Enterprise Software Watch · October 9, 2026
Five announcements in four days point at the same contest: not who has the smartest assistant, but who controls the identity, permissions, context, and audit trail when an AI agent actually does the work.
Core Question
As enterprise software shifts from helping employees complete tasks to letting agents execute workflows across business systems, who controls that execution layer, and can vendors turn it into measurable value?
Last week's coverage looked at the control points in the abstract. This week, five vendors with very different starting positions each made a move on them. One caution applies to everything below: most of this is announcement, not adoption. The table separates what is live from what is preview or planned, and the analysis flags where customer evidence is still missing.
Status Snapshot
| Vendor | Announced | Status |
|---|---|---|
| Google Gemini work agent | Oct 8 | Private preview, enterprise customers only |
| Cisco Webex agentic updates | Oct 7 | Staggered: Claude Managed Agents "soon"; some features from November; Webex Calling personal agents Q1 2027 |
| Apollo GraphOS Agent Services | Oct 7 | Public preview per Apollo's announcement |
| SAP Joule Work and Autonomous Suite | Oct 6 | Rolling out through October; new supply-chain agents planned for general availability in Q4 |
| Salesforce Trusted Enterprise AI Harness | Sept 10 | Component products live; unified experience and AI Control Plane roll out from early fiscal 2028; no pricing |
1. Google's Gemini Work Agent Raises the Stakes
At its Gemini at Work event on Thursday, Google announced a "universal" Gemini agent delivered through the Gemini Enterprise app: one interface where employees assign tasks that run in the background, in the cloud, with context carried across devices. The agent is meant to work inside Gmail, Drive, Docs, Sheets, and Calendar, and to be reachable from third-party surfaces including Slack and Microsoft 365. The status matters: it is in private preview for enterprise customers only. Google separately launched a consumer Gemini agent, Spark, in May, which runs through a different interface.
Spanish-language coverage of the event, which we could not confirm against Google's own materials, adds two notable details: agents that carry a fixed identity, with their own email address, storage, and persistent memory, and routing across model families that today includes Anthropic's Claude alongside Gemini, with other providers promised. If accurate, that makes the work agent less a Gemini feature than an orchestration layer that happens to be sold by a model vendor. We found no pricing.
The contrast with Microsoft is the strategic point. Microsoft 365 Copilot and Agent 365 anchor on Microsoft's own identity and security stack; Salesforce anchors on customer data. Google is trying to anchor on the interface employees already use and let it reach into rivals' applications.
The CODEW Lens: If one agent becomes the primary interface to workplace software, advantage shifts from owning individual applications to controlling the context, permissions, and actions that connect them. A private preview proves intent, not that employees will give up their apps.
2. Cisco Turns Collaboration Into an Agent Venue
At WebexOne in Austin on October 7, Cisco described "digital teammates" that join Webex spaces, meetings, and calls and handle multi-step work spanning Webex and third-party applications; its example is tagging an analytics agent for usage data, then asking a presentation agent to update a deck. Anthropic's Claude Managed Agents will be added to Webex group conversations, with Cisco saying "soon." A Model Context Protocol integration will bring Webex context to OpenAI's Dots agents, which are planned to appear later inside Webex as personal agents. Cisco also launched Dialog, a customer-experience platform in which agents keep working after a conversation ends, and said its devices will natively support Google Meet through a new Google partnership. Some capabilities begin in November; personal agents for Webex Calling are slated for the first quarter of 2027.
Notice what Cisco is not claiming: ownership of the system of record. It is offering to be the place where people and agents from competing vendors work together, secured and governed on Cisco's network and security foundation. One industry analyst quoted in coverage described the goal as keeping Webex the place where people work rather than ceding that role to other agent interfaces.
The CODEW Lens: Cisco is extending collaboration software into a neutral venue for execution, not becoming a broad execution platform. That is a defensible position only if conversations remain where work gets assigned, which is exactly what Google's single-interface agent contests.
3. Apollo Targets the Governance Gap Underneath
Apollo GraphQL's October 7 launch of GraphOS Agent Services addresses a problem the application vendors mostly skip: the APIs agents call were built for human developers, and return opaque responses security teams cannot reason about field by field. Agent Services sits between agents and enterprise systems, modeling those systems as a graph so policy can be applied per field, against the agent's identity and the person it acts for. Agents connect to a single endpoint and never hold upstream credentials; every request and denial is logged for audit, and a misbehaving agent can be cut off through a policy change enforced at the gateway. Apollo names Intuit as an early user; that is the only customer example we found.
Apollo's announcement calls the product a public preview, while its documentation still describes onboarding with an Apollo team member. Treat availability as limited until that is reconciled.
The CODEW Lens: The enterprise agent race is not only about intelligence. It is about a governed bridge between an AI system and the applications where data and permissions live. Apollo is betting that bridge is a neutral layer, not a feature of any one application vendor.
