Emphanic · AI product studio of Tom Suter
Over the years I have built hundreds of products, moving from classic development to low-code and no-code platforms. For the past three years I have built fully with AI. Emphanic is where those AI products live, from the iPhone game Heatwave to a tool for validating product ideas, a daily puzzle and Swiss finance calculators.
Each product starts from a concrete question. Below you will find my latest products, how I work, the tools I use and the latest developments in AI.
Portfolio
My most recent products, all designed, built and launched with AI.
The Battle for Cold
A location based horror game for iPhone, played outdoors with earbuds. Designed for groups of friends and first tested in Appenzell.
Key questionWhether a phone, earbuds and familiar streets can create a more immersive experience than a screen, and whether one builder working with AI can ship a full mobile game to real players.
Turns product ideas into a structured evidence tree. Start from a vision, add test and learning cards, and score every test on time, distance and cost to real data. AI sharpens each card.
Key questionWhether the discipline of Alberto Savoia's The Right It and Strategyzer's test cards survives being turned into software, and whether AI can make a vague bet sharp enough to test.
A daily code-breaking game. One global vault per day, a five-emoji code, six attempts and a par to beat. Days add up to weekly challenges, in five languages and without an account.
Key questionWhether a new puzzle can go live every day that is solvable, fair and engaging, without manual review.
Free Swiss finance calculators in four languages. One question, your figures, an answer in thirty seconds. Behind it sits a maintained Swiss dataset that updates itself when an official source changes.
Key questionWhether a focused, useful tool site can still be found in Switzerland, and how much of the building, translation and maintenance AI can take on.
The original Emphanic product. It measures how often ChatGPT, Claude, Perplexity and Gemini name a business, and which competitors they name instead.
Key questionWhether something as elusive as "does AI recommend me" can be measured reliably, week after week, in a way a business owner can act on.
AI chatbots for Swiss small and medium businesses, trained on a company's own content. The models run locally on a Mac mini with Ollama, connected securely through Tailscale, so the data stays on Swiss hardware.
Key questionWhether a chatbot trained on a company's own pages becomes good enough to sell, and what it takes for a business owner to trust it with customers.
Constructive news: the day's stories collected, rated for impact and retold with the constructive angle included, on the web and in a mobile app.
Key questionWhether a news pipeline can run on its own, from collection to publication, and still be worth reading.
AI trends
What is moving in AI development right now, researched every week from published sources, with what it means for anyone building with AI.
Last updated:
Anthropic relaunched Projects in Claude Code: several agents run side by side in one workspace and share memory, goals and files under a central coordinator.
Why it matters The bottleneck moves from writing code to briefing parallel work well and reviewing what comes back.
Microsoft warned organisations to treat AI agents as potential insider risks: they can reach company data, apps and tools and act on that access quickly. Its advice is to keep controls outside the model, give each agent its own identity and only the access its task needs, and monitor continuously.
Why it matters Every agent you deploy deserves the same access review as a new employee.
Attackers broke into an older system of the open source coding agent OpenCode, affecting 1.4% of users of its OpenCode Console, where some stored their AI provider keys. Affected users were told to replace their keys with each provider and watch for unauthorised charges.
Why it matters Use a separate key per project, set spending limits at the provider, and store keys in as few places as possible.
Meta says its new Muse Wonder models reach frontier level, and the full version is only available through its Muse Code subscription.
Why it matters Model makers now compete on the coding tool around the model, not just on the model itself.
Zuletzt aktualisiert:
Anthropic hat Projects in Claude Code neu lanciert: Mehrere Agenten laufen nebeneinander in einem Workspace und teilen Gedächtnis, Ziele und Dateien unter einem zentralen Koordinator.
Warum es zählt Der Engpass verschiebt sich vom Schreiben von Code zum guten Briefing paralleler Arbeit und zum Prüfen der Ergebnisse.
Microsoft rät Unternehmen, KI-Agenten als mögliches Insider-Risiko zu behandeln: Sie erreichen Firmendaten, Apps und Tools und handeln schnell. Die Empfehlung: Kontrollen ausserhalb des Modells, eine eigene Identität pro Agent, nur der Zugriff, den die Aufgabe braucht, und laufendes Monitoring.
Warum es zählt Jeder Agent verdient dieselbe Zugriffsprüfung wie eine neue Mitarbeiterin.
Angreifer sind in ein älteres System des Open-Source-Coding-Agenten OpenCode eingedrungen und haben 1,4 % der Nutzer der OpenCode Console getroffen, wo manche ihre Schlüssel für KI-Anbieter gespeichert hatten. Betroffene sollen ihre Schlüssel beim jeweiligen Anbieter ersetzen und auf unerlaubte Abbuchungen achten.
Warum es zählt Pro Projekt ein eigener Schlüssel, Ausgabenlimits beim Anbieter, und Schlüssel an so wenigen Orten wie möglich.
Meta sagt, seine neuen Muse-Wonder-Modelle erreichen Spitzenniveau, und die Vollversion gibt es nur im Abo Muse Code.
Warum es zählt Modellanbieter konkurrieren jetzt um das Coding-Werkzeug rund um das Modell, nicht nur um das Modell selbst.
