This week in AI: agents now do real work, but the company still holds the approval

Three developments this week point the same way: AI is moving from demonstration to operations — where every task needs clean input data, a named person to approve it, and a log to check afterwards.

1. IBM and Microsoft launch an industry-specific agent platform

On 10 November 2026, IBM and Microsoft announced Imagine Enterprise, a platform for enterprises to build, run and govern AI agents that follow industry-specific processes. The interesting part is not the model but what surrounds it:

• IBM Consulting brings more than 6,000 prebuilt agents and assets plus delivery experience by industry; Microsoft supplies AI, cloud and security technology.
• The platform targets tightly regulated sectors such as financial services, insurance and retail — where AI cannot simply fail and be fixed later.
• It ships with forward-deployed engineering support for qualifying engagements instead of leaving enterprises to start from zero.

The takeaway: value comes from connecting agents to the right data and the right process with oversight in place — which is why the big vendors sell delivery together with the licence.

2. Google releases Gemini 4 Hydrogen for fast agentic tasks

On 12 November 2026, Google introduced Gemini 4 Hydrogen alongside Carbon and Argon. Carbon and Argon handle long-horizon reasoning; Hydrogen targets fast execution — everyday coding and multi-step agent workflows.

• Hydrogen ranks #5 on Agent Arena, a live benchmark for multi-step tasks such as deep research, data analysis and navigating large codebases.
• It scores 53.4% on the Multimodal Knowledge Agent (MKA) benchmark for end-to-end knowledge work over complex visual files.
• Google added Orchestrator in Antigravity, switching between Carbon and Hydrogen inside a single session instead of using one model from start to finish.

The implication: cost per AI task keeps falling, and splitting work across models — the strong one for hard reasoning, the fast one for repetition — becomes a normal design choice.

3. Vietnam: up to VND 3 billion per AI project

In Vietnam, the National Technology Innovation Fund (NATIF) has opened a funding channel for technology-innovation tasks that apply AI, with support of up to VND 3 billion per task. The core condition is a real business problem with data and a deployment plan — the application has to show what is done, on which data, and who uses it. Separately, a draft for a national AI development fund proposes up to VND 30,000 billion from the state budget, disbursed in stages.

Three things to do now, without waiting

1. Pick one process, not “the whole company”. A process with clear input and output that repeats weekly: receivable reconciliation, attendance and payroll, purchase order handling. A narrow scope can be measured in 4–8 weeks.
2. Clean the data before choosing a model. All three stories revolve around data: industry agents need standardised data, fast models only help when input is tidy, and the funding application must prove the data exists.
3. Design the approval step from day one. AI proposes, an authorised person approves, and every action is logged. This is not a brake on the project; it is the condition for putting AI into a process that involves money and compliance.

How VASCOM reads this week

We do not sell a model. Our work is to put AI where it belongs inside a company: review the processes already running, decide which parts AI can take and which must stay with a person, build an agent per department, connect it to existing data, and run the system on the company’s own infrastructure so data does not have to leave. Each step ends at a decision point: the company agrees before the next step starts.

This is slower than a demo, but it matches what the major vendors announced this week: an agent only earns its place when it runs a real process, with an accountable owner and a log.

Sources

• Microsoft Source, “IBM and Microsoft announce a new AI agent platform designed for industry-specific transformation”, 10 Nov 2026 — news.microsoft.com
• Google, “Introducing Gemini 4 Hydrogen”, 12 Nov 2026 — blog.google
• Google DeepMind, “Gemini 4 Hydrogen — Model Card”, 12 Nov 2026 — deepmind.google
• Vietnam Government Portal, “Call for AI technology-innovation tasks, funding up to VND 3 billion”, 16 Sep 2026 — baochinhphu.vn
• Dan Tri, “VND 30,000 billion from the budget for the national AI development fund” — dantri.com.vn