AI in practiceFrom possibility to product.
Built with AI.
Shaped by experience.
Pondering and Poke were built fully using AI. They show how we turn product rules, workflows and interface decisions into working software.
Explore the projects ↓AI-generated implementationHuman directionPractical validation
Selected development projectsTwo ideas. One AI-led approach.
A clear idea.
A complete game experience.
One frog in each coloured pond. No shared rows or columns. No touching frogs. Pondering turns those simple rules into daily puzzles and progressive levels.
TypeScriptViteCapacitor
Pondering privacy policy ↗
The challengeMake logic feel playable.
A puzzle needs more than a board. Generation, difficulty, controls and progression all have to work together, while every puzzle remains logically solvable.
How AI helpedDevelop the pieces together.
AI was used to create the implementation across puzzle logic, interface behaviour and mobile packaging. It made it easier to explore changes across connected parts of the game and refine them through feedback.
What it made possibleMore room to refine.
A shared daily puzzle, progressive levels, hints, undo, streaks and skins. Using AI throughout development helped shift attention towards how the game behaves and feels.
Inside the product +
Progressive play starts with a 5×5 board and three ponds. The Daily Pond uses an 8×8 board shared by everyone that day. Background workers support generation, and Capacitor provides the Android and iOS application structure.
Puzzle verification has covered 1,000 levels and 400 daily boards, checking unique solutions, logical solvability and repeated maps. Development checks remain part of the workflow as the game evolves.
An event-driven workflowPoke
- 01
Recognise the alertDiscord channel · keywords · cooldown
- 02
Coordinate the PCBrowser session · remote agent
- 03
Keep the user informedScreenshots · notifications · controls
Several tools.
One control surface.
Poke brings Discord alerts, browser actions, remote PC control, screenshots and monitoring schedules together in a private web application.
AutomationDiscordPC agent
The challengeConnect the steps around a task.
Monitoring a browser session involves several actions: responding to relevant alerts, opening a page, checking the screen and knowing when attention is needed.
How AI helpedTurn requirements into a workflow.
AI helped create the app from practical requirements, connecting alert rules, cooldowns, scheduling and remote controls in one interface. Iteration focused on making the controls clear and the workflow easy to follow.
What it made possibleLess switching between tools.
The interface brings monitoring settings and remote actions into one place. Screenshot feedback and notifications help the user check a session and decide what to do next.
Inside the workflow +
Channel and keyword rules identify relevant Discord alerts. The controller offers actions to open a target or alert URL on the PC, request screenshots and set recurring captures. Restock and CAPTCHA notification settings draw attention to events that need a response.
Day-of-week schedules cover starting and stopping monitoring, waking the PC and putting it to sleep. Scheduled waking uses a Windows wake timer. The application is a private development project, presented here as an example of AI-built automation.
Our AI development practiceExperience through application
AI across the build.
Judgement at every step.
Our experience comes from using AI to build, connect and refine real products. Clear requirements, useful feedback and deliberate checks help turn generated code into a coherent result.
01 / SpecifyMake the behaviour clear.
Break a large idea into rules, workflows and concrete examples that can guide implementation.
02 / IterateKeep the feedback close.
Use AI to develop and revise the software, then bring observations from the product back into the next change.
03 / VerifyCheck what matters.
Review generated work against the intended behaviour, from puzzle rules to alert settings and interface states.
