In 2022 I was 10 weeks into training for my fourth marathon. I was following a popular 18-week plan — one of those fixed PDF schedules you download from a running website. Life got busy, I missed a couple of weeks, then tried to cram the missed mileage back in. By week 12 I had shin splints that put me out for six weeks.
I missed the race. Not because I am a bad runner — because the plan had no way to adapt when my life changed. A good human coach would have spotted the warning signs and adjusted the schedule. The PDF just sat there.
After recovering, I hired an online running coach for three months. The difference was dramatic. They looked at my Strava data every week, changed sessions based on my fatigue levels, and explained the reasoning behind every adjustment. My next marathon was a 14-minute PB.
I started running in 2019, initially just a couple of 5Ks a week to manage work stress. My first marathon was a 4:41 suffer-fest that convinced me I was terrible at running. Three years and a lot of reading later, I ran my current PB of 3:28 — a 73-minute improvement, achieved entirely by training smarter rather than training harder.
That transformation came from understanding the science: ACWR and training load, polarised 80/20 intensity distribution, proper periodisation with step-back weeks, and paying attention to HRV and sleep data. None of this is secret knowledge — it is what elite coaches have applied for decades. It just was not accessible to recreational runners at scale.
I have worked as a software engineer for over a decade, with a focus on data systems and machine learning applications. When large language models became capable enough to reason about multi-variable problems — not just generate text — I saw an opportunity to build something that could genuinely replace a spreadsheet-following coaching bot with an actual thinking system.
PacecraftAI is built on Claude (Anthropic) as the reasoning core, with a multi-agent architecture that routes your messages to specialised coaching, analysis, planning, and scheduling agents. The system reads your actual training data before responding — it does not improvise from a script.
Everything on the PacecraftAI blog is grounded in peer-reviewed sports science. I reference Tim Gabbett's ACWR research, Stephen Seiler's polarised training work, and the same periodisation literature that underpins the system prompts in PacecraftAI's coaching agents. If I make a claim about training, it has a citation behind it.
I also write from experience: every injury risk pattern I describe, I have either experienced personally or observed across the runners who tested PacecraftAI during development. The methodology page explains in detail how the coaching system works.
I keep my full identity private by choice — the same way many independent builders do. What matters is whether the coaching system works and whether the advice is grounded in evidence. You can evaluate both of those things directly: try the free tier, read the methodology, check the references.