This guide is for heads of engineering, platform leads and CTOs who own how Kubernetes applications reach production. It sits beside Netbase's cloud platform engineering service and covers one decision inside DevOps and platform engineering work: how manifests are packaged and how they get into each cluster.
In this guide
- What are you actually choosing between?
- Packaging: Helm, Kustomize or both?
- Delivery: pipeline push, Argo CD or Flux?
- Which pattern fits your team maturity and environment count?
- How do you promote a release across environments?
- When should you not adopt GitOps yet?
- Where AI fits in release engineering
- What delivery record exists, and what does not
- Alternatives: other ways to release to Kubernetes
- What are the limits of this guide?
- Frequently asked questions
- Next step
What are you actually choosing between?
You are making two choices, not one. The first is packaging: how one set of Kubernetes manifests becomes the slightly different manifests each environment needs. The second is delivery: whether a pipeline pushes those manifests into a cluster, or a controller inside the cluster pulls them from Git and keeps them in place.
Helm and Kustomize answer the packaging question. Helm installs a chart, a package of templated manifests, and each install creates a release object that can be upgraded and rolled back with helm rollback; environments differ through values files and --set overrides, as the Helm documentation describes. Kustomize patches plain YAML instead of templating it, and the Kubernetes documentation notes that kubectl has supported it since version 1.14 through kubectl apply -k.
Argo CD and Flux answer the delivery question. Both follow the four principles of OpenGitOps: desired state is declarative, versioned and immutable, pulled automatically by software agents and continuously reconciled against what is running. Both read Helm charts and Kustomize overlays, so the packaging choice does not lock in the controller.
Packaging: Helm, Kustomize or both?
| Option | What it does well | What it hides or costs | Choose it when |
|---|---|---|---|
| Plain manifests | Every line is reviewed as written | Copies drift apart with every new environment | One environment, a handful of services |
| Kustomize overlays | Base plus small patches per environment; the diff is plain YAML | Large structural differences become many patches | Your own services across two to five environments |
| Helm charts | Versioned packages, values per environment, release history | Templates hide the final YAML from code review | Third-party software, or software you ship to other teams or customers |
| Helm plus Kustomize | Vendor chart, local patches on top | Two tools to learn and two places a value can come from | You consume vendor charts and must patch what their values do not expose |
A workable decision rule: Kustomize for the services your own team writes, Helm for software someone else maintains, and both together only for vendor charts that need local patches. If reviewers cannot see the final manifest of a change, render it in the pipeline and attach the output to the merge request, whichever tool produced it.
Delivery: pipeline push, Argo CD or Flux?
A pipeline that runs kubectl apply or helm upgrade after a merge is deployment automation. It works, but nothing checks afterwards that the cluster still matches Git, and the pipeline needs credentials to every cluster it deploys to. A GitOps controller reverses that: it runs inside the platform, pulls from Git and corrects drift on its own.
-
How Helm runs
- Argo CD
- Renders charts with
helm template; Argo CD, not Helm, owns the lifecycle, sohelm lsdoes not show the app - Flux
- Helm controller performs real installs, upgrades, tests and rollbacks from a
HelmRelease
-
Helm test hooks
- Argo CD
- Not supported; most other Helm hooks are mapped to Argo CD hooks
- Flux
- Run as part of the release
-
Many clusters
- Argo CD
- One Argo CD can manage and deploy to multiple clusters
- Flux
- Controllers run in the cluster they reconcile
-
Visibility
- Argo CD
- Web UI with a real-time view of application state, SSO and RBAC
- Flux
- Command line and Kubernetes resources; add a dashboard if people need one
-
Rollback
- Argo CD
- To any configuration committed in Git
- Flux
- Automated remediation on failed install, upgrade or test, plus Git revert
-
Ordering
- Argo CD
- Sync waves and hooks
- Flux
dependsOnbetween Kustomizations and HelmReleases
The selection criteria in one line each: pick Argo CD when people outside the platform team need to see what runs where, or when one central team manages many clusters. Pick Flux when charts depend on Helm's own lifecycle, tests and remediation, or when each cluster should carry its own controller and nothing central holds credentials to all of them.
