AI changes *Nothing* — Dax Raad, OpenCode

Other coding agents promise to be the smartest. Dax Raad introduces OpenCode as the “hottest coding agent”: an experience built around good UX, themes, personalization and fun. He describes the project he works on as fully open source and one of the most used coding agents, and pitches its flexibility to work across models. The adoption claim is his characterization; the model pitch is about choice of providers, rather than a guarantee that every model works equally well.

That opening points to a problem that better AI does not automatically solve: how does software become a product people choose to use? Raad grants that some predictions about AI changing software development contain truth. His concern is the work that remains difficult even with those capabilities available. A prospective user starts by knowing nothing about a product, hears about it, tries it, likes it and eventually commits to it. Three transitions in that funnel demand particular care: earning attention, making the value click and keeping the user. These are still daily struggles for him, despite his experience and access to AI.

The person you want to reach is already occupied. They have their own problems, worries and plans; your product is not among them. Marketing has to make that person stop, pay attention and investigate what you offer. That is a much higher bar than explaining a feature accurately to someone already interested.

For a builder, the first obstacle can be psychological. Marketing is an unfamiliar skill, and discomfort easily turns into the conviction that it should not be necessary: a sufficiently good product ought to sell itself, without anything that feels gimmicky. But even outstanding products must earn attention. Treating the task as beneath you does not remove it from the funnel. It only leaves a necessary skill undeveloped.

Raad's test for an attention campaign is whether it has a chance of being shared. A feature announcement, a paid influencer explaining that feature or a billboard can communicate information. But what makes a stranger send it to a friend? His example of a blog post announcing a product's seventy-sixth feature exposes the gap: the company has something to announce, but the reader has little reason to care. Against a feed trained by TikTok, routine announcements are easy to scroll past. In his framing, they do not accomplish the essential work at the top of the funnel.

The material needs to strike a chord: something sufficiently funny, surprising, outrageous or deep that a viewer wants coworkers or friends to see it. Shareability is a creative constraint, not a promise of success. An idea can have a plausible route to that reaction and still miss. The useful distinction is between an attempt that might become a hit and one that never gave anyone a reason to pass it along.

Raad illustrates the payoff with nine embarrassing or unsuccessful attempts followed by one hit seen by a million people; this is a hypothetical, not a reported campaign result. The target audience might be among those viewers or connected to them. Repeated attempts can compound reach while improving the maker's ability to recognize and create something shareable.

This is where his experience with AI has disappointed him. Raad says he has not obtained a good marketing idea from it, even as a brainstorming partner: the suggestions feel corny rather than cool. That is an account of his own creative work, not a measured comparison between systems. He nevertheless treats the ability to make something cool as a skill people can learn, rather than an innate quality that excuses everyone else from trying.

Once someone is interested, the next job is to reach the Aha moment: the interaction that makes the product's value click. The person need not become a customer for that moment to matter. They might recognize that the product is useful for someone else and recommend it. The goal is a clear understanding of why the product deserves to exist.

Getting there requires choosing among features, not introducing all of them. Features A, B and C may each be good, but a first experience cannot give every feature equal priority. Raad's procedure is deliberately severe:

1. Identify the single interaction that best communicates the product's value.
2. Deprioritize competing features in the initial experience.
3. Inspect every step between arrival and that interaction.
4. Cut the steps that do not need to come first.

The cost of extra steps is that people leave before they understand what you built. Raad uses losses of 10% or 20% at individual onboarding steps as illustrative estimates, not measured funnel data. Questions about company size, employee count and job title make the problem concrete: the company is collecting information while the visitor is still deciding whether the product is worth using. Those questions obstruct the very experience that could make answering them worthwhile.

His hypothetical 3% arrival rate at the Aha moment shows how onboarding can squander the attention marketing earned. A compelling campaign does little for a product if almost everyone disappears before experiencing its value.

Sometimes the problem is deeper than onboarding: there may be no Aha moment to reach. That forces a prior question about whether the product is worth the effort of building. Raad turns to two deliberately extreme examples of value that becomes legible through an interaction. He calls ChatGPT the most successful consumer product of his lifetime—a personal judgment—and emphasizes its open-ended input box. A person can enter a question in ordinary language and receive a human-like response. The breadth and immediacy of the interaction make its appeal easy to grasp.

His Facebook example is more socially specific: sending a poke to a crush lets someone flirt while retaining deniability. In that situation, the feature's value does not need a tour or an explanation of the platform. The interaction itself supplies a reason to return.

These are Raad's examples of a singular reason to use a product, rather than explanations of everything that made either product successful.

Finding an equivalent moment in your own product requires creativity and a close understanding of the space. Positioning, clarity about the offering and a sense of what you do uniquely all matter. Raad does not find that having an AI to talk to automatically produces that understanding. People who already excel at this work still have an advantage; everyone else still has to develop the skill.

Attention and activation bring someone into the product. Retention determines whether that work accumulates. Raad's aspiration is to keep customers for life: if new users continually replace departing users, leakage can put a ceiling on the business. Keeping existing customers lets new arrivals add to a growing base.

This changes the design problem. A first-time user needs a short, clear path to value. A longtime user may want configuration, advanced features and the ability to push the product beyond its default behavior. Teams and enterprises introduce further requirements. The challenge is to support those needs without making every newcomer confront them at the start.

Raad draws on several years of open source work, where this range of users is especially visible:

• Individual hobbyists want something simple, free and immediately usable.
• Large enterprises may choose the same project because they want to run and host it themselves and tightly control where their data goes.

Both are legitimate users of the same software. Serving one cannot simply mean dismissing the other.

A common response is to frame simplicity and capability as opposing choices. Raad rejects that framing. Saying a product is simple like Apple's does not justify making it incapable. A simple experience can sit on top of a powerful system, but designing both takes more work up front.

His architectural sequence starts below the feature:

1. Consider a feature users need.
2. Identify the underlying primitives that could be assembled into it.
3. Make those primitives broad and powerful enough to support more demanding uses.
4. Assemble the straightforward default experience from those same primitives.

Raad describes that default as the experience for 99% of users, a design framing rather than a measured usage share. The important relationship is that the simple product uses the capable foundation itself.

As users need more control, expose the underlying primitives more directly. They can then bend the product to fit their needs without abandoning it. Otherwise, a product that users outgrow effectively becomes an acquisition funnel for the competitor they move to. Anticipated future capabilities should therefore inform the primitive set from which the current experience is assembled.

The architectural prescription leaves a hard question unanswered: which primitives are right? For Raad, trying to build them is part of discovering what they need to be. The requirements and abstractions do not arrive fully formed before implementation begins. Working through the design reveals both the necessary building blocks and a clearer understanding of the product itself.

AI can participate in that exploration as a brainstorming partner. But communicating what the system should do already requires understanding it well, and developing that understanding is the central task. Time, energy and work in the domain are what let a designer recognize primitives that will hold across different situations. Access to AI does not, in Raad's experience, remove that intellectual work.

OpenCode is not offered as a finished demonstration of getting all this right. Raad invites people to try it and expects them to find places where he fails to follow his own advice. Being able to articulate the principles does not mean consistently executing them. Aligning attention, the first experience and long-term capability takes sustained effort from a person and a team.

He explicitly credits AI with making previously impossible activities possible. What it has not removed is the daily difficulty of making something great. That remaining difficulty is also his answer to the fear that AI will leave people without purpose: the work still demands judgment, creativity and persistence, and those demands remain a source of both meaning and fun.