Tag: ai coaching
-
Smarter Recovery: How AFitPilot Now Knows When You Actually Need a Deload
Deload weeks only work if there’s fatigue to recover from. This post explains how AFitPilot moved beyond calendar-driven recovery and built a Smart Deload Guard that delays rest weeks until they’re actually earned.
-
Why Afitpilot Started Giving Sissy Workouts
Afitpilot didn’t start giving bad workouts because the AI was dumb. It happened because decision authority was misplaced. This post explains how safety-first defaults, missing memory, and unclear constraints caused underwhelming sessions—and what changed to fix it.
-
Why Your AI Training Plan Feels Generic (And How We’re Fixing It)
Most AI training plans aren’t wrong. They’re just polite, conservative, and forgettable. This post explains why generic input produces generic output, why “baseline week” fails trained athletes, and how Afitpilot is rebuilding its system around entry state, tension, and real-world decision-making.
-
Does Free-Text Workout Generation Actually Improve LLM Output Quality?
Does forcing an AI to output strict JSON actually make training plans better? In AFitPilot, we found the opposite. By shifting workout generation to free-text markdown and reducing prompt complexity, session quality, clarity, and coaching nuance improved measurably. Here’s why fewer constraints lead to smarter AI coaching.
-
Ya Filthy Clanker!
People don’t really fear AI coaching. They fear systems with no accountability. This post explains why fully autonomous AI is a bad idea, what AI is actually good at, and how Afitpilot keeps humans firmly in control.
-
💻 Local LLM Limitations: Why They’re Not Ready for Real Logic
Local LLMs are getting faster, cheaper, and easier to run — but they still fall apart when the task needs real logic, structure, or adaptation. I tested the most hyped models head-to-head. Here’s what worked, what broke, and why hybrid AI stacks are still the way forward.