After Ten Years of Transformation, the Next Chapter Is Operational Excellence

In short. A decade of transformation made our companies more capable and much harder to run. I believe the next step is a new operations layer, where agents do the investigation and people make the decisions, adopted carefully, one team at a time.

Before CloudThinker, I spent years running production systems at startups, at FPT Software and at AWS. Many of my hardest nights followed the same pattern. An alert would fire, and I'd spend hours in dashboards, logs and recent changes before finding the cause. The fix often took minutes. Finding it was the real work.

I still see that pattern in almost every company I visit, and in engineering teams that are tired in a way that has little to do with the hours they work. The systems are bigger and the tools are better, yet the hardest part of the job is still the same: understanding what's actually happening.

Many of the examples in this series come from enterprises I work with in Vietnam and across Southeast Asia, so a few parts look at Vietnamese costs and regulation. The ideas apply wherever production runs.

We built a lot

Over the last decade, most of us led one transformation after another. Cloud. Microservices. DevOps. And now AI. Each of them was the right call, and together they made our companies faster and more capable than we could have imagined ten years ago.

They also gave our teams far more to run. Organizations now estimate that 29% of their cloud spend is wasted, the first increase in five years.[1] Most of us didn't set out to waste money. Our estates simply grew faster than our ability to look after them.

Better and harder at the same time

What strikes me most is that things are improving and getting harder at once. Outages are becoming less frequent, yet one in five major outages now costs more than $1 million.[2] AI helps teams ship faster, and it's also linked to more change failures and rework.[3]

I read this as a shift in where the difficulty lives. It used to sit inside individual machines. Today it sits in the connections between hundreds of services, tools and teams. And the people connecting those dots, often late at night, are our engineers.

Ten years later: more services, more connections, and five pressures growing together. Cost, reliability, security, velocity and people.

Operational excellence comes next

I believe the next chapter is operational excellence: running everything we built, well, and getting a little better at it every day. It's a practice rather than a project, and it moves in a simple loop.

01Organize

Customize

Every team owns its outcomes, with shared goals, in its own workspace.

02Prepare

Review and Cyber

Every change and every weakness is checked before production.

03Operate

Resolve

Every issue is investigated, and the root cause arrives in minutes.

04Evolve

Optimize

Every fix becomes company memory, and every cost is optimized.

Organize, Prepare, Operate, Evolve. Every day.

I was glad to see Google make a similar point in October, in the second edition of its Site Reliability Engineering book. The authors write that the challenge is now "maintaining high reliability in the face of the immense complexity," and they describe AI as a force multiplier for reliability engineers.[4] They also note what hasn't changed: culture. Excellence is still something teams choose to practice together.

Most leaders I speak with already agree with all of this. The difficulty has always been doing it consistently, across every team, with the people we have.

One new layer

This is where I see AI making a real difference. Keep the stack you have. Keep your teams where they work today. Add one layer in between, where people and agents work together.

The operational excellence layer connects what we built to the outcomes the business needs.

In this layer, agents do the investigation. They look at every alert, every change and every weakness, follow the evidence across logs, metrics, traces and code, and come back with the likely root cause and a proposed fix. Engineers make the decisions. They approve, they take the unusual cases, and they set the direction.

Think back to those nights on call. The agent does the hours of searching. The engineer spends a few minutes deciding, and gets their evening back. Gartner sees the same direction for infrastructure and operations, with people shifting from doing the work to supervising it.[5]

At CloudThinker, the agents in this layer are our Frontier Investigation Agents: Review, Cyber, Resolve, Optimize and Customize.

Trust is earned, one team at a time

I want to be careful here, because enthusiasm can run ahead of reality. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, because of rising costs, unclear value or weak risk controls.[6] I don't read that as a reason to wait. I read it as a reminder that how we adopt agents matters as much as whether we do.

In my experience, trust grows from a few simple habits. Each team works in its own isolated workspace. Agents only get the access a task needs. Every action is recorded. And every action waits for a person to approve it, until the results earn more autonomy.

