Autonomous Agents Advance: New Tools, Real ROI, Rising Risks

OpenAI’s July 6 update signaled a strategic shift: instead of showcasing a more powerful model, the company focused on turning its technology into something immediately useful on the job. ChatGPT’s new "workspace agents" can now be shared across teams and entrusted with complex, multi-step tasks, all while operating within an organization’s approved permissions and policies. These agents run in the cloud and don’t need a user babysitting them; they can carry on working even after a human logs off. For instance, OpenAI’s own sales team uses an AI agent to sift through call notes and client data, qualify leads, and draft follow-up emails, allowing employees to spend more time with customers instead of paperwork.

OpenAI is not alone. Just days earlier, rival Anthropic unveiled Claude "Sonnet 5," a new model boasting advanced agent-like capabilities and significant leaps in multi-step reasoning and coding performance. Google, for its part, has retooled its upcoming Gemini platform around agents, including visual workflow designers and persistent memory for long-running agent tasks. In short, the race among AI leaders has moved beyond raw model size and into enabling autonomous action. The major AI players are increasingly positioning these agents not as chatbots that merely answer questions, but as digital workers that can carry out instructions and drive business processes.

New solutions are also emerging to let companies deploy AI agents without deep technical skills. At its Inspire 2026 conference, analytics firm Alteryx introduced “Agent Studio,” a no-code platform that lets business analysts turn existing data workflows and business rules into self-running AI agents. An analyst can, for example, take a pre-defined rule (say, a low-inventory alert or a customer onboarding process) and with a few clicks wrap it into an AI agent that automatically monitors conditions and takes action, without writing any new code. This approach essentially “harnesses” the organization’s current business logic and delegates it to an always-on digital assistant.

Even outside traditional enterprises, the push for easy AI automation is accelerating. Cryptocurrency platform BNB Chain this week launched its own Agent Studio, which it claims can deploy an autonomous AI agent with nothing more than a single text prompt. And major software vendors are re-architecting their platforms to be operated by AI directly. For example, Salesforce’s new "Headless 360" mode exposes its entire CRM system via APIs, enabling AI coding agents (such as Anthropic’s Claude or open-source tools like Cursor) to manage customer records and build applications without a human user interface. In fact, Salesforce says it decided two years ago to rebuild its core products for an "agent-first" era, exposing all functions so that AI systems can tap them programmatically rather than through a screen. The common theme: creating an ecosystem where AI agents can be deployed in minutes and woven into everyday tools, dramatically lowering the barrier to autonomous workflows.

For many organizations, autonomous AI isn’t just a futuristic idea—it’s already boosting productivity and operating metrics. In a recent survey of over 500 tech leaders, 80% reported that their AI agent investments are yielding measurable returns for the business. The companies seeing the biggest gains treat these bots not as R&D experiments but as core parts of their operations, integrating them into key workflows rather than leaving them on the sidelines.

Financial services firms have been among the early movers. In a notable first for Wall Street, Morgan Stanley plans to let thousands of its corporate clients plug their own AI agents directly into the bank’s stock-plan administration platforms. This will allow automated management of complex equity transactions without human intermediaries—a bold step that other banks have yet to take. Meanwhile, established banks are already seeing significant value from internal AI deployments. Lloyds Banking Group reported about £50 million in productivity gains from AI last year and expects over £100 million in 2026 by scaling up both generative and "agentic" AI projects. Specific use-cases are delivering real impact: the bank’s coding assistant sped up legacy code conversion by 50%, and an AI-driven HR virtual assistant now resolves 90% of employee queries correctly on the first contact, relieving staff of routine queries.

Retailers are likewise finding that AI agents can directly drive revenue. This week, Salesforce announced the general availability of its Agentforce Commerce suite, introducing AI-powered "Shopper," "Buyer," and "Merchant" agents to automate online customer journeys and back-end sales operations. The payoff has already been evident: during the 2025 holiday season, AI influenced roughly 20% of global online sales—about US$262 billion in value—and retailers that deployed their own shopper agents saw e-commerce sales grow 59% faster compared to those that did not. These AI-driven interactions also convert customers at a rate many times higher than traditional web or social channels. The message is clear: companies that quickly embrace agent-enabled workflows are capturing a tangible lead in market performance.

Rapid adoption of agentic AI brings not just rewards but also new vulnerabilities. This reality was starkly illustrated by the revelation of the first known case of fully AI-executed ransomware. Dubbed “JadePuffer,” the malicious agent gained access to an enterprise network and proceeded to carry out an entire cyberattack autonomously. After a human hacker picked the target and provided initial instructions, the AI agent exploited a known software flaw, stole credentials, moved laterally across servers, and encrypted 1,342 database records without any further human guidance. It even generated its own ransom demand note. Security experts warn this development lowers the skill and cost barrier for sophisticated attacks—potentially making it “as easy as an API call” for bad actors to deploy intelligent malware.

The incident is a warning sign for business leaders that uncontrolled or malicious AI isn’t a hypothetical risk, but an emerging threat. It also highlights why robust AI governance and security measures are becoming non-negotiable. Companies are responding by extending familiar “zero trust” principles to AI. Cybersecurity firm CrowdStrike, for example, just announced a "Continuous Identity" system to manage machine credentials and privileges for AI agents alongside human users. This ensures that autonomous bots only have access to what they need and can be instantly locked down if they behave unexpectedly. In a similar vein, ServiceNow has partnered with Microsoft to integrate its AI Control Tower into the Microsoft 365 platform. This move will let organizations carefully vet and monitor AI "co-workers" operating across email, documents, and other everyday apps, providing a centralized way to track what agents are doing and impose policies or kill-switches as needed.

Beyond internal governance, external regulators are increasingly shaping the pace of AI deployment. Just last week, U.S. export controls forced Anthropic to temporarily halt access to its most advanced model, known as Fable 5, over national security concerns. The company’s rapid compliance measures allowed it to redeploy the model globally on July 1, but the episode illustrates how geopolitical factors can abruptly constrain cutting-edge AI capabilities. Leaders must now factor regulatory risk into any AI strategy—especially when relying on third-party AI providers.

Regulated industries like healthcare are also moving cautiously. In late June, a startup called UpDoc publicly debuted what it calls the first FDA-cleared clinical AI agent designed for direct patient use. Cleared as a narrow diabetes management tool, this system provides insulin dosing guidance to patients based on parameters set by their doctor. Its approval (via the FDA’s rigorous 510(k) process) shows that even “agent” AIs can gain regulatory acceptance if they operate within strict safety boundaries and human oversight. Healthcare consortia have followed up by publishing detailed AI risk management guidelines, and major hospital systems are investing in AI oversight platforms to monitor algorithms across clinical workflows. The big picture: whether it’s compliance for cutting-edge models or patient safety for medical AI, effective governance will determine which organizations can safely harness autonomous systems at scale.