AI Application-Layer Startup Radar — Week of May 26

Six AI application-layer companies disclosed rounds between May 26 and June 2, 2026, spanning AI coding agents, generative video infrastructure, consumer mini-app platforms, industrial simulation software, and a solo-founder workspace — collectively pulling in roughly $1.26 billion in new capital.

Cognition makes Devin, an AI coding agent that the company says writes 89 percent of the code committed by its own engineers. The two-year-old San Francisco startup closed a $1 billion Series D led by Lux Capital, General Catalyst Partners, and 8VC, at a $26 billion post-money valuation.1

Founder and CEO Scott Wu started coding at nine and made the US International Mathematical Olympiad team three times. He co-founded Cognition in 2023 alongside Harrison Chase and others he met through competitive-programming circles — the same cohort that produced Scale AI founder Alexandr Wang.2 The company also owns Windsurf, the AI coding environment it acquired in 2025.

Wu frames Devin as handling "long-tail maintenance" — migrating apps across platforms, updating dependencies — to free engineers for more original work. Cognition says its current annualized revenue run rate is $445 million, roughly 18 months after launch.3

Why watch: Devin is the only AI coding agent with disclosed enterprise revenue at this scale. Any direct competitor targeting the same "autonomous software engineer" positioning faces a product already embedded at meaningful depth inside paying engineering organizations.

OpenRouter operates a marketplace where developers route requests to 200+ language models through a single API and only pay for what they use. CapitalG, Google's independent growth fund, led the $113 million Series B — the firm's first check into an AI inference-routing company.1 The New York-based company has not disclosed its post-money valuation.

The OpenRouter team was not identified by name in publicly available filings this week. The round underlines the thesis that model-agnostic routing infrastructure — rather than any single model — may become the default interface layer for enterprise AI consumption.

Reactor exited stealth on May 28 with a $59 million Series A led by Lightspeed Venture Partners, with participation from WndrCo (Jeffrey Katzenberg's firm), Amplify Partners, Sky9 Capital, and FPV Ventures.4 AWS is the company's preferred cloud provider.

The company provides a unified SDK and API that lets developers run real-time generative video models in production without managing the underlying inference infrastructure. Lightspeed says the combined seed and Series A total $59 million.

Co-founders Alberto Taiuti (CEO) and Bryce Schmidtchen (CTO) previously led technical work on Apple Vision Pro, and Taiuti earlier co-founded Luma AI, where he served as CTO and built the inference stack behind one of the most-used 3D and video generation platforms.4 The broader team includes engineers from Apple, Netflix, Meta, Google, Adobe, and Replicate. Early customers include film and television studios as well as robotics companies.

Lightspeed partner Bucky Moore cited Taiuti and Schmidtchen's combination of real-time systems expertise and product vision as the reason to back a new category rather than wait for existing video infrastructure providers to build this layer.

Why watch: Reactor's infrastructure-for-AI-video position is directly competitive with any startup building real-time interactive experiences — game engines using generative AI, interactive video advertising, or robotic simulation. Founders building in those adjacent layers should expect Reactor to be a default infrastructure provider within 12-18 months.

Sekai is a mobile platform where users describe an app idea in a text prompt and receive a playable mini-app — similar to how short-form video enabled anyone to publish video content. The company raised $6 million in seed funding (led by Mayfield in 2025) and a $20 million Series A (co-led by Khosla Ventures and Connect Ventures), bringing total funding to $26 million.5

Founder and CEO Lucky Zhang has previously sold two companies: Yi+ AI (video e-commerce) to Apple in 2017, and Blacktail (a Latin America short-video platform) to ByteDance in 2020. He also founded NCT, a music streaming service in Vietnam that is still operating. Sekai, which means "world" in Japanese, began as an anime fan-fiction platform before pivoting to general mini-app creation after AI coding models improved sharply in 2025.5

By the time of the Series A close on June 1, users had created more than 15 million mini apps, with more than 200,000 generated daily. Average session time exceeds one hour per day. Connect Ventures co-founder Nicole Quinn (whose firm was co-founded by Creative Artists Agency and NEA) pointed to that engagement metric as the primary signal.

Zhang says he is building toward a creator monetization model, with CAA relationships potentially enabling celebrity partnerships on the platform.

Paris-based NP Company (NP Co.) builds AI-native simulation software for aerospace, defense, energy, and automotive engineering. It closed a €6 million pre-seed round led by Partech, with participation from the Peugeot family office and angel investors including Guillaume Lample and Cédric O (co-founders of Mistral AI), Florian Douetteau (founder of Dataiku), and Vincent Luciani (founder and CEO of Artefact).6

Founders Emmanuel Menier (CEO) and Matthieu Nastorg (CTO) are both PhD researchers in AI for simulation from Inria and Paris-Saclay University — the same Inria lineage that produced Mistral AI's founding team. The company's pre-trained transformer models run industrial physics simulations in seconds rather than the days or weeks required by legacy solvers.

NP Co. says it has demonstrated 1,000× speedups on benchmarks set by aerospace manufacturer Safran, with a claimed path to 50,000× on full-assembly problems. The funding will expand the research team and advance automated design tools and real-time operational simulators.

The Mistral founders' angel participation three days after Mistral AI acquired a competing simulation startup is notable: it suggests the co-founders are personally tracking the simulation-AI category independent of Mistral's corporate moves.

Harry Howard AI closed a $3 million seed round on June 1, 2026, from eight backers including Inovia Capital, Mantella Venture Partners, Neuralhash Capital, Skyland Ventures, and Vantec Angel Network.7

The Victoria, BC-based company builds a workspace for one-person companies, coordinating multiple specialized AI agents across idea validation, business plan generation, pitch deck drafting, MVP scoping, and daily operations. The founder goes by Harry; no full name was disclosed in available filings.

The platform's stated architecture combines large language models, agent orchestration, and persistent project memory in a cloud-native environment. Funding will expand the engineering team and accelerate feature rollouts.

The six deals point to three distinct application-layer bets playing out simultaneously: AI coding agents continuing to scale toward replacing whole engineering workflows (Cognition at $26B), real-time interactive AI creating new categories that don't yet have incumbent players (Reactor, OpenRouter's routing infrastructure), and consumer creation tools positioning AI output as a new medium of personal expression (Sekai's mini-app model). NP Co. and Harry Howard AI represent the earlier end of the funnel — pre-seed and small seed rounds that will take 18-24 months to emerge as visible competitors or acquisition targets for the larger players.

Geography is also notable: three SF-based companies, one Paris deep-tech, one NYC, one Canadian seed. The Inria-to-startup pipeline in France, visible in NP Co.'s founding story, is producing companies with technical depth that North American funders are beginning to track.