Relevance AI launches Invent 2.0 to build and manage AI agent workforces

Relevance AI said Tuesday it launched Invent 2.0, the latest version of its collaborative agent for designing, building, and managing AI agent workforces that automate business processes end to end.

Most agent tools still leave an operations team to wire prompts and integrations by hand. Invent 2.0 is Relevance AI’s pitch that a collaborative builder can interview the business, stand up a team of specialist agents with tests and monitoring, and keep improving that work after it goes live.

On Tuesday, 29 September 2026, Relevance AI announced Invent 2.0 from San Francisco. Relevance AI calls it the latest version of its collaborative agent for designing, building, and managing AI agent workforces. An agent workforce, in that sentence, is a team of specialist agents that automate a business process from start to finish. The PR Newswire page stamps the release Sep 29, 2026, 09:05 ET, which is 9:05 a.m. Eastern. Those lines are on the release.

What Invent does, as the release states it. Invent works like a solutions engineer, the person a software company sends to learn how a customer’s work actually runs. It interviews users, analyzes documentation, maps the process, and tracks value. It then creates an end-to-end agentic business process. Agentic means the software works toward an outcome, not only a chat reply. That process is a coordinated team of specialist agents, with the tools, integrations, testing, and governance the release says a company needs. An integration is a connection to another system. Governance is the set of rules for who may act, and the checks before a change goes live. The release also says Invent replaces the work of hand-building and maintaining every agent, tool, and workflow. Those lines are Relevance AI’s.

Four capabilities the release names. Process mapping: Invent talks through a business process from start to finish, and marks which decisions need a person and which parts should become tools, agents, or workforces. Agent construction and integration: the system builds a multi-agent setup on its own, wires agents to the right tools, connects them to the outside systems where the work happens, and sets the triggers that keep the work running. A trigger is the event that starts the next step. Quality assurance and performance monitoring: Invent sets performance benchmarks before an agent goes live, then samples live runs against those standards. A benchmark is the bar the agent has to clear. When quality slips, Invent names the cause and proposes fixes to prompts, tools, examples, models, or the way the agents are coordinated, and it tests those fixes before anything ships. A prompt is the instruction given to the agent. Continuous improvement: Invent reviews live activity, error patterns, and evaluation results, then points out more work that could be automated and recommends refinements, instead of leaving that review to a human team. Those lines are Relevance AI’s.

How long the company says the build takes. The release’s summary says customers report agent workforces going from an idea to a production workflow, with evaluation, monitoring, and optimization, in hours instead of weeks. Production is the live system a company actually runs. A caption on the same page says business experts can create in hours what previously took engineers weeks. The same summary says Invent is already helping companies raise process efficiency and effectiveness, and drive revenue. Those lines are Relevance AI’s account of what customers report. The release does not print a revenue dollar for that claim, and it does not name the customers behind the hours-instead-of-weeks line beyond the Secure Code Warrior quotation below.

Where the product was tried. Invent 2.0 was refined over a multi-month enterprise design partner program. Relevance AI says business builders and operations teams put Invent to work on real, high-stakes workflows. Those lines are Relevance AI’s. The release does not name every partner, and it does not print a count of them.

Secure Code Warrior is the customer the release names. The company, which the release calls an AI software governance and secure coding training company, has fully switched to building and running its internal automation on Invent. Matt Wilson, vice president of financial planning and analysis and revenue operations, said the deal desk now runs entirely on a small number of Invent prompts. A deal desk is the team that checks a sale before it closes. He said that change improved speed and quality at a fraction of the cost of the earlier manual process. Quoted: “Invent makes building and managing powerful agent workflows straightforward and easy.” He said he was impressed by the experience, and that Invent cut what used to be weeks of building time alone to less than a day. He said his teams cannot imagine building and managing agent workflows without Invent. Those sentences are his, in the release. The release does not print a dollar figure for that cost.

Daniel Vassilev, co-chief executive of Relevance AI, said the launch makes it straightforward for an organization to move past agents that help one person, such as copilots, to autonomous workflows that raise the output of a whole department or company. A copilot, in that sentence, is an assistant a person still steers. The release describes Level 1 and Level 2 as chatbots and copilots, Level 3 as Autopilot Agents, and Level 4 as Self Improving Agents. Vassilev said the leap from one person’s productivity to Level 3 and Level 4 productivity needs governance, testing, monitoring, and self-improving loops built for a company. He said that is what makes Invent powerful, because it can deliver efficiency and effectiveness on a new level. Those sentences are his, in the release.

Who Relevance AI says already uses the platform. It helps organizations build and manage AI workforces, teams of specialist agents that automate business processes end to end. It says the platform is used by companies including Canva, Confluent, Autodesk, and AVEVA, across sales, customer success, marketing, support, operations, human resources, and finance. Those names are on the release. The release does not say each of those companies is running Invent 2.0. It does not print a customer count, a price, or a revenue figure.

Further reading the release names. It points readers to a post by Paulwyn Devasundaram, head of engineering, titled “Launching Invent: the fastest way to AI agents that actually do the work,” and to a whitepaper, “The Levels of AI Autonomy,” on how agents move from Level 1 to Level 4. Those titles are on the release.

In plain terms, Relevance AI said on Tuesday that Invent 2.0 interviews a team, maps a process, stands up specialist agents with tools and tests, then watches live runs and proposes fixes. Secure Code Warrior’s Matt Wilson said deal-desk build time fell from weeks to less than a day. The release names Canva, Confluent, Autodesk, and AVEVA as companies that use the platform. It does not print a price.

The picture is Relevance AI’s Invent launch graphic. White type on a dark field reads “Launching Invent,” with the line “The fastest way to AI agents that actually do the work.” It is the announcement image for this launch.