Genpact has introduced Onyx, a platform designed to help large companies build and operate artificial intelligence agents within real business processes. Announced on 16 November, the product targets work such as accounts payable, procurement and other enterprise operations in which an AI system must follow rules, work with messy records and know when to involve a human.
The announcement matters because the debate around agentic AI is moving beyond what an assistant can say to what it can reliably do. An agent is software that can take a series of actions to accomplish a goal, rather than respond to a single prompt. In corporate environments, those actions may involve invoices, financial records, supplier data or decisions that create legal and operational consequences.
Why knowledge of the process matters
A company can instruct a model to process an invoice, but exceptions quickly appear. A supplier may have changed its bank details, an order may be missing approval, or the amount billed may not match the delivery. Handling such cases requires an understanding of company policy and how experienced staff resolve disputes.
Genpact says Onyx uses a Process Intelligence Core built from operating knowledge collected across finance, risk, banking, supply chain and other functions. The platform is intended to connect that knowledge with enterprise data and a selection of AI models. Its first proprietary solution is an Accounts Payable Suite, while clients can also commission custom agent workflows.
The company describes an approach called “earned autonomy”. Agents may initially propose actions for a specialist to approve, and take on additional responsibility as their reliability is demonstrated. In theory, this can avoid handing high-stakes authority to a model before its performance has been measured.
The difference between automation and accountability
Traditional automation often follows fixed rules. Agentic systems may interpret information, choose tools and decide the next step, which can make them more flexible but also harder to supervise. Genpact says its platform includes logs of agent decisions, controls around what an agent may do and processes for human review.
For customers, the critical question is whether these safeguards work across routine transactions and unusual cases. An invoice approved incorrectly can cause a financial loss; an account change accepted without proper verification can create fraud exposure. A polished demonstration is not evidence of a dependable production process.
Companies evaluating this technology should ask how the platform measures error rates, what happens when an action fails and how responsibility is allocated between software and employees. The availability of human approval also matters: a requirement to review every minor decision could erase much of the expected efficiency gain.
Business significance beyond the product launch
Genpact has historically provided outsourced business operations and related services. Building an AI platform around that experience suggests a possible shift in how such services are delivered. A client may expect higher transaction volumes and faster processing, while the service provider must demonstrate accuracy and accountability rather than simply reduce staffing effort.
In Singapore and across ASEAN, companies implementing agentic workflows will also need to consider data protection, audit expectations and the integration of older systems. Financial services, multinational shared-service centres and manufacturers can have different requirements even for apparently similar back-office tasks.
Genpact has not provided independently verified evidence that Onyx produces a specified return on investment for every customer. That outcome will depend on workflow selection, data quality, governance costs and how well staff handle exceptions.
The direction is nevertheless significant. Agentic AI is increasingly being sold as an operating model, not just a productivity feature. The companies that gain most may be those that redesign processes carefully, keep humans responsible for consequential decisions and measure results against clear business objectives.
Sources and further reading
Sources: Genpact official announcement, 16 November 2026; Genpact newsroom and product background.