Laatst bijgewerkt:
Anthropic heeft Projects in Claude Code opnieuw gelanceerd: meerdere agenten draaien naast elkaar in één werkruimte en delen geheugen, doelen en bestanden onder een centrale coördinator.
Waarom het ertoe doet Het knelpunt verschuift van code schrijven naar parallel werk goed briefen en controleren wat er terugkomt.
Microsoft waarschuwt organisaties om AI-agenten te zien als mogelijk insiderrisico: ze bereiken bedrijfsdata, apps en tools en handelen snel. Het advies: controles buiten het model, een eigen identiteit per agent, alleen de toegang die de taak nodig heeft, en doorlopende monitoring.
Waarom het ertoe doet Elke agent verdient dezelfde toegangscontrole als een nieuwe collega.
Aanvallers drongen een ouder systeem binnen van de open source coding-agent OpenCode, wat 1,4% van de gebruikers van de OpenCode Console raakte, waar sommigen hun sleutels voor AI-aanbieders bewaarden. Getroffen gebruikers moeten hun sleutels bij elke aanbieder vervangen en letten op ongeautoriseerde kosten.
Waarom het ertoe doet Gebruik per project een eigen sleutel, stel uitgavenlimieten in bij de aanbieder, en bewaar sleutels op zo min mogelijk plekken.
Meta zegt dat zijn nieuwe Muse Wonder-modellen topniveau halen, en de volledige versie is alleen beschikbaar via het Muse Code-abonnement.
Waarom het ertoe doet Modelmakers concurreren nu op de coding-tool rond het model, niet alleen op het model zelf.
Approach
Not from a feature list. Every product answers a question that research alone could not settle, and that question sets the priorities.
A live product with real users tells more than any presentation. Prototypes go out early, as a beta where needed, and data decides what stays.
AI agents write much of the code, copy and translations. I review, test and take responsibility for every result before it ships.
Toolkit
The tools, models and protocols I build with every day, and what each one taught me.
Tooling
My primary development tool. It loads its own skills, runs subagents in parallel, uses a separate git worktree per branch, and automated checks stop a build when a retired design value reappears.
Key insightHow much of a real codebase an agent can handle at once, and where human oversight remains essential.
Tooling
A second tool on the same repository, with its own config file and its own list of MCP servers.
Key insightTwo agents with different habits on one codebase. The repository needed rules that hold regardless of which agent is working.
Design
The visual system lives outside the code, in its own design project, and gets pulled into the repo one component at a time.
Key insightDesign decides colour, type and shape; code decides how they are built. Documenting that split ended most of the debate.
Models
GPT, Claude, Gemini and Perplexity, all behind one adapter that logs tokens and cost on every single call and stops at a monthly budget.
Key insightFour assistants answer the same question differently, which is why the visibility check exists. Central metering keeps costs predictable.
Agents
Eight agents with separate roles, each reporting to a lead agent that ranks the results before I review them.
Key insightA team of agents produces a lot of confident output quickly. Nothing is published without my review.
Protocol
Model Context Protocol. The database hangs off it as a server, so Claude Code and Codex both reach the same Supabase through one shared set of tools.
Key insightA connection described once works in every tool I adopt next. That portability proved more valuable than the connection itself.
Why "emphanic"
Two words pressed into one. Empathic, understanding the person in front of you and what they actually need. And emphanes, Greek for visible, manifest, showing itself plainly.
That combination is the idea behind the studio: technology is valuable when it understands what someone means and makes the result visible. The name fit a tool about being found by AI, and it fits everything else here as well.
I am Tom Suter. I moved from the Netherlands to Switzerland in 2017 and live in Appenzell AI. My career spans web, product, SaaS, startups and agile ways of working, and today I work at 42hacks. Alongside that, I explore the opportunities of AI closely and hands-on. I have built hundreds of products over the years, from classic development to low-code and no-code, and for the past three years fully with AI. Emphanic is where that work comes together.
FAQ
Tom Suter is a Dutch product builder who has lived in Switzerland since 2017 and works at 42hacks. Alongside that, he explores the opportunities of AI closely and hands-on. Over the years he has built hundreds of products, moving from classic development to low-code and no-code platforms, and for the past three years he has worked fully with AI. His background spans web, product, SaaS, startups and agile ways of working.
Emphanic is the AI product studio of Tom Suter. The name combines empathic, understanding what someone needs, with the Greek emphanes, visible. It is where his products and experiments with AI live.
His most recent AI products are Heatwave, an iPhone horror game; Validaid, a tool for testing product ideas; vault.day, a daily puzzle game; kalko.ch, Swiss finance calculators; the Emphanic AI visibility check; swisschatbot.ch; and brightcast.news.
Mainly Claude Code and Codex for building, Claude Design for the visual system, MCP to connect tools and data, and models from OpenAI, Anthropic, Google and Perplexity behind one metered adapter.
Simply get in touch. Whether it is a first question about what AI can do for your business, a second opinion on a plan or a product you want built, I would love to help. Email tsuter5@gmail.com or connect on LinkedIn.