Forward (formerly Forward Networks) has been building out technology for a decade to help enterprises better understand their network traffic. In January, the company launched an agentic AI system built on its network digital twin, giving operators a mathematically grounded way to query, verify, and automate workflows across multi-vendor environments. The company is now extending that foundation with Forward Predict, a capability that models future network state after a proposed change, before that change is executed.
Forward Predict runs proposed changes against the company’s network digital twin, modeling the impact across routing protocols, ACLs, firewalls and NAT configurations at production scale before anything touches the live network.
“The key here is being able to understand the impact of a change you are about to make on an enterprise-scale, multi-vendor network [before] actually executing those changes on the production network,” Nikhil Handigol, chief AI officer and co-founder of Forward, told Network World. “You can have a lab, and you can make some changes, test it out, but those labs are never equivalent to the actual production network, which can be tens of thousands of devices.”
How Forward Predict works
Forward’s network digital twin has historically operated on the data plane, maintaining a complete model of forwarding state across every device in the network. Forward Predict extends that model to cover the control plane. The control plane model is verified against combinations of OS versions, configurations and multi-vendor device interactions.
“We have literally traced where every packet could ever go, proactively, to analyze it to the highest degree that is actually possible,” David Erickson, CEO and co-founder of Forward, told Network World.
A proposed change is fed into the platform, which traces how it alters control plane protocol behavior and models how that state propagates into the data plane. The platform then runs its full packet-path analysis against the predicted snapshot before any change touches the live environment.
The scope extends beyond routing to cover firewalls, ACLs and NAT configurations. “Predict basically analyzes and exposes the end-to-end impact of that firewall or ACL change that you make on those firewall devices,” Handigol said. “It’s not just a local impact that matters, but it’s the end-to-end impact in terms of who can talk to whom.”.
Autonomous networking is the path forward
Forward positions Forward Predict as a prerequisite for agentic AI in network operations. Erickson compared the capability to the software development toolchain, specifically the compiler and unit test framework that allow developers to validate code before it ships. Network change management has until now lacked the equivalent.
“It’s powerful for humans, it’s a superpower for AI agents, because it gives them the ability to vet out changes, understand the impact, take feedback, and keep iterating until they get the change right,” Handigol said. “That ability to iterate at machine speed like that is the key unlock for agents.”
The autonomous loop runs in three steps, predicting the change, executing it, then verifying against the live network that the intended outcome was achieved. Forward Predict capabilities are exposed to AI agents as callable tools via REST API or MCP server.
For organizations with hundreds to thousands of changes queued in IT service management platforms like ServiceNow, Forward’s argument is that running the entire backlog through Predict in parallel shifts the bottleneck from design and approval to deployment.
What’s next
Erickson said the company’s focus beyond the launch is expanding what the platform can drive autonomously. That includes covering both reactive responses to network events and proactive improvements operators have not yet had time to address.
“We are laser-focused on helping get folks to an autonomous networking future,” Erickson said. “There are tons and tons of opportunities for security improvements, for cleanups of complexity that don’t need to be there, for maintaining compliance. The list is almost endless in these incredibly complex networks today.”