If Users Can't Figure Out Your Product, That's on You

01 /The Concept

"Still burning time updating user docs, explainer videos, and online help centers? We're coming for you." That's essentially the pitch from Sammy Labs, which claims to offer "the fastest and most effective tool for onboarding and activating users" of cloud services.

The key innovation: Sammy's AI crawls a customer's service on its own, learning how it works – then uses that knowledge to guide other users through the same service. It does this continuously, so when new features ship, interfaces change, or workflows shift, Sammy automatically updates its internal knowledge base.

Setup on the product side is a single line of code that embeds and brands the Sammy widget. When a user gets stuck, they trigger Sammy – which can see the same screen the user sees and walks them through the steps needed to complete their task. Guidance arrives via Slack or email.

Last week, the startup launched a new version on Product Hunt. Sammy can now hear users' spoken questions in addition to seeing their screen – and responds with voice guidance as well. The demo makes the upgrade immediately legible.

At the same time, Sammy launched a public catalog where users can share guides they've created for services whose makers haven't yet integrated Sammy.

Sammy Labs positions itself as an AI technical support layer – and charges companies per successfully resolved issue, starting at 12 cents each. Only outcomes get billed.

This is a [second look](/review/prishlo-vremja-peredelyvat-a-ne-uluchshat) at Sammy Labs – the first was when it entered Y Combinator.

02 /Why It Matters

As one of Sammy's customers put it: "Less than 1% of our users read documentation, less than 3% watch explainer videos – so all the effort we put into creating and maintaining those things simply doesn't pay off. Sammy changed everything."

The reason is behavioral. What does most people do when they hit a confusing moment in an app or web service? Read the docs? Search for a tutorial video? Contact support? Some do.

But most people ask someone who knows – a partner, a kid, a coworker – and say: "Can you show me how this works?" Sammy steps into that exact role. The familiar behavior pattern doesn't break – it just gets AI instead of a person. And Sammy learned the product the same way a person would: by navigating it and accumulating experience.

The framing of Sammy Labs as a "technical support" tool is worth unpacking. Because AI is dissolving the line between interface and support. When an AI guides you through a task in real time, on screen – is that part of the product's interface, or a support function?

Honestly, it's both. What matters is the business model choice. Position your AI as an interface layer, and developers will probably pay a flat subscription. Position it as technical support, and you can charge per resolved issue – which can mean dramatically more revenue, especially for products with complex workflows and users who won't self-serve.

OpenCopilot, another Y Combinator graduate, originally sold developers an embedded AI assistant for their products – a clear interface framing. It then pivoted to selling companies an AI platform for automating technical support operations. The underlying engine didn't change much. The positioning and pricing model did.

Somewhere between interface and support sits a third category that's growing fast thanks to AI: users asking a product to *do* something complex without specifying which buttons to press. The AI needs to understand the intent, map it to a sequence of actions in the service, execute those actions, and surface the result. Sammy is moving in this direction too – offering developers MCP server integrations so Sammy can handle requests like "show me my most valuable customers" or "have all my team members completed onboarding?"

Startup Adopt ([covered here](/review/tvoi-polzovateli-skoro-tebe-zajavjat-vot-takoe)) raised $6M in its first round in May for a platform that lets developers embed this kind of purpose-built AI agent directly into their products. Technically that agent is an interface – it replaces the traditional button-based UI. But it could also be framed as support: it helps users execute complex tasks without being walked through every step manually.

03 /Opportunities

Two directions emerge from this review.

Sammy Labs puts it directly: "Competition between products is too intense for companies to support users the way they did in 2010." The era of "your call is very important to us" – which in practice means "we'll get back to you with an email you won't understand on the first read" – is over.

Sammy is a more modern, faster, and more intuitive alternative. But the broader category of AI-native support experiences has many directions to explore. Last summer, Hark ([covered here](/review/uluchshat-nuzhno-otsjuda)) raised $5M for a platform that lets customers of physical goods retailers send short video clips showing exactly what their problem or question is, with AI analyzing and routing those clips to make support faster and smarter.

The other direction is new interfaces for software products – ones that let users accomplish complex goals quickly and simply, without navigating traditional UIs full of screens, buttons, toggles, and input fields.

Sammy is an example – for now an overlay on traditional UI, but a meaningful one. Adopt and the earlier version of OpenCopilot point in the same direction.

The trap to avoid: collapsing this idea into a single chat window. An open text prompt gives users freedom but not speed. Interior design platform Madespace ([covered here](/review/samoe-krutoe-sdelat-takoe-chemu-chatgpt-ne-konkurent)) made a deliberate choice to drop the free-prompt chat interface entirely. Its reasoning: most people don't design interiors from scratch – they start from images they found, then copy, combine, and modify. So the Madespace interface is built as a combiner and editor of existing professional designs, not a blank canvas for AI prompts.

The territory of genuinely new AI interfaces is largely unexplored. It can be explored inside your own product, or you can build the platform that lets other developers build new AI interfaces across many verticals.

What would the "dream interface" for your own product – or for the product that frustrates you most – actually look like? That's your starting point