Which pattern fits your team maturity and environment count?
The fit matrix below is the core of this guide. Read across from the row that describes your team today, not the team you plan to have.
-
Start
One small team, one or two environments, one cluster
- Packaging
- Plain manifests or one Kustomize base
- Delivery
- Pipeline push with a manual approval for production
- Move to the next stage when
- A second environment copies the first by hand
-
Overlays
One team, development, staging and production
- Packaging
- Kustomize base and one overlay per environment
- Delivery
- Pipeline push, rendered diff in every merge request
- Move to the next stage when
- Someone changes a cluster by hand and nobody notices for days
-
GitOps
Two or three teams, three to five environments, two or more clusters
- Packaging
- Kustomize for own services, Helm for third-party software
- Delivery
- One controller per cluster or one central Argo CD; promotion by pull request
- Move to the next stage when
- Teams wait on one platform engineer to release
-
Platform
Several teams, many clusters or tenants
- Packaging
- Shared Helm charts or Kustomize components owned by a platform team
- Delivery
- Central Argo CD with generated applications, or Flux per cluster with a shared fleet repository
- Move to the next stage when
- The platform team becomes the bottleneck; invest in self-service templates
Two signals matter more than headcount. The first is drift: if the cluster and Git disagree and nobody finds out, a controller pays for itself. The second is environment count: every extra environment multiplies copy-and-edit mistakes, which is the problem overlays and values files exist to solve.
How do you promote a release across environments?
-
Separate code from configuration
Keep application code and the environment repository apart, so a configuration change does not rebuild an image.
-
Build once
The pipeline builds and scans one image, tags it with an immutable reference and never rebuilds it per environment.
-
Update the first environment
The pipeline opens or merges a change that sets the new image reference in the development overlay or values file.
-
Let the controller sync
Argo CD or Flux pulls the change and reconciles; the pipeline waits on health checks rather than pushing.
-
Promote by pull request
Copy the same reference to staging and then production through reviewed pull requests, with the rendered diff attached.
-
Roll back in Git
Revert the commit that changed production; the controller applies the previous state, and Helm-native remediation in Flux can act first on a failed upgrade.
When should you not adopt GitOps yet?
Do not install a GitOps controller for one environment and one cluster that a single pipeline already deploys reliably; the controller becomes one more system to upgrade, secure and monitor. Do not adopt Helm for your own services only because vendor software uses it; templates make simple changes harder to review.
Do not start GitOps before secrets have a plan. Encrypted secrets in Git, an external secret store or an operator all work, but the choice has to be made before the first production sync, not during it. Do not move to many clusters to solve a problem namespaces and quotas would solve.
Where AI fits in release engineering
AI helps with the reading around a release, not with the release itself. An assistant can summarise a rendered manifest diff before review, draft a Kustomize overlay or values file from an existing one for an engineer to check, and group drift reports so the cause is easier to find. Netbase works with the major commercial and open-source AI tools and models, chosen per project.
Engineers still review every rendered diff, approve every promotion pull request and decide every rollback; no model merges to the environment repository or changes a cluster.
- In delivered work
MLOps pipeline
An MLOps pipeline Netbase delivered for a client that is not named; an AI delivery record, not a GitOps rollout.
- Available capability
AI agents in operations
AI-assisted diff review and drift triage is a way Netbase approaches the work, not yet tied to a published case.
What delivery record exists, and what does not
Netbase works as offshore development partner on a multi-tenant cloud ERP delivered as SaaS to small and mid-sized businesses for a US client, not named, and names AWS, Google Cloud, DigitalOcean and Cloudflare as the cloud providers it uses.
Delivery runs in a six-step lifecycle with weekly reviews and KPI dashboards, and most Netbase projects are delivered on fixed-price contracts. No published Netbase record describes a Helm, Kustomize, Argo CD or Flux rollout, and no release frequency, incident or result figure is claimed here.
