Here is the summer in one sentence, citable from the top: the biggest AI-agent checks in July and early August 2026 skipped the model labs to fund the operational layer around agents — and that money now flows in three distinct lanes: control the agents, own a regulated vertical, or sell the factory that builds the software. A month ago we mapped two of those lanes. The third one just started writing nine-figure checks.
The scoreboard, one month later#
Nothing in the last month broke the July thesis. It sharpened it. On August 3, Zenity closed a $125 million Series C led by Norwest Venture Partners — with SoftBank Vision Fund 2, Hitachi Ventures, LG Technology Ventures, and Qumra Capital coming in new, and existing backers DTCP, Vertex Ventures, Third Point Ventures, and Intel Capital following on. That takes the company to roughly $185 million raised (BusinessWire, SiliconANGLE). It was the biggest disclosed agent round of the window, and it landed in the lane July already told us was on top: control.
The other two lanes didn't go quiet. They clarified. Below is the map a founder should actually keep.
Lane one: control — priced on visibility#
The control lane sells one thing: the ability to see the agents you're already running. Neo left stealth on July 20 with $100M for "agentic software control." Zenity's $125M is the same bet with a bigger customer base behind it — Gartner has called Zenity "the company to beat in AI agent governance," and its customers span Fortune 500 and Global 2000 firms in financial services, healthcare, pharma, manufacturing, and tech. The founders are Unit 8200 veterans Ben Kliger (CEO) and Michael Bargury (CTO); this is a security company, top to bottom.
You cannot secure, bill for, or switch off an agent you cannot see. The freshest money of the maturity phase went, again, to the people who make agents visible.
What it means for you: the buyers just told you agent sprawl is a board-level problem. If your wedge is control, your entire job is to make agents legible and prove it — inventory, attribution, policy. And you can start that discipline at your own scale today: take stock of every agent you run before a customer's security team does it for you. We covered the fresh Zenity round and its founder implications in Zenity raised $125M to police a billion agents.
Lane two: the regulated vertical — priced on liability#
The second lane was July's valuation champion, and the multiple has an unglamorous explanation: liability. Norm AI raised $120M at a $1.2B valuation for "agentic law" — software that interprets rules, monitors compliance, and governs how other AI behaves in regulated environments. Harvey took $200M at $2.1B for owning legal work end to end (TechCrunch).
The premium isn't for a smarter model. It's for a company willing to stand behind a compliance-heavy workflow that a horizontal chatbot will never touch. When Norm AI's backer Blackstone is also its customer, that's the whole thesis in one line.
What it means for you: if you're in a vertical, go narrower than feels comfortable and own the risk. The defensibility is the part your competitors are afraid to sign up for.
Lane three: the software factory — priced on throughput#
This is the lane that graduated to nine figures. 8090 — launched in 2024 by Chamath Palihapitiya — raised a $135M Series A led by Salesforce Ventures on June 26 for its "Software Factory," a platform that pulls people and agents into a single environment spanning business intent, requirements, architecture, work orders, code, testing, and production maintenance (BusinessWire). We unpacked what that model implies in the AI software factory, explained.
Lyzr rhymes with it from the control-plane side: a Series B of roughly $100M at a ~$500M valuation, selling infrastructure to build and operate agents inside a customer's own private cloud so data never crosses the perimeter. The detail everyone quoted — Lyzr's own agent, SivaClaw, fielded 130-plus investors and drafted the memos before humans closed terms (TechCrunch) — is a demo disguised as a press release. The product ran the raise.
What it means for you: the factory lane doesn't sell a feature, it sells shipped software. If that's your wedge, your north-star metric is throughput — intent to production — and your competition is the customer's own engineering team, not another chatbot.
The one decision the map forces#
Three lanes, three scarcities: control prices visibility, the vertical prices liability, the factory prices throughput. They are not interchangeable, and the ~$1.8B-plus that moved this summer is unanimous on the part that is interchangeable — the model. The durable money went to the layer around it.
So the decision isn't "which model." It's "which lane," and the honest version of that question is which scarce thing am I actually selling — and can I be the one who owns the mess agents leave behind in it. Pick the lane your wedge already sits in, price against its scarcity, and stop trying to out-raise a lab. The operational layer is reachable at any size. The model isn't.
