Live from the show floor at AWS re:Invent 2025, I sat down with Hong Wang, co-founder and CEO of Akuity, and the original creator of the Argo Project, to talk about a shift that’s showing up everywhere at re:Invent this year: less “AI theater,” more practical automation that reduces toil for SREs and platform teams.
Hong’s message was clear: the enterprise software delivery problem is no longer just Kubernetes delivery. Most organizations are operating across a messy reality of Kubernetes + VMs + serverless + Terraform, and they want a single, GitOps-style promotion model that can bridge old and new without forcing a rip-and-replace modernization.
From “Run the Model” to Practical AI for SRE and Platform Engineering
Hong described 2025 as a year where the market is getting more grounded about AI. Instead of focusing on model hosting as the end goal, teams are increasingly asking: where’s the ROI? Which workflows get faster, safer, and cheaper?
“Right now we’ve reached the era of looking at something very practical,” Hong said. “We are providing AI SRE capability on top of our platform now… to help the SRE or platform engineer be more efficient.”
That framing aligns with what we consistently see in our research: complexity and skill gaps are the most persistent blockers in modern operations. Platform engineering teams don’t lack tools; they lack time, consistency, and enough experienced humans to manage change safely at scale.
GitOps Promotion Across More Than Kubernetes
At re:Invent, Akuity announced expanded support for multi-cloud, multi-region, and multi-environment promotion, and critically, promotion that isn’t limited to Kubernetes.
Hong explained that customers want a unified approach that can also deploy and promote workloads tied to:
- Serverless
- VM-based deployments
- Terraform
- OpenTelemetry tooling (as part of the broader delivery and operational system)
Kubernetes still represents the center of gravity for modern platforms, but it’s rarely the whole estate. Enterprises are asking for a delivery control plane that reflects the real world: some workloads are containerized, some aren’t, and many are mid-migration.
As Hong put it, “They are looking for a unified solution… not just deploy and promote to Kubernetes, but also deploy and promote to the other environments.”
Bridging Heritage Systems and Cloud-Native Futures Without a “Big Bang” Cutover
One of the most practical parts of the discussion was Hong’s perspective on modernization timelines, especially for large enterprises.
Financial services and other highly regulated, high-scale sectors can’t “move everything to Kubernetes” in a quarter. They’re running systems that are literally tied to multi-billion or trillion-dollar businesses. The migration path is long, and the reality is hybrid for years.
“They don’t want… a cut over one night,” Hong said. “It’s actually a much longer journey.”
What’s interesting here is the framing. The maturity model isn’t purely organizational; it’s application-by-application.
Hong made the point that some systems are “money-making machines,” stable, low change velocity, and not worth disturbing until there’s a strong business driver. Meanwhile, new services move onto Kubernetes quickly. So the delivery platform needs to support both the stable legacy core and the fast-moving modern edge, without fragmenting operational workflows.
The Day 2 Problem Gets Worse as AI Generates More Code
Hong also connected the dots between AI-assisted development and operational burden. If AI is producing more code, teams will ship more features, and that shifts the bottleneck downstream into operations.
“A lot of code is being auto-generated… people are getting more productive,” he said. “Then the problem is starting shifting to another area… who is going to maintain it? Who is handling Day 2, Day 3?”
This is a critical insight. Even if AI accelerates Day 0 (creation) and Day 1 (release), the long-term cost is paid in Day 2 operations: drift, misconfigurations, noisy incidents, and fragile production changes that propagate quickly.
Why Human-in-the-Loop Still Matters for Agents in Production
We also talked about trust. When agents can take actions across infrastructure, mistakes can propagate rapidly. Hong’s position is pragmatic: human-in-the-loop remains the default for the next several years, especially for platform and infrastructure layers where blast radius is high.
“I do think that human in the loop will still be the trend… the main reason is context,” he said. Every enterprise has its own preferences, standards, and “local best practices” that aren’t naturally encoded in foundation models.
To bridge that gap, Akuity is emphasizing human-readable runbooks as a practical control mechanism: explicit guidance on what automation is allowed to do, what is not allowed, and what the expected outcomes should be.
“You want to be specific about what do you allow the AI to do, what do you not want the AI to do,” Hong said.
This runbook-driven approach is an important middle step: it gives teams a structured path to adopt AI for operations without handing over the keys to production before risk tolerance and governance are ready.
Analyst Take
Akuity’s re:Invent message is well-timed. GitOps and promotion workflows must evolve past a Kubernetes-only worldview. Enterprises aren’t choosing between “legacy” and “cloud-native;” they’re running both simultaneously, and they need a delivery approach that can bridge the gap without multiplying tools or rebuilding processes for every environment type.
Two signals stood out:
- GitOps is becoming an enterprise promotion layer, not just a Kubernetes pattern.
Supporting serverless, VMs, Terraform, and multi-environment promotion reflects what real estates look like. Unified promotion matters more than ideological purity. - AI accelerates software creation, but amplifies Day 2 risk.
If AI generates more code, operational load increases unless teams also invest in AI-assisted SRE capabilities, automation guardrails, and human-auditable workflows.
The practical path forward looks like: adopt AI where it reduces toil, keep humans in the loop for high-blast-radius actions, and codify “how we operate” into runbooks so automation can be scaled safely. If the industry is moving from “copilots that write code” toward “agents that run systems,” Akuity is positioning around the reality that operations still needs control, provenance, and context, not just speed.
Akuity Expands GitOps Beyond Kubernetes
Nutanix NC2 Becomes Generally Available on Google Cloud
Axonis Exits Stealth with Federated AI Architecture
Google Extends MCP Across Cloud Services
Databricks Advances Agentic AI with Agent Bricks and Unity Catalog Governance