4. SAP Puts Vendors in the Execution Business
At SAP Connect in Las Vegas on October 6, SAP said its October update delivers on the "Autonomous Enterprise" vision it set out at Sapphire in May, when it pitched more than 50 assistants and 200 agents. Joule Work, which turns Joule from a chat window into a single work experience backed by agents, and the Autonomous Suite architecture are rolling out through October. New Joule agents for supply chain are planned for general availability in Q4, and new governance, risk, and compliance assistants are meant to keep agents safe as they execute. SAP says agents can connect with outside agents through the Agent2Agent protocol and that Joule Studio offers more than 70 models. It names Novartis, Nestlé, Morgan Foods, and PwC as customers, but those are SAP's own statements.
The difference from a general-purpose assistant is grounding. SAP's agents act inside finance, procurement, and supply-chain processes whose data and approval logic SAP already holds, which is also why SAP frames this as autonomy to be earned gradually rather than switched on.
The CODEW Lens: Industry-specific data and workflows may be the most durable advantage an established vendor has as models become interchangeable. The open question for SAP customers and implementation partners is whether agents shrink the services work that surrounds SAP projects or simply reshape it.
5. Salesforce Names the Control Layer
Salesforce's September 10 Trusted Enterprise AI Harness is an architecture, not a single product, wrapped around Agentforce rather than inside it. It bundles six capabilities (trusted context, agency, action, governance, security, and models) and adds an AI Control Plane meant to discover, register, monitor, and control agents across the enterprise, including those built outside Salesforce. Salesforce's stated thesis is that the model matters less than the proprietary context a company brings and its ability to turn that context into action safely.
The caveats are specific. Much of what sits under the Harness, such as Data 360, MuleSoft, and Tableau, already ships; the unified experience is slated to begin rolling out in early fiscal 2028, and there is no pricing or packaging. Reviewers have flagged two buyer questions: whether Salesforce hits that timeline, and whether the Control Plane governs third-party agents as well as it governs Agentforce. We found no independently verified customer outcomes for the Harness itself.
The CODEW Lens: Salesforce wants to manage not just the assistant but the conditions under which AI may act across business systems. That is the same ground ServiceNow and Microsoft are contesting, which makes interoperability the real test.
The Cross-Story Analysis: Three Contests
Who owns the interface? Google wants one agent in front of every application; Cisco wants the conversation to be that front door; SAP wants Joule Work to be the single experience for SAP processes. Each is a bid to become the place employees start. They cannot all win, and none has shown that employees will abandon their existing tools.
Who controls execution? This is where the stack is most contested and least settled. Salesforce and SAP govern agents from inside their applications' data; Apollo governs from the API layer beneath all of them; Cisco governs from the network and collaboration layer above. Google's agent must reach into others' systems and so depends on their permission models. An enterprise running several of these would have several overlapping answers to the same question: what is this agent allowed to do, and where is the record?
Who captures the value? The evidence is thin. We found no Google pricing and no Salesforce pricing, and the customer proof points (SAP's named customers, Apollo's Intuit example) come from the vendors themselves. If agents reduce the number of human interactions with software, per-seat licensing faces pressure; if governance, identity, and API controls become mandatory, vendors that supply them gain pricing power. Which effect dominates is not yet visible.
The CODEW Takeaway
The competition is no longer about who has the most capable assistant. It is about who owns the layer that decides what an agent may see, do, and be held accountable for.
Every vendor above is making a version of the same argument: durable advantage lies in controlling context, permissions, and execution together, not in the interface. What differs is the layer each is positioned to own, and that difference is what to watch.
| 01 | Interoperability tests: whether Salesforce's Control Plane and Google's agent work as well across rivals' systems as inside their own. |
| 02 | Preview-to-production conversion: Google's private preview, Apollo's preview, and Salesforce's fiscal 2028 timeline are all unproven until customers report results. |
| 03 | Pricing: the first published agent pricing from Google or Salesforce will show whether vendors expect to defend per-seat revenue or replace it. |
Sources
→ The Verge (excerpt): Google is launching a one-stop Gemini agent for your work tasks
→ Ecosistema Startup: Gemini de Google se vuelve agente único en Workspace
→ Cisco Newsroom: Cisco unveils new agentic collaboration experiences
→ SiliconANGLE: Cisco broadens agentic features in Webex
→ InfoSec Today: Cisco brings collaborative agents and Claude into Webex chats
→ Apollo GraphQL: Introducing GraphOS Agent Services
→ Apollo documentation: GraphOS Agent Services
→ CIO: SAP advances its autonomous AI agenda
→ SiliconANGLE: SAP expands Joule into an agentic work layer
→ SAP News: Joule assistants and agents for autonomous supply chain
→ Salesforce Ben: Salesforce launches Trusted Enterprise AI Harness
→ Martech Notes: Salesforce previews Trusted Enterprise AI Harness
Editorial Note
Enterprise Software Watch tracks what is changing now in enterprise software. This edition covers announcements dated September 10 and October 6 to 8, 2026, and treats them as recent developments, not October 9 launches. Availability is stated as each vendor describes it; private-preview and planned features are not presented as generally available.
Educational content only. Not investment or purchasing advice. Product capabilities, customer names, and performance claims are vendor statements and have not been independently verified by The CODEW. Two items in this edition involve Anthropic's Claude as a partner model; the analysis is editorial and independent of those relationships.
Reviewed by Erwin Castro
on
Friday, October 09, 2026
Rating:
No comments: