> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cloudthinker.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Chapter 10 · The Road Ahead

> Plan for five near-term shifts in agentic operations, and choose between hyperscaler-native agents and one platform across every cloud.

*Five shifts to plan for, why timing matters, and the one architecture choice every 2026 buyer faces: a native agent per cloud, or one platform across all of them.*

This chapter covers five shifts to plan for, how to choose between hyperscaler-native agents and one platform across clouds, and what to do first. The CloudThinker mapping for each practice is on [About CloudThinker](https://docs.cloudthinker.io/learn/aio/about-cloudthinker.md).

## 10.1 Five near-term shifts

Each row pairs a shift with one thing you can do this quarter.

| Shift | What changes | What to do now |
| :- | :- | :- |
| **Response to prevention** | Once agents remember past incidents, effort moves from fixing incidents to stopping them: risk checks before a deploy, capacity moves before a spike, design advice drawn from patterns across your estate. | Feed the memory layer on purpose: topology, conventions, past incidents ([§9.3](https://docs.cloudthinker.io/learn/aio/chapter-9-roadmap.md)). |
| **Agent-to-agent operations** | Your operations agents will hand work to vendor agents, such as a cloud provider's support agent. MCP (Model Context Protocol) already connects agents to tools. The A2A (Agent2Agent) protocol, now under the Linux Foundation, covers messages between agents across company boundaries. AWS DevOps Agent already escalates to AWS Support with its investigation attached. | Prefer platforms that speak MCP and A2A. In CloudThinker, an [MCP connection](https://docs.cloudthinker.io/guide/connections/mcp.md) reaches tools you add later, and the [A2A API](https://docs.cloudthinker.io/guide/api/overview.md#a2a-api-for-outside-callers) lets an outside system, such as ServiceNow, ask a CloudThinker agent a question (Scale plan). |
| **Governance becomes law** | AI governance is moving from voluntary practice to legal requirement in some sectors, led by the EU AI Act. In regulated industries, a platform that can't produce audit evidence won't pass procurement. | Build the controls in [Chapter 6](https://docs.cloudthinker.io/learn/aio/chapter-6-trust-governance.md) before a regulator asks for them. |
| **Autonomy ratchets up** | As evidence builds up, more action classes move from approval to pre-approved, and agents take on whole workflows in bounded domains. | Raise autonomy one action class at a time, on measured results ([§7.3](https://docs.cloudthinker.io/learn/aio/chapter-7-humans-on-the-loop.md)). |
| **The operating model becomes the product** | As models become interchangeable, the difference between vendors moves to orchestration, domain depth, accumulated context, and trust controls. | Judge vendors on those four, not on which model they use. |

## 10.2 The strategic window

Two Gartner forecasts look like they disagree:

* [Task-specific agents in 40% of enterprise applications](https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025) by the end of 2026, up from under 5% in 2025.
* [More than 40% of agentic AI projects canceled](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027) by the end of 2027 ([§9.4](https://docs.cloudthinker.io/learn/aio/chapter-9-roadmap.md)).

They don't conflict. Together they sort organizations into two groups: those that move from experiment to production with staged autonomy, real governance, and honest measurement, and those that stall.

Getting to production early compounds:

* Every resolved incident adds to what the agents know about your systems.
* Every hour of toil removed goes back to engineering work.
* Operating cost stops growing in step with the estate.

## 10.3 Hyperscaler agents or one platform across clouds?

The hyperscaler agents covered in this book (AWS DevOps Agent, Azure SRE Agent, Gemini Cloud Assist) are good at what they were built for. Each goes deepest on its home cloud, follows its vendor's pricing model, and focuses on investigation, with automated fixes arriving more slowly.

If you run a single cloud, the native agent is a strong default. Most enterprises don't. Many regulated estates, including much of Southeast Asia's financial sector, mix hyperscalers with local or sovereign clouds and on-premise systems. Three single-cloud agents mean three consoles, three governance models, and three audit trails: the same swivel-chair problem this book argues against, one layer up.

| Dimension | Hyperscaler-native agent | One platform across clouds |
| :- | :- | :- |
| Coverage | Deepest on its home cloud, partial elsewhere | Every cloud you run, including local clouds and on-premise systems |
| Scope | Investigation first, fixes arriving gradually | The full detect, analyze, resolve, validate loop under one policy |
| Governance | Separate controls and audit trail per vendor | One approval policy, one audit trail, one place to approve |
| Data control | Processing in the vendor's cloud, controls vary | Identifier masking, your own model account, and private deployment options |
| Incentives | Optimizes inside its vendor's ecosystem | Cloud-neutral, including on decisions that cut a vendor's bill |
| Depth on a single cloud | Deepest: built by the cloud's own team | Usually shallower on any one cloud than that cloud's own agent |
| Setup effort | Low on its home cloud, inside an account and vendor you already have | A connection and credentials per cloud, plus a new vendor to review |

### Pick a column

Answer four questions:

1. How many clouds, local clouds, and on-premise environments do you run today, and how many will you run in two years?
2. Does your regulator or security team need one audit trail across all of them?
3. Must data stay inside a specific country or inside your own perimeter?
4. Do you need fixes carried out under approval, or is investigation enough for now?

Count the answers that point beyond one cloud or beyond investigation:

* **None:** start with the native agent.
* **One:** start with the native agent, unless that answer is a hard requirement, such as data that must stay in one country. Then treat it as two.
* **Two or more:** shortlist platforms that run across clouds, and test each with the eight data-control questions in [§6.3](https://docs.cloudthinker.io/learn/aio/chapter-6-trust-governance.md) and the five-question vendor test in [§3.3](https://docs.cloudthinker.io/learn/aio/chapter-3-what-is-aio.md).

### A disclosure

This field guide is published by CloudThinker, which builds in the second column. We kept that interest away from the evidence: benchmarks are attributed, the hyperscaler agents are shown at their strongest, and the framework chapters apply whatever platform you choose.

Hold us to the same two tests from the list above. [About CloudThinker](https://docs.cloudthinker.io/learn/aio/about-cloudthinker.md) maps each practice in this guide to a product page you can check.

## 10.4 What to do Monday

1. **Capture a baseline.** Export 90 days of MTTR, alert volume, and pages from your paging and ticket tools ([§8.2](https://docs.cloudthinker.io/learn/aio/chapter-8-measuring.md)).
2. **Pick one bounded domain** with frequent, repetitive, reversible incidents ([§9.2](https://docs.cloudthinker.io/learn/aio/chapter-9-roadmap.md)).
3. **Start every write on approval.** In CloudThinker, set the Approval mode to **Manual** before you connect anything, and under [Run types](https://docs.cloudthinker.io/guide/approvals/approval-mode.md#set-a-mode-for-a-run-type) override **Automations** to **Manual** too. A chat can pick its own mode, so check it there.
4. **Name the autonomy policy owner** before the pilot starts ([§7.2](https://docs.cloudthinker.io/learn/aio/chapter-7-humans-on-the-loop.md)).

## Next steps

Read [About CloudThinker](https://docs.cloudthinker.io/learn/aio/about-cloudthinker.md) to see how the ideas in this guide map to the product and how to start the 90-day pilot from [Chapter 9](https://docs.cloudthinker.io/learn/aio/chapter-9-roadmap.md) on your own environment.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.