Three September 17 Series A rounds, totalling $93 million, share a structural premise: agent deployment is not coming — it has already happened, and these are the businesses that only exist because of it.
17 September 2026 — Three Series A Rounds
$93M
Total raised, three rounds
$35M
Raindrop Series A, led by CRV
$34M
Comp AI Series A, co-led by Roo Capital & Grand Ventures
On September 17, 2026, three companies announced Series A rounds on the same day. Raindrop closed a $35 million round led by CRV, bringing its total funding to $50 million, with Lightspeed Venture Partners and Y Combinator participating alongside lead researchers from OpenAI, Anthropic, and Thinking Machines. Its product reads production agent trajectories to detect silent failures — hallucinated answers, tool misuse, and behaviour changes introduced by a model upgrade — and the round accompanies the launch of Simulations, which tests agent changes against production traffic before they ship.
Comp AI raised $34 million in a round co-led by Roo Capital and Grand Ventures. Founded in January 2025 and headquartered in Miami with a New York office, the company sells AI agents that write security policies and gather audit evidence, and is expanding from compliance automation into continuous cybersecurity — real-time monitoring, control validation, and security testing across applications and infrastructure. The company states it has more than 1,000 customers and 15× year-on-year ARR growth; both figures are the company’s own and have not been independently verified.
Kastle raised $24 million led by Insight Partners, with Y Combinator and Commerce Ventures continuing and Fifth Wall plus a group of founders and financial-services executives joining as new investors. It builds an AI workforce for financial services, beginning with consumer lending, running agents across existing core systems rather than replacing them. The company states its agents have processed more than $1.8 billion in transactions — a figure that is the company’s own.
Round Timeline
January 2025
Comp AI is founded, headquartered in Miami with a New York office.
Pre-17 Sept 2026
Raindrop accumulates $15M in prior funding before closing its Series A; total reaches $50M on announcement.
17 September 2026
Raindrop ($35M), Comp AI ($34M), and Kastle ($24M) announce Series A rounds on the same day — $93M in aggregate.
18 September 2026
California Executive Order N-9-26 is issued the following day; among its proposals under consideration: safety frameworks verified to an external standard. The proposals are under consideration and not requirements in force.
The key insight: These are not three bets that AI will get better. Each of the three businesses only has a market if agent deployment has already happened. The rounds price a fact rather than a forecast — and the product requirements are the tell.
None of the three is a bet on models improving. All three are bets that agents are already doing work whose consequences someone has to account for.
The Structural Read
The Business Engineer lens here is pricing a fact rather than a forecast. Read together, these three rounds describe the shape of a market that requires prior deployment to exist at all. That is the analytical move worth making precisely — it is a reading of what each product requires in order to have customers, not a claim about how widely agents are deployed in production generally.
Silent failure is a category that only exists where agents run unattended at volume. A supervised system fails loudly, because a person is watching. Raindrop’s product — reading trajectories to catch hallucinated answers, tool misuse, and behaviour changes introduced by a model upgrade — has no addressable problem in a world where humans review every agent action. The product requirement is the deployment tell.
Comp AI’s proposition depends on a point-in-time audit having become a poor approximation of the thing it approximates, which is only true where the configuration being audited changes between audits. Static environments do not degrade periodic checks. The move from gathering evidence once to validating controls continuously is only commercially meaningful if the underlying system is already moving fast enough to make the snapshot stale.
Kastle’s company-stated figure of more than $1.8 billion in transactions processed is the most direct statement of the same premise. It is not a forecast. It describes a pipeline that has already run.
Business Engineer — Pricing a Fact Rather Than a Forecast
The Deployment-Tells Framework
When a product’s existence requires a prior condition rather than a future one, the funding round is evidence of that condition having been met — not a bet that it will be. The product requirements are not market projections; they are market readings. Silent failure, stale audits, and processed transaction volumes all read the same way: the prior condition is deployment at scale, unattended and consequential.
The second structural observation is about what each company says it sells. Raindrop and Comp AI both sell the ability to observe and attest — trajectories read, controls validated, evidence gathered continuously rather than at a single moment. That is a verification layer being capitalised as a commercial category. Kastle sells something structurally different: it sells the labour itself, describing hybrid teams with agents on repeatable volume and people on judgement.
