Agentic Defined Networking is moving from research talk into funded product form. Miami-based doxx.net closed a $38 million Series A led by Andreessen Horowitz, with Animo Ventures and Focal.vc joining. The company also opened a beta of its Agentic Defined Networking platform, often shortened to ADN. For India and USA delivery teams, the story matters because AI agents already browse and act with client credentials.

This guide walks through what ADN claims to do, which facts are verified from public sources, and how agency teams should pilot agent networking without creating new risk. You will leave with a practical checklist, not hype.

Abstract illustration of a private mesh network and security shield for AI agents — doxx.net / Andreessen Horowitz announcement artwork. Image credit: a16z investing announcement assets.

What Agentic Defined Networking means in practice

AI agents now open pages, call APIs, move files, and talk to services. They often do this under a user’s accounts and authority. doxx.net says those agents still struggle to tell safe destinations from unsafe ones while they work.

Agentic Defined Networking is the company’s answer. ADN gives agents an environment with defined connectivity and built-in threat protection. The goal is to stop agents from reaching known malicious destinations while they browse and act.

Andreessen Horowitz described the same idea in its investment note. Agents expand the surface you must protect because they carry your data and permissions. Where an agent connects, and who can observe that path, becomes a delivery concern for every agency project that ships agent features.

Defined paths: Agents operate inside connectivity rules you set.
Threat blocking: DNS-level controls try to stop bad destinations early.
Human plus agent use: The network is built for people and agents together.
Policy ownership: Users set privacy terms through architecture, not a policy PDF alone.

Verified funding and company background

According to SecurityWeek’s October 3, 2026 report, doxx.net raised $38 million in Series A funding. Andreessen Horowitz led the round. Animo Ventures and Focal.vc participated.

The firm was founded in 2025 by Barrett Lyon, who serves as CEO. SecurityWeek notes the company operates across six continents. The funding announcement arrived with the ADN open beta.

Andreessen Horowitz’s investment post (October 1, 2026) adds founder context. Lyon helped pioneer modern DDoS defense with Prolexic, later acquired by Akamai. He later founded BitGravity, acquired by Tata Communications, and Defense.net, acquired by F5. That history is useful when you evaluate whether the team can run global networking claims.

Joel De La Garza, a partner at Andreessen Horowitz, sits on the doxx.net board. In SecurityWeek’s coverage he said the internet was never designed for privacy, and that doxx.net aims for privacy by architecture rather than marketing language.

How the ADN platform is described

Public reporting describes ADN as a parallel, private networking platform. Humans and AI agents can connect, communicate, and browse without relying on central intermediary servers for those flows.

doxx.net argues intermediary servers create several problems:

They create a single point for interception and surveillance.
Privacy promises often rest on policy, not architecture.
They can raise the chance an agent lands on a malicious site.
They can be censored or blocked.

Feature claims covered by SecurityWeek and a16z include:

Serverless peer-to-peer communication with end-to-end messaging and file transfers
Threat protection that blocks malicious destinations, malware, and phishing sites
Agent integration with API control
Privacy and anti-tracking features
Native apps for iOS, macOS, and Android, plus Windows and Linux support
A global backbone operated by the company

Customers also get a private naming system. Reporting cites 196 custom top-level domains and doxx.net’s own certificate authority. ADN networks can resolve domains the public internet does not know, without third parties in that resolution path.

Diagram of an AI agent path filtered by DNS threat protection with safe and blocked destinations — Conceptual view of agent traffic screened before it reaches the open web. Illustration created for Debugged Pro.

DNS threat protection and the 38 million claim

Threat protection is described as DNS-level filtering. Known and emerging malicious destinations are blocked before an agent or person connects.

doxx.net claims it blocked more than 38 million threats since December 2025 during closed beta. Treat that as a company-reported figure. Independent researchers have not publicly audited the counting method in the sources used for this article.

For agency delivery, that distinction matters. You can still use ADN-style controls in a pilot. You should not put an unaudited metric into a client security slide as proven efficacy.

Infrastructure choices that change diligence work

Reporting says customers can run end to end on infrastructure owned by doxx.net. That list includes its own IP space and autonomous system number, its own DNS root, its own certificate authority, local AI models, and bare-metal servers.

Accounts do not require a username, password, or other personal information in the company’s description. Accounts are created through proof-of-work instead. a16z also notes users can start without providing an email address or phone number.

Barrett Lyon told SecurityWeek that many products sell privacy while routing traffic through rented infrastructure they do not own and cannot verify. He said doxx.net built a platform where privacy comes from architecture, so the company never can know what users search, where they go, or who they talk to.

Agency teams should hear both sides of that design:

Upside: Fewer rented hops and less PII at signup can shrink some privacy exposure.
Tradeoff: A parallel internet with its own TLDs, DNS root, and CA concentrates trust in one provider.
Client ask: Enterprise buyers will want independent verification before agents carry production credentials on that path.

