Relevance AI's Invent 2.0 Builds Enterprise Agent Workforces Fast

Relevance AI today unveiled Invent 2.0, a major update to the collaborative agent the company pitches as the fastest route from a business idea to a working team of AI agents. The San Francisco company announced the launch on September 29, 2026, positioning the release as a gateway for enterprises moving past chatbots and copilots into genuinely autonomous agent workforces.

Invent works less like a dashboard and more like a solutions engineer. It interviews the user about a business process, studies their documentation, and figures out where value can be tracked. From there it builds an end-to-end agentic workflow: a coordinated team of specialist agents, wired together with the right tools, integrations, testing, and governance for enterprise use. According to the announcement, which was carried by PR Newswire via Morningstar, customers report going from concept to production workflow with full evaluation and monitoring in hours instead of weeks.

From a conversation to a coordinated agent team

The core workflow breaks into four capabilities, each designed to replace labor that human teams currently do by hand. First comes process mapping: Invent holds a collaborative dialogue to understand a business process end to end, flagging which decision points need human judgment and which parts of the workflow should become tools, agents, or full workforces.

Next is agent construction and integration. The system automatically builds multi-agent architectures, wiring specialist agents together with the tools they need, connecting them to the external systems where the real work happens, and orchestrating the triggers that keep everything running. The promise is that no one has to hand-build and maintain every agent, tool, and workflow themselves anymore.

Agents that monitor their own quality

The two capabilities that set Invent 2.0 apart address the problem that keeps enterprise AI teams up at night: agents that work on day one and degrade on day thirty. Invent sets performance benchmarks before an agent goes live, then samples live runs against those standards on an ongoing basis. When quality drifts, it diagnoses the root cause and proposes fixes to prompts, tools, examples, models, or orchestration logic, with built-in testing before anything ships.

On top of that sits a continuous improvement loop. Invent reviews live agent activity, error patterns, and evaluation results to surface new automation opportunities and recommend refinements over time. Reporting by AIThority notes the company describes this as the gateway to agentic workflows that can evaluate, monitor, and improve themselves without a human team doing the tuning.

Real deployments, not just demos

Invent 2.0 was refined over a multi-month enterprise design partner program, where business builders and operations teams put it to work on real, high-stakes workflows. One partner, Secure Code Warrior, has fully switched its internal automation over to Invent. Matt Wilson, the company's VP of FP&A and Revenue Operations, says the deal desk function now runs entirely on a small number of Invent prompts, improving speed and quality at a fraction of the cost of the previous manual process.

Wilson's verdict, as quoted in the announcement: Invent makes building and managing powerful agent workflows straightforward, reducing what used to be weeks of building time alone to less than a day. His teams, he says, cannot imagine building and managing agent workflows without it.

The push toward Level 3 and Level 4 agents

Relevance AI frames Invent 2.0 as the pathway from today's common deployments to something more ambitious. Most organizations sit at what the company calls Level 1 and Level 2 agentic deployments: chatbots and copilots that improve individual productivity. Invent is designed to carry them to Level 3, Autopilot Agents, and Level 4, Self-Improving Agents, which elevate the productivity of an entire department or company.

Co-CEO Daniel Vassilev argues that making that leap requires enterprise-class governance, testing, monitoring, and self-improving loops, and that Invent is the package that delivers them. Whether or not every enterprise is ready for self-improving workflows, the direction of travel is clear: the industry is racing to give agents the evaluation and oversight scaffolding that production software has had for decades. You can read more about how enterprises are getting a handle on sprawling agent deployments in our earlier coverage of Dataiku's agent management tooling, and follow ongoing developments on the AI News topic page.

The broader context matters too. Agent workforces are becoming a crowded category, with every major platform vendor racing to own the orchestration layer. Relevance AI's bet is that the differentiator will not be the models or the agents themselves, but the discipline around them: process mapping up front, benchmarks before launch, and continuous evaluation after. If Invent 2.0 delivers on that promise at the speed its design partners report, the bottleneck for enterprise AI may shift from building agents to deciding which processes deserve them first.