tugas
Three bullet points in a note became an app on my phone. Built with an AI, with care and intention — this is everything in between.
My task system was a weekly paper diary — Monday to Sunday, the same format every year. It worked, except for the two things paper can’t do: reset a recurring chore by itself, and remind me. I’m only human — I forget.
So for weeks I trawled the App Store for a simple app to track recurring chores. Everything I found wanted something from me first — a sign-up, a subscription, a paywall in front of the basics — and every screen was doing too much. Nothing was simple enough. And there’s one more reason that matters: I’m dyslexic, and a crowded task list reads like noise — seeing the whole mountain at once is overwhelming. Seeing only today is calm.
I wanted my diary, digitised — no logins, no distractions, one week at a time. It didn’t exist, so I built it.
TUGAS is a deliberately tiny task app: one scrolling week, one accent colour, no accounts. What makes it mine isn’t the feature list — it’s that every design decision traces back to something I believe. This is the app’s DNA:
| What I believe | The design decision it produced |
|---|---|
| Overwhelm comes from seeing the whole mountain | Slice the load into daily, sectioned pieces |
| I plan my life one week at a time (the diary) | Week view, Monday to Sunday; this week only |
| Paper can’t reset recurring tasks | The recurrence engine — the core of the app |
| Paper doesn’t speak to me | On my phone, with reminders, always in my pocket |
| Show me only what’s mine right now | Opens on today; everything else a scroll away |
| Adding a task shouldn’t be another chore | A ghost “Add a task…” row — instant capture, no friction |
| My dyslexic brain needs less noise | One accent colour, big day headers, one task = one line |
| Chores aren’t deep — don’t make them deep | No accounts, no projects, no prep; set and forget |
| I don’t want stress to take away something I enjoy | Quiet UI; no streaks, no guilt, no gamified pressure |
Everything here ran the same loop: brief → design → build → verify → ship → document. Honestly? Never as one clean pass — the loop ran dozens of times, in small pieces. Six beats, each with a receipt and something you can take with you.
The questions found what the bullets hid.
TUGAS started as three bullet points: add a task, give it a category, make it repeat. Instead of building that, we interrogated it — one question at a time. What happens to an unfinished task at midnight? What does “every 2 days” mean when you tick it late? The three bullets were hiding at least six decisions I hadn’t made yet.
3 bullets in → 6+ hidden decisions found before anything was built
The AI shows options. I keep the pen.
Nothing went straight to real code. I asked for quick throwaway versions first — cheap to make, cheap to bin. When the task-row swipe looked rough on my actual phone, I had the AI mock up six different fixes to compare… then drew my own seventh. That’s the one that shipped.
Small pieces, each proven before the next.
The app grew in small, reviewed steps — never one big leap. The trickiest part, the rules for what happens to a repeating task at midnight, was written and tested before a single screen existed. Today 50 automated checks re-prove those rules every time anything changes. On the busiest day, five improved versions of the app shipped between morning and night.
50 automated checks · five builds shipped in one day
No green test, no upload.
Build 7 shipped with two buttons that looked perfect and did nothing — every check had passed except the one that mattered: actually tapping them. That failure became a rule. Now a simulated finger taps through the real app — swipes, edits, deletes — before any build is allowed out the door.
The rule itself: no green test, no upload
Shipping is a skill you practise.
The first release took a whole evening of certificates and ceremony. By the ninth, releasing was routine. Nine real versions went to my iPhone through TestFlight — Apple’s channel for pre-release apps — in six days, each one a little better than the last.
Build timeline: 1.0 (1) → 1.0 (9), six days
Write the loop down — it compounds.
Every session starts where the last one ended, because everything is written down: decisions, feedback, what worked. My feedback measurably improved on the record. Round 1’s “the calendar is incorrect” wasted a whole round — the fix matched the words, not what I meant. By round 3 I was sending annotated screenshots of current vs desired, and fixes landed in one shot. The playbook now has a rule at the top: show the destination, not just the problem.
TUGAS shipped. It’s on my phone, on TestFlight, and it runs my week — used daily by the person it was built for. But the app is only half the outcome. The other half is what this practice did to the practice itself. In March I ran my first AI-built project, A Designer’s AI Field Notes; TUGAS, four months later, is the test of what stuck:
| Skill | March · first project | July · TUGAS |
|---|---|---|
| Starting | No plan, a barely-formed prompt | A validated spec and a step-by-step build plan before any code |
| Giving context | Light briefs, content scattered elsewhere | A living spec, a project log, a written brief for every round of work |
| Asking | A prompt, barely formed | Self-contained briefs — each fix scoped with its cause and how to check it |
| Giving feedback | Discovered the hard way that feedback is a skill | Sketches, reference images, real-phone screenshots of current vs desired — one-shot fixes |
| Designing | Rebuilt my design file properly only after a messy first go | Design-system-first from day one; binned a weak attempt cheaply |
| UX judgement | Content-driven, layout-led | Opinionated defaults, minimalism with reasons, empty-state thinking |
| Directing the code | Could read it well enough to follow along | Directed the architecture; caught fixes that solved the wrong problem |
| Verifying | Mostly by eye | A screenshot for every fix, 50 automated checks, always run the real thing |
| Rigour | First real app, learning the tools | The tricky logic written as tested rules before any screens existed |
| Working with AI | Named the roles: lead, ask, override, step away | Internalised — approach agreed first, clear go-aheads, judgement calls kept |
| Shipping & backing up | GitHub from day one | Shipped much further — a real iPhone, TestFlight — but backed up later. Still working on that one. |
And this is the part that compounds.
One more thing. Not long before TUGAS, Satya Nadella wrote that the real opportunity in AI isn’t picking the best model — it’s “building a learning loop on top of models where human capital and token capital compound.” In plain words: your judgement, plus a way of working with AI that gets better every round. That’s what this page is. TUGAS is just the app it produced.