How Your Plan Is Built
Why the structure comes from code, not AI — and how every session connects back to your specific gaps.
The plan is built in stages, and the key thing to know is: the structural decisions are made by code, not by AI.
- 1Your goal is translated into training dimension targets (what fitness do you need?)
- 2Your current fitness is measured against those targets (where are your gaps?)
- 3The training architecture is determined — phase, session types, intensity distribution, how to allocate your available days. This is fully rule-based.
- 4Sessions are assigned to days with specific power targets, durations, and rep structures. Pure computation.
- 5The AI writes coaching narratives — workout names, descriptions, "why this session matters" explanations. It does NOT choose sessions, intensities, or structure.
About 85% of the plan is deterministic. The AI explains the decisions; it doesn't make them. This is deliberate — LLMs are great at explaining training but unreliable at computing zone boundaries or distributing weekly load.
Each block is 3 weeks of progressive loading followed by 1 recovery week at 60% volume.
The critical design choice: only one thing gets harder each block. Either your long ride gets longer (+15 min/week), or your sweet spot duration grows, or your threshold intervals get longer, or your VO2 work increases. Never multiple at once.
Why? It's what good human coaches do. Stressing one system while others consolidate prevents the "everything is hard and nothing is adapting" spiral. It also makes it clear what's working — if you progress on the lever, the stimulus is landing.
Recovery weeks are real recovery: sessions capped at 75 minutes, no intensity, and some sessions are marked droppable so you can take extra rest without feeling like you're falling behind.
Three factors:
Your gaps. If your FTP is close to target but your longest ride is way shorter than your event, durability gets the focus. The biggest gaps get the most attention.
Your event. Mountain events emphasize durability and VO2max. Flat events emphasize threshold. Long events emphasize aerobic efficiency. The event profile sets the baseline importance of each dimension.
Your timeline. Early in training (12+ weeks out), aerobic base and durability get extra weight — build the foundation first. As your event approaches, threshold and VO2max get more emphasis. If you're close to your event but still have a large aerobic gap, Steady will keep you in base-building rather than prematurely adding intensity. Building speed on a weak aerobic foundation doesn't work.
Zones are anchored to your FTP and (when available) your aerobic threshold:
| Zone | Range |
|---|---|
| Easy | 45–55% FTP |
| Endurance | 55% FTP → LT1 (or 75% FTP if LT1 unknown) |
| Tempo | Endurance ceiling → 87% FTP |
| Sweet Spot | 87–94% FTP |
| Threshold | 94–105% FTP |
| VO2max | 105–120% FTP |
When Steady has a confident LT1 estimate, it directly caps the endurance zone. This matters because many cyclists ride "Zone 2" at a power that's actually above their aerobic threshold — defeating the purpose. Most apps just use a fixed FTP percentage; Steady uses your actual physiology when it can.
Power targets also progress within each block — starting at the lower end of the zone in week 1 and moving toward the upper end by week 3. Recovery week drops back down.
All targets are stored as percentages of FTP, so when your FTP changes, every workout auto-adjusts without restructuring the plan.
A template plan assumes everyone needs the same training at the same time. That's not how fitness works.
If your FTP is strong but you've never ridden over 3 hours, you need durability work — not more threshold. If you can ride all day but can't push above tempo, you need threshold and VO2max. If you're preparing for a mountain granfondo but your 5-minute power is low relative to FTP, you need VO2 work for the climbs.
Steady measures each dimension, compares it to what your goal demands, and builds the plan around the gaps. Every session has a reason tied to your specific profile.
This is also why LLMs alone can't reliably build training plans. Ask an AI to "generate a 4-week plan for a granfondo" and you'll get a different structure every time, with no guarantee it addresses your actual weaknesses. Steady's structural decisions are deterministic and computed — the AI explains the reasoning in plain language, but doesn't make the training decisions.
See your plan in action
Connect Strava and Steady will build a training plan around your specific gaps — every session tied to what your goal demands.
Try Steady Free for 14 Days