AI application-layer startup radar: July 15–22, 2026

Executive read

Nine dollar-denominated funding announcements in the July 15–22 window disclose $333.2 million in current-round capital, alongside €15.06 million and one undisclosed pre-seed. The largest check went to Glow, which emerged from stealth with a $180 million Series A for endpoint security around AI agents and developer tools. The more useful pattern is underneath the headline: application companies are filling the control gaps that appear after an agent leaves the demo. They are handling permissions, deployment, payments, workflow-specific data, and human handoffs.

The second cluster is narrower vertical software. Travel, education, energy, construction, legal services, fashion retail, and creator marketing each received a new application-layer bet. Several have a clear wedge into an existing operating workflow; others are still early enough that the product description is doing more work than the traction disclosure.

Quick scan

Company	Sector	Window signal	Product positioning	Founding-team signal
Glow	AI agents / security	$180M Series A at a $1.2B valuation; July 22	Endpoint security for software, agents, and developer tools	Former Meta, Snowflake, and Claroty leaders 1
Fora	Vertical SaaS / travel	$60M Series D at a $1B valuation; July 16	Platform and AI assistant for independent travel advisors	Evan Frank is the only named co-founder in the opened release; prior background not disclosed 2
Natural	AI agents / fintech	$30M Series A led by Forerunner; July 20	Payment rails for agents that move and store funds	Kahlil Lalji previously founded Ivella; co-founders include an Ivella partner and a former Nextdoor engineering manager 3
Arrakis	AI agents / industrial	$30M Series A led by Blossom; $38M total; July 22	Forward-deployed agents for industrial operations	Ex-Accel, Palantir, and Delivery Hero team 4
Passionfroot	Vertical SaaS / marketing	$15M Series A led by Insight Partners; July 22	B2B creator marketplace with an AI campaign agent	CEO and co-founder Jen Phan; full founding background not disclosed in the opened source 5
Modo Energy	Vertical SaaS / energy	€14.9M growth funding; July 22	Benchmarking and valuation for energy storage and renewable assets	Quentin Scrimshire and Tim Overton; prior affiliations not stated 6
Aina	Consumer AI / hardware interface	$5.5M round; July 16	Dune macro keypad for controlling meetings, apps, and agent workflows	Apoorv Shankar previously led hardware at Ultrahuman and founded LazyCo 7
DeweyLearn	Vertical SaaS / education	$5M Series A led by SJF Ventures; July 16	Multimodal assessment of real-world skills	Luyen Chou is a former 2U chief learning officer and Pearson chief product officer; Dirk Liebich is co-founder and CTO 8
Vikk AI	Vertical SaaS / legal	$700K pre-seed plus $3.5M seed; July 21	Legal answers, document help, and sponsored lawyer discovery	Walid Romaya is the named co-founder and CEO; earlier affiliations were not disclosed 9
Cascade	Vertical SaaS / construction	$3.5M seed led by a16z Speedrun; July 21	Predictive project discovery for architecture, engineering, and construction firms	Hannia Zia and Joana Ferreira are former Google operators who met at UnlikelyAI 10
Sowilo	Vertical SaaS / fashion	Undisclosed pre-seed; July 21	Catecut turns product images into retail metadata and copy	Heiðrún Ósk Sigfúsdóttir; prior affiliations not disclosed 11
goNEON Agentic Systems	AI agents / infrastructure planning	€160K from Venture Kick; July 20	Regulation-aware infrastructure design generation	Raphael Eder and Dr. Lukas Ballo; source gives broad experience only 12

Funding profiles

Glow: endpoint security moves inside the agent workflow

Glow emerged from stealth on July 22 with a $180 million all-equity Series A at a $1.2 billion valuation. Sequoia Capital, Cyberstarts, Greenoaks, and Redpoint Ventures led the investor group, with Index Ventures, Swish Ventures, Lux Capital, Operator Collective, and Holly Ventures also participating. The company is building an endpoint-security platform that monitors and controls software, AI agents, and developer tools running inside enterprise environments. Its stated emphasis is prevention before a risky action reaches the environment, rather than detection after the fact. 1

The founding group includes CEO Roi Tiger, formerly a vice president of engineering at Meta; Omer Singer, formerly Snowflake's head of cybersecurity strategy; Ophir Arie, formerly Claroty's vice president of R&D; and Arnon Joseph, formerly an engineering leader at Meta. The team background is unusually legible for an early security company. The diligence question is whether Glow's endpoint position gives it enough telemetry and enforcement authority to become a control point, rather than another monitoring console.

