Context
AI Coach is my personal beta application for planning the day, capturing reality, and evening reflection. It is not a finished product or a universal answer to productivity. It is a system I use in my own life and keep adjusting according to what actually helps.
It came from a simple problem: a normal todo list captures what I should do, but it does not show me where the whole day really disappeared. A plan can look reasonable in the morning, but only comparison with reality shows whether it was doable, what broke it, and what makes sense to change next time.
What the application does today
The core of the application is a daily timeline. The plan is not built only as a list of tasks, but as a concrete layout of the day across 24 hours. Even while planning, it becomes visible what can realistically fit into the day, where the fixed blocks are, and where a person is only pretending to have more time than they actually do.
Alongside the plan, the application stores reality: what really happened, when it started, how long it took, and which area of the day it belonged to. Because of that, AI Coach works not only with intention, but also with facts about the actual day.
Other parts build on top of that: todo tasks, recurring planning blocks, a journal, notes, rituals, a daily reflection on three positive moments, and a chart that shows where time goes.
Closing the day is not just a form at the end. It is the moment when the reality of the day turns into feedback: what stayed open, what was worth recording, and what should change the next plan.
Plan vs. reality
The most important principle in AI Coach is the difference between plan and reality. A plan can look good by itself, but until it is compared with what really happened, a person cannot see whether they planned realistically.
The goal is not to punish deviations from the plan. A day can change for good reasons. What matters is seeing what broke the plan, which things regularly do not fit, where energy disappears, and whether reality replaces the plan with something meaningful or only with random drift.
Technical and design state
AI Coach currently runs on Next.js and React over Supabase/PostgreSQL. I use Next.js for both the UI and the backend part, but I do not treat it as the final boundary of the architecture. I expect that part of the server-side and application logic may later move to C#/.NET, so I already try to separate the UI, application logic, repository layer, and data model.
Part of the work is also a custom design layer. I do not design the application only ad hoc for the current screen. I am gradually building a system for it in Penpot: from foundations such as colors, typography, spacing, and icons, through general UI components and repeated patterns, to components for specific parts of the application, page templates, and the screens themselves. I want to add user flows later. This helps me keep direction even while the application grows quickly and individual parts are created gradually.
The application is web-based, but I also use it like a mobile app through PWA. I have it installed in Chrome on my phone, so it behaves more like a standalone app than a regular browser page. That is why I also consider mobile use in the design, and some screens have their own mobile variant.
Current state
AI Coach is not a final product today. It is a used beta application and also a place where I test what type of planning really helps me. The value is not only that the application can store a plan or tasks, but that it gradually collects a truer history of the day.
Paradoxically, AI Coach did not start as an application, but as a conversation with AI. At first, I talked with Codex about planning, reflection, and what happens during the day. Later, I gave it access to a database, let it create the data structure, and started recording data into it. Only after that did it gradually become a web application.
Today, the application itself has no AI in it. That is an intentional intermediate step: first I need to reliably capture the plan, reality, tasks, and daily reflection. The AI layer should come only once it has something concrete to work from, not as a generator of generic advice, but as a helper over the specific history of my day.