AI agents now handle finance, supply chains - even chat with customers.
Breakthrough models supercharge AI agents.
The race to build more capable AI agents accelerated with new releases that push the boundaries of scale and efficiency. On April 29, Anthropic unveiled a 10-trillion-parameter model designed for high-stakes reasoning in areas like cybersecurity and advanced coding ([1]). At the same time, OpenAI disclosed that its updated GPT-5.4 “Thinking” model scored at human-expert levels on a complex business benchmark – highlighting a shift from simple Q&A towards AI that can formulate goals and carry out multi-step tasks autonomously ([2]).
Another milestone this week came from NVIDIA, which launched a 30B-parameter multimodal AI called "Nemotron 3" that processes text, images, audio and video within one unified system ([3]). By eliminating the need for separate specialized models, it achieves up to 9x higher throughput than previous open multimodal models at similar accuracy ([4]). This innovation promises significantly faster, more cost-efficient agent performance on tasks like document analysis or call center automation.
Finally, researchers are tackling AI’s efficiency from a different angle: extreme compression. New "1-bit" large language models – which use only a single bit to represent each parameter – have demonstrated up to 100x reductions in energy use while preserving advanced reasoning capabilities . Such progress hints at a future where powerful autonomous agents can run even on everyday devices and IoT sensors, making AI-driven workflows ubiquitous.
AI agents take on finance & HR workflows.
Enterprise software is rapidly integrating AI agents to perform tasks once reserved for human teams. At its April 28 "Sage Future" event, business software firm Sage announced new built-in AI 'Intelligence' agents across its finance, HR and operations products ([1]). These agents can execute routine processes – from invoice entry and payroll runs to flagging anomalies in accounting data – directly within the company’s core systems ([2]).
This represents a step change from passive analytics to active automation. For example, Sage’s AI-driven Finance Intelligence Agent can prepare payment reminders and purchase approvals inside the normal workflow, providing an explanation for each recommendation and requiring a human manager’s sign-off before finalizing actions ([3]) ([4]). By ensuring transparency, control and an audit trail, the system lets CFOs and controllers trust the AI while it accelerates processes.
Early enterprise adopters are already seeing significant efficiency gains. Global audit firm EY’s internal AI platform now processes 1.4 trillion lines of audit data across 160,000 client engagements per year ([5]). Meanwhile, JPMorgan’s deployment of AI agents for research and document drafting has cut analysts’ research times by 83% and automated over 360,000 hours of work annually ([6]). These examples suggest that as AI agents prove their reliability in back-office roles, organizations can redeploy human experts to higher-value tasks such as analysis and strategy.
Agents tackle supply chain & industry challenges.
AI’s growing autonomy is also solving problems in industry-specific operations. At its SAS Innovate conference this week, analytics leader SAS announced a $1 billion investment in "industry accelerators" – packaged AI agent solutions targeting complex domain challenges ([1]). The centerpiece is a new Supply Chain Agent that will automate sales and operations planning (S&OP) for manufacturers and retailers – an intensive process that many companies have only been able to execute on a monthly cycle until now ([2]).
The Supply Chain Agent runs continuously, adjusting production and inventory plans in real time as conditions change. A manager can ask the AI to model a hypothetical 15% drop in demand and immediately see how it would rebalance production, with the agent explaining its recommendations for transparency and trust ([3]). By moving from static, spreadsheet-based planning to an always-on, AI-driven process, businesses can respond faster to shifts in demand and supply, potentially reducing stockouts or waste.
Industry experts note the significance of this shift. As one IDC analyst observed, compressing such a complex process into an autonomous agent "could deliver significant value" ([4]), bringing decades of supply chain expertise into an AI that never takes a day off. The broader trend is that major software vendors (from ERP providers to cloud platforms) are rolling out similar domain-specific agents – in finance, healthcare, customer service and beyond – to help organizations tackle their most stubborn operational challenges with AI.
Securing & governing autonomous agents.
Even as capabilities grow, companies are learning that successful AI agents require strong oversight. Fewer than 15% of enterprise AI agent projects have fully scaled out of pilot stage, with over 85% failing to achieve lasting impact . Analyses indicate the main obstacles are not model accuracy, but integration hurdles, unclear ROI, and a lack of robust governance and training processes .
Recognizing these risks, industry groups are stepping in. On April 28, the FIDO Alliance – a consortium of tech and financial firms – announced new standards to ensure agents perform only authorized actions online ([1]). The effort will define secure methods for AI agents to log in to services and initiate payments on behalf of users, using verifiable user instructions and distinct agent authentication protocols ([2]). The goal is to prevent autonomous systems from “going rogue” with sensitive data or transactions and to reassure organizations that agent-driven workflows can be trusted.
Experts also emphasize that enterprises must rethink how they design AI systems for reliability. Rather than just building bigger models or writing better prompts, leading AI teams are now adopting "harness engineering" – constructing a structured environment around the AI that provides memory, tool access, constraints, and human feedback loops ([3]). Without such scaffolding in place, even advanced AIs are prone to stray off course during complex, multi-step tasks – forgetting context, violating policies, or compounding errors in ways that humans must later unwind ([4]). As AI agents become more powerful co-workers, C-suite leaders will need to invest as much in governance, auditing and training frameworks as in the technology itself.
Conversational agents reimagine customer engagement.
A final frontier for agent adoption is customer-facing roles in marketing and service. This week, Snap Inc. began testing “AI Sponsored Snaps,” a new conversational ad format that places branded AI chatbots into Snapchat’s direct-messaging channels ([1]). These AI agents can engage nearly a billion monthly users in natural dialogue – answering questions, offering recommendations, and even facilitating purchases – all without the user leaving the chat.
Initial results are promising. In a pilot campaign with credit bureau Experian, Snapchat’s chatbots drove 22% higher conversion rates and reduced cost-per-conversion by about 20% compared to standard digital ads ([2]). Snap’s own data shows that its community is open to AI interactions: over 500 million users have tried the app’s built-in "My AI" assistant, and users sent 950 billion messages in Q1 2026 alone ([3]). This consumer comfort with AI-driven conversations suggests new opportunities for retailers, banks and healthcare providers to offer personalized, automated customer service and marketing. However, executives should approach these front-line agents with the same care as internal ones – ensuring they stay on-message, comply with regulations, and escalate to humans when necessary.