Natural: agents need a payment system they can actually use

Natural raised a $30 million Series A led by Forerunner, bringing its total funding to $40 million. The startup is building financial infrastructure that lets AI agents move and store funds, make autonomous payments, collect money, and transact with people or other agents. It has operated in beta and is positioning itself against payment incumbents that were designed around human authorization. 3

CEO and co-founder Kahlil Lalji previously founded Ivella, a YC-backed banking product for couples, which was sold to Earnin in 2023. He later worked there as an engineer. Co-founder Eric Wang also worked with Lalji at Ivella, while Walt Leung was an engineering manager at Nextdoor. That is a payments team with a prior product loop and a clear technical wedge. The open question is less whether an agent can initiate a payment than who owns authorization, disputes, fraud liability, and the customer relationship when the buyer is software.

Arrakis: industrial agents arrive with people attached

London-based Arrakis has raised $30 million in a Series A led by Blossom Capital and reports $38 million in total funding after just over three months. Accel participated, alongside Datadog co-founder and CEO Olivier Pomel, OpenAI business-product leader Olivier Godement, and other individual investors. Arrakis is building and deploying agents for aerospace, energy, logistics, manufacturing, construction, and telecommunications companies. Its model-agnostic platform uses forward-deployed AI engineers to work with a customer's data and operational constraints. 4

The founders bring a finance-and-operations profile: Rafael Quintanilla was previously an investor at Accel; Haroun Beltaifa and Romain Fouilland came from Palantir; and Mikhail Galkov was an engineer at Delivery Hero. Arrakis is making a services-heavy bet at a time when many agent startups promise self-serve deployment. That can be an advantage in industrial settings, where process knowledge and integration work are part of the sale, but it can also cap software-like margins if each deployment stays bespoke.

Aina: a physical control surface for agents

Aina raised $5.5 million in a round led by Redstart Labs and 360 ONE, with participation from MIXI Global Investments, Antler, Blume Founders Fund, and several angel investors. Its first product, Dune, is a three-button, context-aware macro keypad designed to control meeting microphones and cameras, run application-specific shortcuts, and trigger scripts or agent workflows. The company plans to pilot the device in the coming weeks. 7

Founder Apoorv Shankar previously served as Ultrahuman's VP of Hardware. Before that, he founded LazyCo, a hardware-interface company later acquired by Ultrahuman. Aina is adjacent to the software application layer rather than a pure SaaS company, but it fits the week's control theme: the product is designed to initiate actions, not merely capture context. The near-term diligence test is whether a dedicated input device earns a place beside keyboard shortcuts, voice commands, and native agent interfaces.

goNEON: a small round around a concrete planning workflow

ETH spin-off goNEON Agentic Systems secured €160,000 from Venture Kick on July 20. Its agentic AI platform generates infrastructure designs against engineering requirements, local regulations, and real-world constraints, with the stated goal of shortening feasibility work from weeks or months to minutes. Founders Raphael Eder and Dr. Lukas Ballo combine startup, AI, and urban-planning experience; the opened source does not name prior employers or institutions. 12

The amount is small beside Glow or Fora, but the product boundary is easier to evaluate. A diligence call can ask which regulations are encoded, which design decisions remain human-approved, and whether a generated plan survives review by an engineering authority. That is a better test than a generic claim that an agent can automate planning.

Fora: AI as leverage for a human travel-advisor network

Fora closed a $60 million Series D at a $1 billion post-money valuation. Forerunner and Tactile Ventures led the round, with Thrive Capital, Insight Partners, Heartcore Capital, PLUS Capital, BlackPines Capital Partners, and Tribeca Venture Partners participating. Fora provides infrastructure, training, and community for people who want to become travel advisors, while also helping travelers connect with those advisors. Its Via assistant handles research, destination and supplier knowledge, and itinerary work. 2

Fora said advisors on the platform have booked more than $3 billion in travel since launch. The company is using the round to expand Via and add categories such as cruises and flights. The release names Evan Frank as a co-founder but does not disclose a complete founding team or his prior affiliations. That gap matters because the company is underwriting a network business: the defensibility question is whether AI improves advisor productivity and retention, or simply lowers the cost of producing itineraries for any competing marketplace.

