What happened
Cisco says MyAgent is now rolling out to all 90,000 employees, moving beyond chat into supervised autonomous workflows across tools including Outlook, Webex, Jira, and SharePoint. The system can take an employee’s objective, determine the required sequence of steps, work across applications, and retain persistent user context.

Why it matters
Enterprise agents just crossed an important line from departmental experiment to workforce infrastructure at one of the world’s largest technology companies. Cisco also says agentic interactions on the underlying Circuit platform grew nearly 350% quarter over quarter, suggesting usage is scaling alongside access.

What’s next
Cisco explicitly wants its internal deployment to become a blueprint for customers and partners, making MyAgent a live test of whether supervised autonomy can scale without creating an equally large governance and security burden.

What happened
Visa expanded its open-source Vulnerability Agentic Harness, or VVAH, from vulnerability discovery into remediation and validation. The updated system adds closed-loop remediation, broader model choice, and visibility into long-running workflows; Visa says some resolution cycles can shrink from weeks to hours.

Why it matters
Security agents are moving from producing reports to participating in the actual repair loop. That can compress response times dramatically, but it also makes permission boundaries, validation, auditability, and human control part of the security product itself rather than an afterthought.

What’s next
VVAH is being contributed to Nvidia’s Open Secure AI Alliance while Visa expands related enterprise advisory services, so the next signal is whether autonomous remediation becomes a standard layer in corporate application-security pipelines rather than a specialist experiment.

What happened
Google upgraded AI Mode in Search with flight-price tracking, rewards-rate search, and hotel booking inside the conversation flow. Users can ask it to monitor fares across data from 300+ airline and travel partners, while U.S. users are beginning to get hotel checkout through partners including Booking.com, Expedia, Hilton, Marriott, and Priceline. 

Why it matters
Google is pushing Search from answering questions toward executing multi-step consumer workflows finding options, monitoring prices, narrowing choices, and moving users through a transaction. That turns one of the internet’s biggest discovery surfaces into an agentic action layer, potentially reshaping how travel platforms compete for both traffic and transactions. 

What’s next
Hotel booking is rolling out in English in the U.S. over the coming weeks, while flight-price tracking is already available across more than 180 countries and territories; Google also plans to expand the loyalty programs supported for points-and-miles searches. Watch whether this transaction model spreads deeper across Search, because the bigger shift is AI moving from recommending the next step to helping complete it. 

What happened
Salesforce and Anthropic announced Claudeforce, a two-way integration that puts Salesforce data and governed actions inside Claude while embedding Claude across Salesforce products. The first Salesforce-in-Claude plugin has 37 prebuilt sales skills spanning meeting preparation, deal health, pipeline reviews, and CRM updates.

Why it matters
The enterprise AI battle is moving closer to systems of record. Instead of forcing employees to copy data into a chatbot, Claudeforce lets a frontier model reason over live business context and execute actions through Salesforce’s permissions and workflow rules, a model for how AI interfaces could begin replacing portions of traditional software navigation.

What’s next
Salesforce in Claude is already with select pilot customers and is expected to enter open beta in September, with more prebuilt skills beginning in Q3 and deeper integrations planned across Claude, Salesforce, and Slack.

What happened
Workday reported that AI generated more than 25% of its new annual contract value in the quarter and that more than 5,500 customers now use at least one of its organic agents, up more than 35% from the previous quarter. Total quarterly revenue reached $2.649 billion, up 12.8% year over year.

Why it matters
Enterprise AI is usually measured in pilots, seats, or vague productivity claims; Workday is tying agents directly to new contract value and customer adoption. That makes this one of the cleaner indicators that agentic products are becoming a meaningful commercial driver inside established enterprise software.

What’s next
Workday raised its full-year subscription-revenue outlook to $9.94 billion–$9.95 billion while saying it will continue prioritizing its agentic roadmap. The metric to watch now is whether AI can continue contributing more than a quarter of new contract value as the customer base expands.

What happened
Anthropic opened a research preview of the Model Hardware Standard, or MHS, a model-agnostic specification that lets AI agents discover, communicate with, and operate programmable physical equipment including microscopes, liquid handlers, robotic arms, and manufacturing machinery. Anthropic says integrations that can normally take weeks or months can be reduced to hours or minutes.

Why it matters
Physical AI has a plumbing problem: every robot, laboratory instrument, and industrial machine tends to have its own interface. MHS attempts to create a common agent-to-hardware layer analogous to what Model Context Protocol did for software tools, potentially making heterogeneous machines far easier for autonomous agents to coordinate.

What’s next
Anthropic is testing MHS with scientific labs and manufacturers before making it open source. Early Genentech work showed Claude coordinating a liquid handler, robotic arm, and plate reader and recovering from some hardware errors but also exposed weaknesses in real-world physical reasoning that still required expert intervention.

Agents are moving deeper into the systems people already use. Cisco is scaling personal agents across the workforce, Visa is letting security agents close the repair loop, Google Search is edging into transactions, Claude is moving into the CRM, and Anthropic is extending agent interfaces into physical equipment. The next phase of AI is less about answering better questions and more about becoming the action layer across software, commerce, and the physical world.

See how far an AI can get without you clicking around.

Pick one real workflow you’d normally complete across several apps like trip planning, CRM prep, inbox cleanup, account research, or meeting preparation and try to complete it through an agentic interface instead of navigating each tool manually.

A few options from today’s stories:

As it works, notice where it still asks you to step in:

💡 Insight
The interesting shift is from asking AI to help with individual steps to asking it to own the outcome across the workflow.