Functional Threshold Power (FTP) is a handy proxy for how hard you can ride for roughly an hour. Raise FTP and most rides—from group efforts to long climbs—feel easier or faster. Generic plans can help, but they often miss the small day‑to‑day adjustments that make the biggest difference. That’s where AI‑driven tools, powered by Large Language Models (LLMs), can speed up progress while keeping training sustainable.

Why generic plans plateau

• One size fits many: Progressions assume an “average” athlete, not your schedule, recovery, or history.
• Rigid calendars: Miss a key workout? The plan usually keeps marching, even if that breaks the logic of the week.
• Guessy intensities: Zones are broad. If your benchmarks are old or conditions change, targets drift.
• No context: Travel, sleep, and stress rarely translate into intelligent changes.

Where AI actually helps FTP

• Personalized starting point: Uses your recent rides and benchmarks (FTP, threshold pace/HR, heart rate drift) to set realistic targets—neither undercooked nor heroic.
• Adaptive progression: Progresses interval duration, density, and intensity based on execution quality (e.g., power stability, decoupling) and how you report the session felt.
• Microcycle tuning: If you miss a day, feel flat, or travel pops up, the week reshapes to preserve the stimulus that drives threshold gains.
• Anchored intensities: Converts goals like “sweet spot” or “over‑unders” into power/pace/HR you can hold on the day—indoors or outdoors.
• Early course‑correction: Flags “too easy” or “too hot” workouts by comparing power with HR and RPE, then adjusts the next session, not next month.
• Recovery and fueling nudges: Small reminders tied to session load help you absorb work and come back ready.

Why this improves FTP

• Progressive overload: Threshold improves when you add stress gradually—more time near the boundary where lactate balance, ventilation, and muscular endurance are challenged.
• Time in the right zones: Sweet spot and threshold work are effective, especially when balanced with plenty of low‑intensity volume to keep fatigue manageable.
• Distribution matters: Many athletes do best with a polarized or pyramidal mix (lots of easy; a moderate amount near threshold; a little very hard), adjusted to the season.
• Individual response varies: Some respond quickly to longer steady efforts; others thrive on over‑unders or broken intervals. AI can “notice” which sessions you complete well and lean into them.

A practical blueprint AI might produce

• 2–3 quality bike sessions/week (rotate focus by block):
  • Week 1–3: Sweet spot repeats (e.g., 3×12–15 min @ 88–92% FTP) + endurance long ride.
  • Week 4–6: Threshold building (e.g., 3×10–12 min @ 95–100%) or over‑unders around FTP.
  • Week 7: Deload (reduced volume and intensity) before the next block.
• Progression knobs: Add minutes per interval, reduce recovery, or add a rep—one knob at a time.
• Outdoor translation: For hilly routes, target average power for the effort with cadence and RPE anchors; avoid “spiking” that turns threshold into VO₂.
• Strength and durability: Short neuromuscular sprints in endurance rides and basic strength keep mechanics efficient.

None of this is magic; the advantage is timely, consistent adjustments when life happens.

How to work with an AI coach for faster FTP gains

• Share benchmarks: Provide FTP or an estimate, plus heart rate thresholds if you have them.
• Log RPE: A simple 1–10 feeling score helps adjust next sessions.
• Note constraints: Weekly hours, preferred days, equipment access, and travel.
• Ask for specific changes: “Make Tuesday easier,” “Swap long ride to Sunday,” or “Try longer intervals next week.”
• Retest on a rhythm: Every 4–6 weeks, or use trend‑based checks (e.g., a steady 30–40‑minute effort) to recalibrate without an all‑out test.

Common pitfalls AI helps avoid

• Too many hard days: Piling intensity raises fatigue faster than fitness; AI spaces quality to protect consistency.
• Progressing on two fronts: Increasing duration and intensity at once backfires; adjust a single variable.
• Monotony: Repeating the same session stalls adaptations; rotate over‑unders, steady threshold, and sweet spot blocks.
• Outdated zones: When fitness moves, targets shift with it instead of trailing by weeks.

Limits and safety

• AI isn’t a doctor. Pain, illness, or injury require medical guidance.
• Consistency, sleep, and fueling still do most of the work—AI just helps you point that work better.
• Sparse data means conservative changes until the system “learns” how you respond.

The bottom line

AI and LLMs don’t invent new physiology—they help you apply solid training principles to your real life, in real time. For most triathletes, that means getting to more high‑quality threshold work with fewer detours, and raising FTP faster than a one‑size‑fits‑all plan.

Curious where you stand right now? Try the Watts‑per‑Kilogram Calculator.