Two questions deserve an answer before you start. The first is cost: Gartner expects AI inference costs per agentic workflow to rise more than fivefold through 2028, so the value has to be measured, not assumed.[7] The second is control: where your data goes and which models you rely on. Choosing between frontier, open-source and self-hosted models should stay in your hands, especially as data and AI sovereignty rise on the CIO agenda.

The memory stays

There's one more benefit that I think is underrated. Every loop between agents and engineers leaves something behind: a root cause, a fix, a decision, a correction. Over time, that becomes company memory.

Agents deliver the root cause and the fix. Engineers decide, and their feedback makes the agents better. Every loop becomes company memory.

Today, much of what our teams know lives in a few people's heads and in postmortems that are hard to find when they're needed most. A shared memory changes that. A root cause found this month can quietly protect next month's release. Teams will change, as they always do. What they learned can now stay.

What this means for a CIO

The CIO conversation has already moved on from experimentation. As one Gartner analyst put it, "2026 will be about delivering agentic AI ROI." The same research found 57% of CIOs under pressure to improve productivity, 52% to reduce costs, and only 48% of digital initiatives meeting their business targets.[8]

That's why I think operations deserves a place near the top of the AI agenda. The work is measurable, the cost is visible, and the people doing it are already stretched. I see five shifts taking shape.

Shift	What it means
From cost reports to cost control	Waste is found and fixed every day, not reviewed once a quarter.
From uptime targets to resilience you can price	Time to understand and recover becomes a business number, linked to revenue at risk.
From headcount to capacity	The same team takes on more, and skilled people spend less of their week on toil.
From AI pilots to measurable return	Operations is one of the clearest places to measure AI value: hours returned, recovery time, spend reduced.
From knowledge in heads to a company asset	What the team learns stays, even as people change roles.

I've turned these shifts into five questions for your next leadership meeting, along with a value model you can fill in with your own numbers.

If I were starting today

I'd start small and stay curious. I'd pick the measures that matter to the business and take a baseline. I'd ask agents to investigate last week's issues with read-only access, then compare their findings with our own postmortems. And I'd grow one team at a time, letting the results, not the excitement, decide how fast we move.

The series

This post is the map. Each part goes deep on one piece of it.

The hidden cost of incident time: where the minutes in an incident go, and why investigation is the clock to measure.
Why incident investigation is slow: complexity, team silos and fragmented knowledge, and a metric called time to remember.
The operational excellence layer: Well-Architected, the four areas and the eight design principles, mapped to agents.
Agents on a unified layer: what agents do in operations, the shared layer they run on, and how company memory forms.
Control before trust: the risks of agents in production, the controls that answer them, and a 90-day rollout.
From single agents to a shared platform: agent sprawl, what to standardize once, and who owns what.
The economics of operational excellence: the payroll cost of incident time, a value model for your own numbers, and the weekly scorecard.

Why this matters to me

We spent ten years building remarkable systems. I'd like the next ten to give something back to the people who run them: calmer nights, fewer repeated problems, and more time for the work that made them choose this profession in the first place.

Engineers get their time back. For creative work.

Sources

Flexera, 2026 State of the Cloud Report.
Uptime Institute, Annual Outage Analysis 2026.
Google Cloud DORA, 2025 State of AI-assisted Software Development.
Betsy Beyer and David Huska, "10 Years of the SRE book: Announcing the second edition of Site Reliability Engineering," Google Cloud Blog, October 12, 2026. cloud.google.com
Gartner, Predicts 2026: AI Agents Will Reshape Infrastructure & Operations.
Gartner press release, June 25, 2025: over 40% of agentic AI projects will be cancelled by end of 2027.
Gartner press release, August 17, 2026: AI inference costs per agentic workflow will increase more than fivefold through 2028.
Gartner, "The CIO Agenda 2026" and press release, October 21, 2025, on the 2026 CIO and Technology Executive Survey (2,501 respondents).