DeweyLearn: assessing skills that are hard to reduce to a quiz

DeweyLearn secured a $5 million Series A led by SJF Ventures. Its multimodal platform combines audio, video, and learning data with domain knowledge and learning science to assess real-world performance, then deliver expert feedback in clinical education, higher education, professional learning, and K–12 settings. 8

Co-founder and CEO Luyen Chou previously served as chief learning officer at 2U and chief product officer at Pearson, and founded The School at Columbia University. Co-founder and CTO Dirk Liebich is described as an AI, data-analysis, and predictive-analytics specialist. The team gives DeweyLearn a plausible distribution path into education, but the diligence burden is high: a model that scores observable performance still needs validated rubrics, institution-level trust, and a defensible process for appeals.

Modo Energy: an AI analyst inside a regulated data product

Modo Energy secured €14.9 million in growth funding from CIBC Innovation Banking, its largest single commitment to date according to EU-Startups. The London company provides benchmarking and valuation tools for energy storage, renewable assets, and flexible loads. Its platform supports project finance, investment diligence, and operating strategy, and includes an AI analyst called Ko. Modo says it is an FCA-regulated benchmark provider for battery energy storage. 6

The named founders are Quentin Scrimshire and Tim Overton. The opened source does not provide prior affiliations. Modo is a useful example of an application company where the AI layer is not the entire product. The underlying benchmark, regulatory status, and customer workflow may matter more than the assistant interface. Investors should separate the value of the data product from the incremental value of Ko.

Cascade: selling foresight before a construction tender exists

Cascade raised a $3.5 million seed round led by Andreessen Horowitz's Speedrun, with Ada Ventures, Blitzscaling Ventures, Indico Capital, shuckerVC, G2C Ventures, and Snowball VC also participating. The company uses AI to identify construction and engineering projects before they become public opportunities by reading bond filings, permits, budget announcements, capital plans, property transactions, earnings transcripts, and meeting minutes. It says it has surfaced more than $10 billion in project opportunities for customers. 10

Hannia Zia and Joana Ferreira are former Google operators who met at UnlikelyAI. The product sits in a familiar but valuable place in a vertical workflow: before the bid, not after the contract. The diligence question is whether Cascade's signals produce win-rate or pipeline improvements that customers can attribute to the system, rather than simply giving sales teams a larger list of possible projects.

Vikk AI: legal search becomes an advertising market

Vikk AI disclosed $4.2 million across a $700,000 pre-seed and a $3.5 million seed. Its consumer-facing legal assistant answers questions in natural language, reads documents, drafts simple legal files, and identifies lawyers handling similar matters. The company's commercial focus is a Lawyer Ad Center that sells sponsored placements matched to user needs and location, plus a directory and data product for law firms, insurers, and legal-aid organizations. 9

Walid Romaya is the named co-founder and CEO. The opened report does not provide his prior company affiliations or a complete founding team. Vikk therefore has two separate diligence questions: whether its legal answers are reliable enough to create user trust, and whether sponsored discovery can be introduced without weakening that trust. The second question is a product-design issue, not just a revenue-model detail.

Sowilo: product metadata as a retail workflow

Iceland-based Sowilo raised an undisclosed pre-seed to expand Catecut, its fashion-product intelligence platform. Catecut identifies apparel products and design attributes from images, then generates titles, descriptions, metadata, tags, multilingual copy, and image alt text. It is also designed to update product pages and improve discovery in search and generative-AI systems. 11

The named founder is Heiðrún Ósk Sigfúsdóttir; the source does not disclose prior affiliations or a complete founding team. The company is targeting a repetitive merchandising task with a measurable output: cleaner, faster product information. The unanswered question is whether retailers will pay for the metadata workflow itself, or expect it to be bundled into commerce platforms and catalog-management systems.

Passionfroot: a creator marketplace for a noisier B2B market

Berlin-based Passionfroot raised a $15 million Series A led by Insight Partners, with Creandum, Supernode Global, and s16vc participating. The company connects B2B creators with brands, provides a wallet for creator payments, and uses a proprietary creator graph to match campaigns with reach and performance data. Since its last fundraise, it has released Zest, an AI agent that helps brands create, execute, and monitor campaigns. 5

CEO and co-founder Jen Phan said revenue grew 13 times over the past year and that the company added clients including ElevenLabs, Figma, Replit, Framer, and Gamma. Those are company-reported figures. Passionfroot also says it paid at least $10 million to creators in the last 18 months. The investment case depends on whether the marketplace's data and payment layer compound as AI increases the number of software products competing for attention, or whether creator discovery remains a fragmented service business.

