On Products
what is a product today, and tomorrow?
somewhere in PRsomewhere in PRI wrote this during my coding vacation. I really like to surf first thing in the morning and then go all in in my many rabbit holes. If you like longboard surfing pls reach out with any reccs!, April 2026
My favorite products of all time: the iPhone, Claude Code, my old 1999 Suzuki Trooper, my Kawai grand piano, and Notion.
The iPhone, because it just works. Claude Code, because it just works. My old 1999 Suzuki Trooper, because after 30 years, it just still works. My Kawai grand piano, because it is beautiful and allows me to be human and express myself as a human via music. And Notion, because it is beautiful and allows me to make writing feel effortless. Again, it allows me to be human and do a human-centered action.
What do these all have in common? They disappear. The best products get out of the way and let you do the thing you actually wanted to do. The piano does not make you think about the piano. The Trooper does not make you think about the Trooper. They just serve you.
I see a big shift here though
I see a big shift here though. Software products from this list are: Claude Code and Notion.
The first one, I love it because: it just works. The second one: it is pretty.
Those are two very different reasons. And I think that distinction tells you everything about where we are headed.
I believe in pretty
I thoroughly believe in pretty. Pretty can be moatmoatA competitive advantage that is difficult to replicate. Originally a castle defense metaphor, now widely used in business strategy. A moat protects your market position the way a water-filled trench protects a fortress.. For example: I pay for Notion but I use no features. I only use the writing feature, I do not use anything else. The only reason I use it: it is pretty, and Google Docs is ugly.
I grew up in Italy, home to so much art that shaped me profoundly and made art and bellezzabellezzaItalian for "beauty." Not just aesthetic beauty, but a cultural value that permeates how Italians approach everything, from food to fashion to architecture to daily life. just second nature to how I think and operate. Beautiful is a no-brainer, ugly is never acceptable: from how a city is designed, to how you host a dinner, to how you plate your tiramisu, from the Sistine Chapel to a perfectly crafted aperol spritz to an impeccable evening listening to opera at the Arena di Verona in a summer night, to how you cook a perfect piatto di pasta and plate it with love and craft, to everything. I love beauty. I still believe that ugly is unacceptable and that in the future all ugly UIs will be gone. B2B SaaSB2B SaaSBusiness-to-Business Software as a Service. Enterprise software sold on subscription to other companies. Historically prioritized functionality over aesthetics, because purchasing decisions were made by procurement teams rather than end users. softwares like Azure, AWS, and Oracle will have one day to become more aesthetic and user friendly. At least, that is the future I dream about. I dream of beauty and seeing it everywhere.
But, is that necessary anymore?
But, is that necessary anymore? The goal is not anymore to learn how to use a UI, but to rather achieve a goal. In a world of GUIsGUIGraphical User Interface. The visual layer of software: buttons, menus, drag-and-drop, icons. Everything you click on and see on screen. GUIs made computers accessible to non-technical users starting in the 1980s., pretty is necessary, because if a UI is ugly, the user is less likely to learn how to use it and there is more friction to usage, i.e. to learn the buttons and clicks needed to get a thing done. But when software can become an agent (or an orchestration of agents) using CLICLICommand Line Interface. A text-based way to interact with software by typing commands rather than clicking buttons. No visual interface needed. Developers use CLIs daily, and increasingly, so do AI agents. commands or MCPsMCPModel Context Protocol. An open standard that lets AI models connect to external tools and data sources through a unified interface. Think of it as USB-C for AI: one protocol, many tools. It allows agents to call APIs, query databases, and interact with software programmatically., when all the functionalities are programmatically accessible and are suddenly tool callstool callsWhen an AI agent invokes an external function or API to perform an action. Instead of a human clicking a button in a UI, the agent makes a structured function call. The agent decides what to call, with what parameters, and interprets the result., does the user even need to learn how to use the UI manually?
What prevents a user from just using a Claude Cowork connected to a Salesforce via natural language, and stop using the Salesforce UI entirely? And what prevents a user from now not even having to type anymore, but just dictates what they want?
But then, what now? What is a product if not how it is perceived by a user? What is perception now?
I still believe in beauty
I still believe in beauty, but I think that friction before was un-intuitive/ugly UI. Now, friction is evals-lacking, poorly-performing, and poorly-orchestrated AI agents behind an AI product experience.
Beauty is not going away. But the definition of what makes something beautiful is shifting. A product that anticipates what you need, that does the right thing on the first try, that never hallucinates or fumbles. That is the new beautiful. Correctness is the new elegance.
The bar just got raised
I think that the bar for software products just completely re-anchored in a new dimension.
EvalsevalsShort for evaluations. Systematic tests that measure whether an AI system is producing the right outputs, behaving correctly, and meeting quality bars. Think of them as the product's immune system: they catch regressions before users do. are what UXUXUser Experience. The discipline of designing how a product feels to use: intuitive navigation, clear feedback, minimal friction. For the past two decades, great UX was the primary differentiator between winning and losing software products. was to the software products of yesterday.
Evals are what track whether your agents are saying, generating, and doing the right thing, more on this hereon-evals. Since doing that is the whole point of using these tools, it does not matter how pretty they are. If they do not work properly, they will not be used repeatedly. See PiPi (by Inflection AI)An AI chatbot launched by Inflection AIinflection.ai in 2023, known for its warm conversational tone, beautiful pastel UI, and empathetic personality. Despite critical acclaim for its design and human-feeling interaction, it struggled to retain users against competitors with stronger underlying models. Try Pi here.hey.pi.ai, from Inflectioninflection.ai. I absolutely loved its UI. I absolutely loved the pastel shade and soft visuals. They were the very first in the wave of human-feeling AI branding, even before Anthropic branding team which followed this similar branding direction with 'keep thinking' and the Air Mail pop ups
I wrote more about why evals matter so much On Evalson-evals.
UI/UX and QA is to the product of yesterday, as Evals is to the products of tomorrow.
Re-imagining software products
Your product, as a user, becomes the outcome. Not the interface, not the features, not the FE/BE stack. The outcome. This applies to both consumer products and enterprise products. What used to be your product (the app, the dashboard, the UI) is now just the playground your agent operates in. The agent interacts with the traditional product (UI, FE, BE, SDKs, etc) on your behalf, and what you experience is the result of that interaction. That is the product now.
The magic moment of a product now is how many clicks can you avoid.
If the user never touches the UI, then the UI is not the product anymore. The outcome is the product. And whoever orchestrates the best outcome wins, regardless of what the UI looks like.
That is why "software" as we know it, is essentially dead. Because a traditional FE/BE/database/SDK kind of software is not the product anymore. As a startup founder or as a big lab or existing incumbent creating new tools, you do not need to create new "products" or new tools. At least, not products in the traditional sense (FE/BE and a FE with clicks/buttons that a user will need to learn how to click and how and why, etc). The new product is an agent (or orchestration of them) that leverages existing products and tools to the max.
We do not need necessarily new things. We need someone to use them for us instead.
If you are thinking about what products look like in the age of agents, or building something where evals matter more than pixels, I would love to hear from you.