AI agents stop being experiments and start being staff
AI agents in enterprise automation are specialised software workers that own well-defined tasks end to end—such as compliance checks, credit memo drafting, or analytics workflows—under human supervision, with the aim of shrinking routine workload and speeding up decisions while keeping people responsible for judgement and final outcomes. On paper this sounds like yet another hype cycle, but the reality is more pointed: these agents are cutting task time by half or more in production, not in demos. The pattern across banking, ride-share, and streaming is the same. Humans keep control of goals and risk; agents grind through the mechanical work at scale. The question for enterprises is no longer "if" they should adopt agentic AI, but whether they can afford not to as competitors turn pilots into live workflows.
Compliance: from eight-day onboarding to minute-scale decisions
In bank compliance, AI agents are quietly becoming a parallel workforce. McKinsey estimates banks could unlock productivity gains of up to 20 times by having one compliance professional supervise 15–20 specialised agents rather than doing every task themselves. These are narrow workers: onboarding agents read KYC documents and run sanctions screening, due diligence agents map beneficial owners and draft risk narratives, monitoring agents rank alerts and draft suspicious activity reports, while perpetual KYC agents track registry changes over time. A worked example shows the payoff: one business onboarding flow went from eight days of manual back-and-forth to under a minute once agents handled registry pulls, structure mapping, and clearing a false-positive sanctions match before an analyst even opened the case. That is not a marginal improvement; it is a redesign of compliance work, turning analysts into supervisors of AI compliance automation rather than form-fillers.
Analytics: Grab and Netflix prove agents can think with data
Analytics teams used to spend most of their time on mechanical chores. At Grab, AI agents now automate large parts of these workflows, cutting the share of mechanical tickets handled by analysts from 44% in February to 30% in June for tasks like data preparation, alerting, and reporting. The company’s autonomy model lets agents discover data, write and execute queries, validate results, and draft analysis while humans frame the question and review gates, so analysts shift toward self-service workflows and deeper interpretation instead of pipeline babysitting. Netflix goes after a tougher problem: causal inference. It has open-sourced an agentic workflow for Observational Causal Inference that uses an actor-critic loop to estimate causality, run sensitivity analyses, and write reports, deliberately reducing toil in repetitive causal analysis while keeping humans in charge of question framing and result evaluation. This is workflow automation productivity gains in the heart of knowledge work, not only in back-office reporting.
Lending: DBS turns credit analysis into an agentic assembly line
Credit assessment is classic white-collar grind: gathering documents, extracting figures, reading industry reports, and writing long memos. DBS has deployed an agentic AI tool that helps about 1,500 employees worldwide handle corporate clients seeking financing, with specialised agents taking on more than 70 separate credit tasks. The system pulls information from annual reports, industry research, and internal records to produce a first draft of a credit memo, then responds to follow-up research requests and revisions from relationship managers and credit risk managers. The bank aims to cut time spent on credit memos and related work by at least 30%, work that previously consumed up to 40% of a relationship manager’s time. According to Han Kwee Juan, Group Head of Institutional Banking, the goal is to "level up the quality of our credit analysis at scale" while giving managers more time for strategic client discussions and portfolio-level risk thinking.
From pilots to production: what enterprises should copy next
These examples share a clear signal: enterprise AI deployment has crossed from pilot novelty into production reality. Research shows only about 10% of financial institutions have deployed AI agents at scale so far, with fraud detection and customer onboarding ranked as leading use cases, but the ones that have moved first are already reporting measurable gains. Grab’s self-service analytics answered without human involvement climbed to as high as 90% of data pulls and 81% of SQL requests in only a few months, supported by dozens of production deployments and features added to its analytics portals. DBS is extending agentic AI beyond lending into customer service bots used by tens of millions of customers. Netflix is making its causal inference agents available to others, inviting scrutiny and reuse. Enterprises that still treat AI agents as sandbox toys are ceding workflow advantages to rivals who are turning them into real staff. The next competitive frontier is not building an agent; it is redesigning work so agents and humans share the load by default.