Product launch checkpoints

Product Hunt's latest 24-hour snapshot surfaced several application-layer launches. These are positioning and distribution signals, not funding evidence or proof of adoption.

Product	Positioning in the launch listing	Signal
Humalike x Hermes	Social-intelligence plugin for an agent that works in Slack, Telegram, and WhatsApp	Agents are being packaged for group-chat context and memory 13
Buzzy	Agentic infinite canvas for video ideas, moodboards, and long-form production	Creative agents are moving from generation toward coordinated editing and consistency 14
Migma AI	Agent that creates, localizes, sends, and learns from email campaigns	Marketing software is presenting the agent as the campaign operator, not a copy assistant 15
Kastra	Runtime authorization for agents across tools such as Claude Code, Cursor, and Codex	Permissions and policy enforcement remain a distinct application category beside Glow and Natural 16

What changed this week

1. The agent control stack now includes money and endpoints

Glow controls what software and agents can do on enterprise endpoints. Arrakis takes responsibility for deployment inside industrial operations. Natural is building the payment layer needed when an agent must complete a transaction. Aina puts an action-oriented control surface on the user's desk. These are different businesses, but they address four different points of failure: visibility, deployment, authorization, and execution.

The Product Hunt signal points in the same direction. Kastra frames runtime authorization as a product, while Humalike frames group-chat context and memory as a product. The category is spreading horizontally. That creates room for specialists, but it also makes integration and ownership more important than a generic claim of agent autonomy.

2. Vertical AI is moving closer to the system of record

The vertical entries are not all the same kind of company. Fora owns a travel-advisor network and adds an assistant. Modo combines an AI analyst with a regulated benchmark product. DeweyLearn couples model output to assessment methodology. Cascade builds a project-opportunity data set before a construction bid. Vikk is combining legal assistance with a lead marketplace. Sowilo is embedding generation into catalog operations.

The common diligence question is where the workflow lives. A thin assistant can be copied by a horizontal platform. A product that owns the benchmark, the customer network, the evidence trail, or the transaction may have more room to compound.

3. The capital bar ranges from program money to a billion-dollar security bet

This week's disclosed amounts span €160,000 for goNEON, $3.5 million for Cascade, and $180 million for Glow. That spread is more informative than a single total. Investors are funding small, domain-specific experiments while also placing very large bets on control points that may sit across every enterprise agent deployment.

The large rounds are not proof that the application layer is settled. They are bets that the winning control surface will become hard to remove once it is attached to identity, endpoints, deployment, or payments. The early rounds are testing whether narrower workflows can become that control surface within one industry.

4. Founder evidence is uneven, and that is itself a diligence signal

Several teams have a strong operating history in the problem area: Glow's security leaders, Natural's prior fintech product, Arrakis's Palantir background, DeweyLearn's education operators, and Aina's hardware experience. Other profiles in the opened sources name the founder but do not provide prior affiliations, including Modo, Sowilo, and Vikk. That does not weaken the companies by itself, but it does mean the public record is not yet sufficient for a confident team assessment.

For a time-constrained investor, the next step is straightforward: separate disclosed facts from claims that still need a call, a data-room check, or a second source. The gap is part of the map.

Diligence questions for this week's shortlist

For agent-control companies, who owns the permission, the audit trail, and the liability when an action goes wrong?
For vertical products, what customer workflow or data asset would be difficult for a general-purpose model vendor to reproduce?
For companies reporting fast revenue or usage growth, which figures are company-reported, and what is the denominator: registered users, active customers, paid accounts, or processed volume?
For network businesses such as Fora and Passionfroot, does AI increase repeat usage and take rate, or only reduce the cost of a service that remains hard to scale?
For early companies with limited public team history, which founder-background claims are still unverified before a partner meeting?

Coverage note

This radar covers public funding and product disclosures dated July 15–22, 2026, using the source publication date where available. Companies already covered in the July 15 radar were not repeated. Model-layer and AI-infrastructure rounds were excluded, as were physical-AI infrastructure companies whose primary product is robotics deployment rather than application software. Product Hunt entries are included as launch-positioning signals only; they do not establish financing, traction, or a complete founding history.

Where an opened source did not disclose the full founding team, prior affiliations, or a funding amount, the gap is stated rather than filled from an unverified database.