# Justus

Founder & Product Lead · 2026 · Independent product

## Reduce the planning burden

Justus is an AI fitness coach that stays with you through the ups and downs (and across all fitness modalities!). Workout data and generic plans are easy to find. The hard part is deciding what to do when work, family, fatigue, and the week’s actual training don’t match the schedule.

It's a private coaching experience that helps someone choose a sensible week, understand what happened, and decide what to change next. I own it end to end: product definition, implementation, release, user feedback, and iteration.

## Keep coaching after the plan is made

[Justus app demo video](https://justus.health/media/Youtube%20version%20of%20app%20video.mp4#t=1)

A plan is only useful if the person can follow it. Justus starts with a conversation about someone’s goals, history, and circumstances, then drafts a plan they can review and change by talking to the coach.

A weekly planning session connects the past week to the next one. Daily check-ins and post-workout feedback bring in what actually happened. Workouts sync from wearables, and the athlete adds how a session felt and the context a device can’t see.

Endurance and strength work live in a single plan, and the coach keeps a running profile of each athlete, so later conversations build on earlier ones.

I kept the experience focused: no public feed, leaderboards, or social comparison, and no abstract scores. The coach is designed to push back. So far its most common advice has been to slow down.

## Build AI with AI

All of Justus’s code is written by AI. I direct the work through product decisions, implementation plans, and feedback on the running application. The stack is React Native and Expo, with Supabase for the backend and Portkey for model routing.

I’ve built a development workflow with separate planning, implementation, and independent AI review stages. Each feature starts with a user-visible outcome and a clear definition of done. Coding agents work from shared product and engineering references, write tests, and exercise the actual application before a change is ready to ship.

My product judgment shapes those instructions and the criteria for accepting the result. I care about whether the coach understands someone’s circumstances, whether a proposed change leaves them in control, and whether the interaction makes a decision easier. Working code is necessary, but it doesn’t settle those questions.

## Close the loop from use to improvement

I built an observability harness that brings together application errors, athlete feedback, AI conversation traces, and evidence of what happened in the product. Agent-assisted review workflows investigate these signals and turn findings into proposed fixes and improvements. I decide which belong in the backlog and what needs a product decision before implementation.

For changes to prompts, orchestration, tools, or memory, I record the expected behavior and how we’ll recognize it after release. These “change watches” connect a shipped change to the evidence needed to assess it. The review process distinguishes a regression from a change that hasn’t had enough use to judge. A successful model call alone doesn’t establish that the intended product behavior occurred.

## Evolve the system as the models improve

I keep revisiting both the models inside Justus and the agents building it. Model selection is task-specific: a good choice for routing tools may not be a good choice for generating a training plan. I compare quality, latency, and the cost of the completed workflow, including retries and failures.

The development process has evolved alongside the product. More capable agents let me delegate more implementation and verification, while better-specified plans make some work suitable for less expensive models. I continue shaping the coaching behavior, interface, and priorities rather than letting the available model or its default style define the product.

## Explore the work

[Visit Justus](https://justus.health/)

## Scope of my work

Solo founder responsible for product direction, coaching and interaction design, AI-directed development, evaluations, observability, and release. Justus is in active beta.

---

[Gideon Emmanuel](https://gideon-emmanuel.com/about/) · [Let’s talk](mailto:gideon.m.emmanuel@gmail.com)

Canonical page: https://gideon-emmanuel.com/work/justus/