Personal Best

Designing and engineering Bod.Coach: a fully-realized AI personal trainer that plans, adapts, and coaches over text message.

Personal Best

The Challenge

To make real personal training (the planning, the adaptation, the accountability) accessible to anyone with a phone, at a fraction of the cost of a human trainer.

The Solution

An AI coaching platform built on a multi-agent architecture: specialized agents onboard new members, plan and reschedule workouts, respond mid-workout, and follow up proactively, all over SMS.

The Outcome

A production AI coaching platform serving paying subscribers, built on an agent architecture the Bod.Coach team can evolve for years to come.

A good coach learns who you are, plans the work, and shows up in your text messages.

Personal training works. It works because a good trainer learns who you are, adjusts when life happens, and checks in when you go quiet. It also costs more than most people can spend, which is why the fitness industry keeps shipping workout apps instead: libraries of content that put the burden of planning, scheduling, and showing up squarely back on the user.

Bod.Coach set out to close that gap, and partnered with Braid to design and engineer the platform. The premise is simple: a good coach shows up for you. So together we built an AI personal trainer that lives in the most human channel your phone has: your text messages.

Your coach learns about you, plans every workout, adapts when you travel or tweak a shoulder, and follows up when you miss a session. There’s nothing to install and no interface to learn. If you can text a friend, you can train with Bod.Coach.

StrategyStrategy
Strategy

Every product decision followed from one strategic bet: meet people where they already are. SMS requires no download, no account juggling, and no new habit, and it makes proactive coaching natural: your coach can start the conversation. From named coach personas with distinct voices, to an onboarding flow that is itself just a conversation, we designed Bod.Coach to feel like a relationship with a coach who knows you.

One phone number, one conversation. Behind it, a team of specialized agents, each an expert in its own job.

From the member’s side, Bod.Coach is one phone number and one continuous conversation. Behind that conversation is a multi-agent architecture: a primary router reads every incoming message and hands it to the right specialist: an onboarding agent that gets to know new members, a planning agent that builds and schedules workouts, an in-workout agent that answers questions mid-session and swaps exercises when a machine is taken, and a re-engagement agent that reaches out when someone goes quiet.

Each agent has its own carefully engineered prompt and its own set of tools: real procedures that create workouts, move schedules, log activity, and send links. Agents chain to one another when a conversation crosses domains, so the member never feels a hand-off.

Because an AI coach is only as good as its judgment, we built an evaluation harness alongside the product. Every prompt change must clear a battery of real conversation scenarios before members ever see it.

EngineeringEngineering
Engineering

Bod.Coach runs entirely on Cloudflare’s edge platform: Workers for compute, D1 for data, Durable Objects for conversation threads and scheduling, and Queues for reliable SMS delivery through Twilio. The AI layer is multi-provider by design: requests fall back across several frontier model providers, so a single provider outage never takes the coach offline.

Production AI has to work every time a real person texts their coach.

An AI product that thousands of real messages flow through demands a discipline of its own — and it’s the part of the Bod.Coach build we’re proudest of. The conversation engine streams model responses, validates every tool call, retries intelligently, and degrades gracefully when a provider misbehaves. Scheduled follow-ups fire from Durable Objects with exactly-once semantics, so “I’ll check in with you tomorrow” is a promise the system actually keeps.

The companion web app, built with Vue.js and Nuxt, handles everything that works better on a screen: rich workout tracking with rep, duration, and distance metrics, progress history, and subscription management.

User ExperienceUser Experience
User Experience

When it’s time to actually work out, Bod.Coach opens up beyond the text thread: a guided Play Mode walks members through each exercise with timers and intelligent auto-advancement, workouts subscribe directly into Apple and Google calendars, and every plan adjusts to the equipment on hand. Ask for a substitution mid-set and your coach handles it instantly.

Every AI engagement we take on is backed by agent systems we’ve designed, shipped, and proven in production.

Bod.Coach is a real business with real subscribers, and every part of the platform has to earn its keep. Building it sharpened our craft across the full discipline of agentic AI: conversation routing, tool design, evaluation, provider resilience, and the unglamorous operational work that separates products from prototypes.

Every AI engagement we take on benefits from that experience. When we tell a client an agent architecture will hold up in production, it’s because we’ve built one that does.


Let’s build something together. We’re excited to hear what you’re dreaming up and how we can help.

UVA DardenPepsiCoHealth WarriorTeak & TwineUVA McIntireSun Tribe SolarInternational Tennis Hall of FameDevils Backbone Brewing CompanyCosaic