Separately, and without claiming any connection, causation, or coordination, verification is also the instrument California reached for the following day in Executive Order N-9-26, whose proposals under consideration include safety frameworks verified to an external standard. Those are proposals under consideration, not requirements in force. Nothing here suggests the funding anticipates regulation, or that regulation will create demand, and nothing is predicted.
The third structural observation is about the word doing the most work across all three announcements: continuous. Raindrop adds Simulations to test changes against production traffic before they ship. Comp AI describes moving from gathering evidence for an audit to validating controls in real time. This is a general property of periodic versus continuous services worth naming clearly: a periodic check prices as a project, and a continuous one prices as a subscription — a different business with a different retention profile and a different relationship to the customer’s calendar. That is not a claim about any of these three companies’ pricing, contracts, revenue, or retention, none of which is established here.
Kastle’s stated constraint deserves naming precisely, because it is the part most often skipped. Its agents run across existing core systems rather than replacing them. An institution that cannot replace its core systems has historically had to choose between living with their limits and financing a multi-year rebuild. A layer that operates on top of those systems is a third shape — not a better version of either option. Nothing here claims that this approach works, is superior, is novel, or is defensible. No institution is named, and no outcome, saving, efficiency, or quality effect is claimed.
Three Implications
IMPLICATION 1 — SELLING EVIDENCE IS A COMMERCIAL CATEGORY NOW
Raindrop and Comp AI both sell the ability to observe and attest to what agents have done or are doing. That a verification layer is being capitalised at this size, at this moment, is the structural signal. The question for any enterprise buyer is not whether they need visibility into agent behaviour — it is what form that visibility takes and how continuously it is produced. A periodic audit that was sufficient when systems changed quarterly is a different instrument from one that captures configuration drift in real time.
IMPLICATION 2 — THE PERIODIC-TO-CONTINUOUS SHIFT CHANGES THE BUSINESS SHAPE
The general property is this: a periodic check is scoped to a moment and priced as a project. A continuous one is scoped to a relationship and priced as a subscription. These are not the same business. The customer’s procurement cycle, renewal dynamic, and dependency on the vendor all change when the service is always on rather than episodically engaged. That is a structural observation about the shape of the category being built — not a claim about any of these three companies’ contracts or revenue.
IMPLICATION 3 — THE ON-TOP-OF-THE-CORE APPROACH IS A THIRD ARCHITECTURAL SHAPE
Institutions with legacy cores have historically faced a binary: operate within their constraints or fund a replacement programme measured in years. Kastle’s stated approach — agents running across existing systems rather than replacing them — is structurally a third option. Whether that option proves durable, defensible, or commercially superior is not established here and is not claimed. What is established is the shape of the problem it addresses: the constraint is real, the binary has been expensive, and a third path is being attempted.
Business Engineer Framework
Map of AI — Where Raindrop, Comp AI, and Kastle Sit in the Stack
The Map of AI traces 200+ companies across 9 layers of the AI stack — from infrastructure and models through to orchestration, observability, and application. These three rounds cluster at the layer the Map calls the governance and verification surface: the stratum that emerges when consequential work is already being done by agents and someone has to be able to attest to what happened. Understanding where a company sits in this stack determines who its real competitors are, where its pricing leverage comes from, and how sticky its relationship with the customer is likely to be. The Map gives you the orientation to read the next round before the press release does.
Ninety-three million dollars moved on September 17 into three businesses that have no customers in a world where agents are not already running unattended at volume. That is not
Comp AI’s figures of more than 1,000 customers and 15× year-on-year ARR growth, and Kastle’s figure of more than $1.8 billion in transactions processed, are the companies’ own statements and are not independently verified here. All three companies are private. No share price, market capitalisation or valuation is stated for any of them, and no revenue figure in currency, customer name, headcount, price, market size or competitor comparison appears above. The argument above concerns what each product requires in order to have customers; nothing above claims how widely AI agents are deployed in production generally. No investor’s reasoning or motives are characterised, the three companies are not ranked, and their quality, prospects and defensibility are not compared. Where California’s Executive Order N-9-26 is mentioned, its proposals are under consideration rather than requirements in force; no connection, causation or coordination with this funding is claimed, and nothing above suggests that regulation will create demand. Nothing above claims that any of these approaches works, is superior, novel or defensible, names any bank or institution, or claims any outcome, saving, efficiency or quality effect. Nothing is predicted. No quotation is attributed to any founder, investor or executive. This is businessanalysis. It is not investment advice, no view is expressed on any security, and no recommendation is made.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
