Your Users Are About to Tell You: Ditch the Interface

01 /The Concept

When a web service has broad functionality or handles moderately complex tasks, its interface tends to become genuinely intimidating for casual users. Getting up to speed requires deliberate effort that most users aren't willing to invest. Many drop off shortly after signing up, and others never unlock the full capability of the tool – which becomes its own reason to churn.

Adopt is addressing this by letting developers quickly and easily add AI agents to their web services – agents that dramatically simplify how users interact with those services.

Once Adopt is integrated, a familiar chat window appears in the product. Users describe what they want in plain language – run a query, filter data by specific criteria, trigger an action – and the AI agent carries it out. The same interface handles product questions: instead of hunting through documentation, users just ask and get an immediate, accurate answer.

The agent also builds memory across interactions, so it can remind users of relevant past work and suggest refinements. If someone asks how to send reminder emails to inactive customers, the agent might note that the user already created a matching segment three days ago – and offer to either trigger the send against that segment or update the criteria first.

Users can also set up recurring automations in plain language: "On the first of every month, generate a department expense report, flag all items over $10,000, and email it to the CFO for review."

For all of this to work, developers first need to teach the agent the service's capabilities. For simpler actions, the agent can learn by scanning the product's pages and documentation. For complex workflows, developers define them in plain language with the agent acting as an interactive configurator.

Adopt was founded last year and is currently in closed beta, with enough developer companies already participating to validate that the core product works. Despite the early stage, the company has raised a first round of $6 million.

02 /Why It Matters

The concept behind Adopt is solid. But it's worth noting that earlier startups building very similar things have since pivoted.

In late 2023, OpenCopilot ([related review](/review/skoro-on-pojavitsja-na-kazhdom-sajte-v-internete)) – an AI assistant for web services that went through Y Combinator – was doing essentially the same thing. Today, that company's primary focus has shifted to AI agents for customer support teams.

In spring 2024, TruvaAI ([related review](/review/interfejs-na-milliard-dollarov)), another Y Combinator graduate, raised $4 million for an "AI concierge for apps" built on smart chatbot principles. Today, it operates as an AI assistant for sales teams.

One likely explanation: those startups may have been early. Traditional web interfaces – with their toggles, menus, and buttons – are deeply familiar to users, and familiarity is hard to displace even in favor of something more powerful.

But a year or two has passed since those attempts, which is a long time in the current environment. What's changed?

Andrej Karpathy, one of the defining voices of the current AI moment and the person who coined the term "vibe coding," argues that the primary audience for web services is no longer humans – it's AI agents.

This is happening for three interconnected reasons.

First, people want more accurate answers from AI chat interfaces like ChatGPT than a conventional web search can provide. For that, AI systems need to pull structured information from web services – data currently locked in internal databases. The only current path is having the AI understand how to formulate a search query, then scrape and parse results from standard web pages. That's slow and brittle.

Second, people want their AI assistants to answer questions not just from the open web but from the specific services they use. This requires not just data access but authenticated access – logging into services on users' behalf. Anon ([related review](/review/bez-jetogo-oni-nichego-poleznogo-dlja-tebja-ne-sdelajut)) raised $6.5 million specifically to solve the authentication piece for AI agents acting on behalf of users in third-party applications.

Third, people want their AI to not just answer questions but take actions inside other services – from purchases to workflow tasks. For that to work reliably, services need machine-friendly interfaces. Having AI simulate human clicks in a browser works in theory but is error-prone, and the cost of mistakes is higher than for read-only access.

Now consider what happens when a developer adds Adopt to their service: they're effectively creating an interface that external AI agents can use.

Adopt reinforces this by letting developers expose their service as an MCP server – enabling external AI agents to interact with it via the MCP protocol introduced by Anthropic late last year.

So adding Adopt delivers two benefits in one package: human users can use the service to its full potential without studying its interface, and external AI agents get clean, standards-based access through MCP. In that combined form, this is something that could genuinely take off.

03 /Opportunities

A standardized interface through which external AI agents can interact with web services is, in effect, the "new API." This concept has more tailwind behind it now than earlier generations of API tooling did – because an ever-growing share of users wants to interact with the digital world through AI chat interfaces.

That means the same categories of platforms that succeeded in the API economy are likely to succeed again – in a new form. Platforms like Adopt itself are the most direct play: tools that help developers add AI agent interfaces to existing services. One layer up, the opportunity is building unified interfaces that translate many disparate service APIs into a single call – the way Plaid did for bank connections. Deeper still, platforms like Deck ([related review](/review/hochesh-stat-monopolistom)) take it further by building API access to services that have no public API at all, using AI to parse web interfaces and expose the data through a standardized endpoint. And for B2B SaaS developers who need to ship integrations fast, platforms like StackOne ([related review](/review/vzletajushhaja-tema-dlja-nastojashhih-ajtishnikov)) offer pre-built modules for hundreds of enterprise services without building each one from scratch.

The broader opportunity is the market for "new APIs" – infrastructure that enables AI agents and traditional web services to interoperate. Including, eventually, transforming web services into agent-native products the way "API-only" products emerged – no conventional web interface at all, just programmatic access.

This space is large, new, and still early. There's room for many things Adopt is doing, because there are a lot of developers who need it – and one platform won't be enough.