When Shopify revealed in May that an internal AI agent had authored one in every eight merged pull requests, the statistic landed somewhere between impressive and unsettling. But it also hinted at something more fundamental than a productivity boost: the possibility that the next entity evaluating your API documentation, testing your authentication flow, or deciding whether to integrate your software might not have a pulse.
That possibility is now hardening into reality, and a small San Francisco startup thinks it's spotted a business opportunity in the gap. Scope, which came out of stealth in mid-May with backing from Y Combinator, is building what amounts to quality assurance infrastructure for a world where AI agents don't just use software—they discover it, assess it, and increasingly make the call on whether to adopt it.
"Every internet product now has two users: the human who clicks and the model that reads," the company wrote in its launch memo. It's the kind of line that sounds like marketing copy until you consider the data piling up around it.
AI-powered search traffic climbed from less than 2% to over 9% of desktop queries in 2026, according to an April report from TechRadar Pro. Google's AI Overviews now surface in roughly 16% of results. GitHub said in May that more than one in five code reviews involve an agent. The shift isn't hypothetical anymore.
Scope's founder, Anand-Arnaud Pajaniradjane, spent time working on interpretability research at Princeton before pivoting into machine learning engineering focused on generative engine optimization—a corner of the industry concerned with how AI models surface and rank information. He started Scope in 2025 with a thesis that feels both obvious and surprisingly underexplored: if agents are becoming a primary interface for software discovery, then optimizing for agent experience is a distinct problem, not just an extension of human UX design.
The platform does two things, more or less. First, it simulates buyer queries across various AI models to track how often a product shows up, which competitors get mentioned alongside it, and which sources the agents cite. Then it runs actual browsing and coding agents through workflows—sign-ups, authentication sequences, API calls, error handling—to pinpoint where things break down. Visibility matters, but so does conversion. An agent that finds your product but can't figure out how to use it isn't much of a win.
What Scope outputs, according to its Y Combinator launch materials, is concrete: data on when an agent picks you over a competitor, where it stumbles, what specifically needs fixing. That might mean changes to documentation, adjustments to API design, tweaks to authentication flows, or clearer error messages. The company targets products that agents interact with directly—APIs, infrastructure tools, command-line interfaces, and servers built for the Model Context Protocol.
The platform organizes its work into three broad categories: monitoring (simulating how agents discover and complete tasks, running daily across real workflows), analytics (capturing tool calls, errors, friction points, latency, and the reasoning agents use), and what the company calls "action"—prescriptive fixes based on the data. It's a framework that mirrors traditional web analytics, adapted for a user base that doesn't browse so much as ingest.
At launch, Scope named two early customers: Blaxel, a YC-backed startup building persistent sandboxes for AI agents, and an unnamed decacorn. The company has raised funding from Y Combinator and, according to a June LinkedIn post from the European venture firm Newfund, secured backing from that investor as well, though terms haven't been disclosed.
The language around this emerging category is still unsettled, which is often a sign that the category itself is still forming. Scope uses "Generative Engine Optimization" and "Answer Engine Optimization" to describe discovery metrics, borrowing terminology that's cropped up in various corners of the industry over the past year. For usability testing, the company leans on "Agent Experience." Builder.io published a piece in early June titled "Developer experience is dead. Long live agent experience," arguing that coding agents require a distinct layer of context, tools, and feedback loops. Whether the terminology sticks remains to be seen.
The competitive landscape is fragmentary. OtterlyAI launched a public API and Claude skill in June. Profound offers an enterprise platform for multi-engine visibility. Several startups have clustered under the "GEO" and "AEO" labels, though most focus on content and brand visibility in AI search results rather than infrastructure. Scope distinguishes itself—at least in its positioning—by targeting developer tools and testing actual agent workflows instead of just tracking mentions.
Pajaniradjane runs the company with one other team member, operating out of San Francisco. The website offers a launch video, early access, and demo bookings. The YC directory lists Nicolas Dessaigne as the company's primary partner.
Perhaps what's most telling is Scope's framing of the problem: instrumenting and improving "agent experience" so AI agents can correctly discover, evaluate, and use software products. It's a bet that as agents become participants in procurement—reading docs, testing APIs, filing support tickets—the interface layer they navigate will matter as much as the one designed for humans. Maybe more.
Whether that's worth building a company around is the question Scope is in the process of answering. The two-person team has a thesis, some early traction, and a market that's moving fast enough that the definitions are still being written. That's either the perfect time to build infrastructure or the worst time to assume you know what the infrastructure should look like. Probably both.