Why India and USA agency teams should care now

Delivery teams in India and the USA already ship agent workflows for support, research, ops, and GTM. Those agents hit the open web. They also touch SaaS tools under client SSO. A malicious redirect or poisoned site is no longer only a browser hygiene issue. It is a delivery incident.

Agentic Defined Networking frames a control at the network layer. That sits beside prompt rules, tool allowlists, and human approval gates. It does not replace them.

Practical situations where ADN-style thinking helps:

Research agents that crawl vendor sites for RFPs
Coding agents that fetch packages or docs outside a mirror
Ops agents that open incident links from chat
Sales agents that visit lead websites at volume

If you build or integrate agents through agentic AI development, network boundaries belong in the same design review as model choice and tool scope.

ADN versus a VPN for agency delivery

Clients will ask if ADN is “just another VPN.” Public materials position it as broader than tunnel privacy:

Private naming and custom TLDs inside the parallel network
Own certificate authority for that namespace
Peer-to-peer messaging without a central communications server
Explicit agent APIs and agent-assisted network configuration
DNS blocking aimed at agent misadventure under user authority

a16z notes agents can help configure the network itself. That includes connecting devices, setting firewall rules, and opening private connections under user policies. For delivery teams, that is powerful and risky at once. An agent that can change firewall rules needs change control, logging, and rollback.

Delivery checklist for India and USA pilots

Use this checklist before you put client agents on a parallel network:

Map agent authority. List accounts, OAuth scopes, and secrets the agent can use.
Define allowed destinations. Prefer allowlists over open browsing for production tasks.
Separate pilot traffic. Keep ADN pilots off production customer data until review is done.
Test DNS blocks. Confirm known bad domains fail closed and false positives are logged.
Review the trust model. Document what doxx.net controls (IP, ASN, DNS root, CA) and what you still verify.
Check identity posture. Proof-of-work signup may help privacy. Confirm how your client’s SSO and audit needs map to it.
Add human gates. High-risk actions still need approval, even on a private path.
Write an exit plan. Know how to pull agents back to corporate network controls if the vendor path fails review.

Teams that need policy and oversight scaffolding can pair this technical pilot with responsible AI and governance practices so network controls stay tied to risk tiers.

What to tell clients without overselling

Keep client language factual:

doxx.net raised $38 million Series A led by Andreessen Horowitz.
ADN is in open beta as a private network for people and agents.
Threat protection is described at the DNS layer.
The “38 million threats blocked” figure is a company claim from closed beta.
Enterprise use should wait on independent verification and your own tests.

Avoid promising that ADN solves alignment, prompt injection, or tool misuse. Those are different layers. Network isolation helps when the failure mode is “the agent went to a bad place.” It helps less when the failure mode is “the agent did the wrong thing on an allowed site.”

How Debugged Pro teams would stage this

On a typical India–USA delivery split, we would stage ADN evaluation like any other security control:

Week 1: Read primary sources, map agent use cases, and choose one non-production workflow.
Week 2: Run browsing tasks with logging. Compare blocked events against your threat intel feed.
Week 3: Review identity, CA trust, and data residency questions with the client security owner.
Week 4: Decide go, no-go, or wait. Capture findings in the project risk register.

If you need help scoping that evaluation, start with a free consult through our software development consulting page. Bring your agent inventory and current network controls to that call.

Key Takeaways

Agentic Defined Networking from doxx.net is backed by a $38 million Series A led by Andreessen Horowitz.
ADN aims to give AI agents defined connectivity and DNS-level threat protection on a private parallel network.
Public sources describe own IP space, ASN, DNS root, certificate authority, and proof-of-work accounts.
The 38 million blocked threats figure is company-reported and should be labeled that way for clients.
India and USA agency teams should pilot with allowlists, logging, and an exit path before production credentials ride the mesh.

FAQ

What is Agentic Defined Networking?

Agentic Defined Networking is doxx.net’s term for a private networking model where people and AI agents get defined connectivity, peer-to-peer communication, and threat protection so agents are less likely to reach known malicious destinations while acting under user authority.

Who funded doxx.net’s Series A?

Andreessen Horowitz led the $38 million Series A. Animo Ventures and Focal.vc also participated, according to SecurityWeek’s October 3, 2026 coverage.

Does ADN replace application-level agent guardrails?

No. ADN targets network paths and destinations. You still need tool scopes, human approval for sensitive actions, evaluation, and incident response for mistakes on allowed destinations.

Is the 38 million threats number independently verified?

Not in the sources used here. doxx.net reports more than 38 million threats blocked since December 2025 in closed beta. Agencies should treat it as a vendor metric until audited.

Should India and USA delivery teams adopt ADN now?

Treat it as an open-beta evaluation. Run a limited pilot, verify blocks and false positives, and complete security diligence before agents carry production client credentials on the parallel network.

Author: Nitin Kumar is Lead Project Coordinator at Debugged Pro. He works with India and USA delivery teams on AI agent projects, security reviews, and practical rollout plans for client software.