Alternatives: other ways to release to Kubernetes
| Alternative | Fits when | Selection criterion |
|---|---|---|
| Managed containers without Kubernetes | Few services, no cluster skills on the team | The platform's own deploy and rollback cover the need |
| Pipeline push with rendered diffs | One or two environments, one cluster | Drift is rare and the pipeline holds the only credentials |
| A hosted delivery or platform product | The team wants a portal more than a controller | It reads the same Git repository, so leaving it stays possible |
| Stay on the current process | A migration or a major release is under way | Change one variable at a time |
Teams that want to know which stage they are at can score release and recovery practice with the DevOps maturity assessment scorecard. The health checks a controller waits on come from the signals in observability stack design with OpenTelemetry, and when the release pattern changes during a move to the cloud, the cutover plan for a cloud migration covers the window itself. For the migration as a whole, see cloud migration and modernization.
What are the limits of this guide?
- It compares patterns and decision rules, not tool versions; check the current Helm, Argo CD and Flux documentation before you commit.
- The stage boundaries are judgement, not thresholds; one team with strict compliance needs may belong a stage higher.
- It covers application releases; cluster creation and infrastructure as code are separate decisions.
Plan the next step with a Netbase consultant
Frequently asked questions
Use Kustomize for services your own team writes, because the change under review is plain YAML. Use Helm for third-party software and for software you ship to other teams, where a versioned package and values files are worth the hidden templates.
Neither is better in general. A small team with one cluster may not need either yet; once it does, Argo CD suits teams that want a web view of every application, and Flux suits teams that rely on Helm's own lifecycle and tests.
Yes. Kustomize can patch the rendered output of a vendor chart, and both Argo CD and Flux read Helm charts and Kustomize overlays. The cost is two places where a value can come from, so document which tool owns which setting.
Usually three or more, or two or more clusters, or any setup where clusters are changed by hand without anyone noticing. Below that, a pipeline that pushes reviewed and rendered manifests is often enough.
Next step
If you want a second opinion on the release pattern before you add a controller, book a solution review with Netbase.
Related services and solutions
AI-ready cloud platform engineering for SaaS and commerce workloads Netbase provides cloud platform engineering for SaaS and commerce teams that need a secure, repeatable home for their software and its AI features, on AWS, Google Cloud, DigitalOcean or Cloudflare. We design the architecture, write the environments as code, connect managed and AI services and hand over a platform your team can run, built the way we run ours.
Learn More
AI-assisted cloud migration with a cut-over plan and a way back Netbase provides cloud migration services for companies running business applications on ageing servers or hosting to move them to AWS, Google Cloud, DigitalOcean or Cloudflare without losing data or control. AI speeds up the code and dependency assessment; every migration still has an inventory, a rehearsed cut-over, cost controls from day one and a tested rollback point.
Learn More
AI-assisted DevOps and platform engineering for routine releases Netbase provides DevOps and platform engineering for SaaS and commerce teams to make releases routine instead of risky. We set up CI/CD pipelines, repeatable environments and observability on AWS, Google Cloud, DigitalOcean or Cloudflare, with AI assistance for tests, reviews and incident triage, so every change ships the same way and problems surface before customers notice.
Learn More
SaaS product accelerator: launch an AI-ready SaaS on proven Netbase modules A SaaS product accelerator is a set of reusable Netbase modules for accounts, billing, roles and integrations that helps founders and product teams launch subscription software faster, with room for in-product AI from the first release. Reusing these modules can shorten development time, and the approach is proven on Printcart, Netbase's own web-to-print SaaS.
Learn More
Work with Netbase remotely
Every Netbase engagement is delivered remotely, so clients in any country and any industry can work with us; Netbase has worked with clients in 80 countries. Contracts and invoices are issued by Netbase JSC, and payment is made by bank transfer to Netbase JSC's company account or PayPal Business. Talk to the CEO, Huy Nguyen (David), on WhatsApp or by email.
Discuss a project
Netbase JSC helps organizations design, build, modernize, and operate digital products and AI-enabled business systems.project@netbasejsc.com
+84 937 869 689
91 Nguyen Chi Thanh, Dong Da, Hanoi, Vietnam
Get in touch
Tell us what you want to build, modernize, or operate.