AI Planning for Athletes
• The Athlete AI Planning Framework
• How Amateur Athletes Use AI Planning
• Why AI Cannot Replace a Coach
• Measuring Goal Progress with AI

Tags: marathon training AI, amateur marathon case study, AI periodization, training plateau, AI running plan