AI in Business, October 2026: Agents Grow Up, Pilots Stall

In October 2026, Google and OpenAI released workplace agents that run with their own identities, budgets and audit trails. In the same period, Gartner reported that most companies still have not scaled AI and warned against depending on vendor engineers to build agents. For business leaders, the lesson is to own your agents and govern them centrally.

What happened in AI for business this month?

The EU item is older, but its deadlines shape planning for any company with European customers, so we include it.

What did Google announce for Gemini business agents?

What happened. According to TechCrunch (8 October 2026), Gemini agents can take an objective, plan, use custom skills and delegate to subagents. They connect to Google Workspace, Microsoft 365, Slack, Jira, BigQuery, Snowflake and any MCP server. Each agent gets its own Workspace account and email address, plus an audit trail. Admins get real-time spending caps, and the system routes work between models. Businesses get access first.

Why it matters. An agent with its own account can be given permissions, monitored and switched off like an employee account. Spending caps acknowledge that autonomous work can run up costs quickly.

What we take from it. Agent identity, audit trails and budget caps are now offered by the largest vendors, so buyers should expect them from every supplier. For Vietnamese companies with data residency obligations, the next questions are where the agents run and which models see which data.

What did OpenAI launch at DevDay 2026?

What happened. According to Reworked (1 October 2026), OpenAI used its 29 September DevDay to introduce Dots, always-on agents that connect to more than 4,000 apps. Dots are off by default for Enterprise, Edu and Healthcare customers until admins enable them. OpenAI also launched ChatGPT Space, a shared workspace for people and agents, an Agents API with computer use, and a Decisions API.

Why it matters. Agents are moving from chat windows to background work. The off-by-default setting for enterprise customers shows that admin control over agents is now expected.

What we take from it. Admin gating at the workspace level is the right default. Shared spaces also match what we see in practice. Agent output is more useful when the team can see it than when it sits in one person's private chat.

How many companies have actually scaled AI?

What happened. According to Gartner (1 September 2026), only 22% of organizations have scaled AI across multiple business units. The survey covered 1,303 respondents at firms with at least US$50 million in revenue, from January to April 2026. 85% of functional leaders plan to raise AI spending, and about 11% do not know what their functions spent on AI.

Why it matters. Spending is rising faster than scaling. A company that cannot see what it spends on AI cannot tell which uses are worth expanding.

What we take from it. Scaling is an operating problem. It needs shared skills that every team can reuse, budgets by team and clear visibility of spend. Before raising the AI budget, find out who uses what.

Why does Gartner warn about vendor-built agentic AI?

What happened. According to Gartner (29 September 2026), 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineers (FDEs) by 2028. Gartner warns of "FDE washing" and advises companies to settle governance, IP ownership, knowledge transfer and exit terms up front.

Why it matters. Agents built by a vendor's engineers on the vendor's stack can become hard to maintain once those engineers leave. The company pays for the build and then cannot change it.

What we take from it. We agree with Gartner's advice. In our deployments, company know-how is captured as skills the company owns and its own staff can edit in plain language. The platform runs on the customer's infrastructure. When we help set things up, the result stays with the customer.

What changed in the EU AI Act timeline?

What happened. According to Cooley (3 August 2026), the EU Digital Omnibus on AI, Regulation (EU) 2026/1744, was published on 24 July 2026 and entered into force on 27 July 2026. It moves the deadline for Annex III high-risk systems to 2 December 2027 and for Annex I systems to 2 August 2028. Article 50 transparency duties still apply from 2 August 2026.

Why it matters. Vietnamese companies that serve EU customers or belong to European groups have more time for high-risk obligations. Transparency duties already apply.

What we take from it. Delays to high-risk deadlines do not remove transparency duties. Vietnam's AI Law 134/2025/QH15 has applied since 1 March 2026, and according to legal summaries its Article 11 requires labeling of AI-generated media that simulates real people or events. Companies working across both markets should plan for the stricter rule.

How is AI adoption progressing in Vietnam?

What happened. According to Báo Đầu tư (30 September 2026), an AWS and Strand Partners study found that about 26% of Vietnamese businesses, roughly 245,000 companies, now use AI, up from 18%. 61% are exploring or testing AI, but only 8% have built it into their processes. 38% are aware of agentic AI and 19% are piloting agents. 72% of AI users report productivity gains, and finance leads adoption at 41%.

Why it matters. The distance between 26% and 8% is the gap between pilots and processes. Vietnam shows the same pattern Gartner found among large global firms.

What we take from it. Most of the value sits in closing that gap. Recurring work in finance and operations is a good place to start, because it has clear inputs, clear owners and outputs that are easy to check.

What are the three takeaways for business leaders?

1. Agents are becoming accountable colleagues. Google and OpenAI both give agents identities, admin controls and audit trails. Treat an agent like a new hire: give it a role, permissions, a budget and a supervisor.
2. Owning beats renting. Gartner's FDE warning and the scaling data point the same way. Companies that own their agents, skills and knowledge can keep improving them. Companies that rent vendor engineers often have to start again.
3. Scale comes from processes and governance. Only 22% of large firms and 8% of Vietnamese businesses have moved AI into processes. Governance at the workspace level, with budgets, data rules and audit, is what lets a pilot become daily work.

How Redstone reflects these trends

Redstone was designed around the same ideas.

• Agents with identity and an audit trail. In the AI Factory, each agent acts as the person assigned to it, with that person's permissions in Odoo, 1Office, Jira, Google Workspace and Outlook. Every step is written to an audit log with CSV export.
• You own your agents and knowledge. Redstone is self-hosted and can run on-premise. Skills live in a shared, versioned library that the agent proposes changes to and a person approves. The knowledge base answers with citations, and company data is not used to train models.
• Governance at the workspace level. The governance console sets token quotas by user, group and cost center, warns at 80% usage and switches to a local model at the cap. Four data levels route work to the right models, and a kill switch stops all agents at once.
• From pilots to processes. The AI Factory assigns one agent to each stage of the value chain, and autopilot creates recurring work every day. The scheduler, missions board and morning briefing turn one-off experiments into routines people review each morning.

Book a briefing to see how one of your workflows would run in Redstone.