Wabi's app-builder-to-messenger turn: how to test a category bet

When your users start asking for outcomes the product was not built to own, the real decision is whether to ship a fix or make a category bet. The test is whether the reposition changes the user's core job, the product's identity, and the competitive field; if only features change, treat it as a fix.

Eugenia Kuyda, who previously built Replika, presented Wabi 2.0 as a personal agent that assembles the needed interface in real time, and Wabi is shifting its main product to a messenger-style agent that creates task-specific interfaces on the fly. The case is useful because the shift is visible in the interface, not just in the roadmap.

The user's job changes before the code does

A feature fix improves how the same job gets done. A category bet changes what the user is hiring the product to do. Wabi's new shape makes that difference visible: the user asks for a task, and the agent creates the surface needed for the result. Wabi 2.0's on-demand interface examples include a real-time calorie tracker and shared family calendars. Those are task surfaces that appear when the job needs them.

Founders often wait too long because the old product still works. It can still load, still convert, still earn. The behavior that matters is whether the user's next request fits the old category. When users keep asking for outcomes the product was not built to own, the product has become a doorway to a different job. The new job is the real asset, and the old interface may be the wrong vehicle for it.

The interface changes what the product is

Identity is the surface the user learns to trust, not the logo. The old interface may carry the habits of the old job. The new interface carries a request, a generated screen, and a result. Kuyda's argument is that chat-only agents become harder to use over time because past messages and active tasks sink into a single scroll, and that people prefer tapping, scrolling, and looking as well as typing. That points to a different product: one that generates the right screen for the task instead of keeping everything in a conversation. A generated screen changes more than the surface. It changes what the user expects to find after the first tap.

The competitive field is the hardest tell

The fair objection is that the new product is still close to the old version. The field around it has moved. Wabi's pivot is occurring while demand for AI agents such as Meta's Muse and Instinct is rising. OpenAI announced at DevDay a ChatGPT plug-in update that lets apps provide interactive experiences inside ChatGPT and pins the app's name in a more visible sidebar spot. The category Wabi is entering has more moving parts in practice. Staying in the old category means competing with products in that category. Becoming a personal agent means competing with agents, platforms, and the surfaces those platforms control.

The category-bet test is the decision tool

A fix can be shipped in a sprint. A reposition asks the team to abandon the name, the habits, and the proof that the old product worked. Wabi 2.0 is available only through invite codes distributed on X. The invite-only release gives the company a way to test the new identity before it commits the whole brand to it. The test is quick to run and hard to ignore.

• Write the old job and the new job in one sentence each. If the old job still explains why users open the product, keep the identity and treat the change as a fix.
• Check the first action. If the core action is unchanged, keep the interface; if it is a request, a generated screen, or a delegated task, test the new identity before you commit.
• Name the competitors. If the list is the same, the field has not changed; if it includes platforms, agents, or distribution surfaces, abandon the old category story and reposition.