Private aviation people understand tool maturity better than most industries. A airframe earns trust over years of dispatches, not in a launch-day press release, and the upgrades that matter are the ones you barely notice at the annual. The GitOps world has just passed through its own equivalent of that maturing curve, and late 2026 is a good moment to take stock of what actually ships, what graduated, and what is still taxiing. Short version for the busy reader: the two flagship tools are steady and feature-complete for most operators, the cloud vendors have stopped pretending this is a niche, and the interesting action has moved to the edges — promotion, secrets, and progressively delivered releases.
For anyone new to the term: GitOps is the operating model where the configuration you want is declared in a version-controlled repository, and an automated agent continuously pulls that declaration and reconciles your live systems to match. Changes go through reviewed merges. That is the machinery humming underneath everything discussed below, and if you want the formal principles behind it, the maintained GitOps entry at xOps School is the reference this piece was checked against.
GitOps Flagships: The Two Names Everyone Still Says
Argo CD and Flux completed their graduations within the Cloud Native Computing Foundation back in the winter of 2022 — the 30th of November for Flux, the 6th of December for Argo — and four years on, both projects behave like mature fleet types rather than experimentals.
Argo CD, currently at version 3.5.5, remains the hub-style choice: a central control plane with a UI that non-engineers can actually read, AppProjects to fence tenants apart, ApplicationSets to stamp out applications from templates, sync waves to order what gets applied first, and optional self-healing when you want drift corrected without asking. The interesting new entry is the Source Hydrator, in beta, which pre-renders manifests so that reviewers read finished output instead of template spaghetti.
Flux, at version 2.9.6, keeps the opposite shape: a slim agent on each cluster, controllers split by responsibility, dependencies expressed through dependsOn, with drift hunted down on whatever interval you set — and, quietly important for regulated operators, verification of PGP and SSH commit signatures before anything is applied. If your auditors ask who is allowed to change what, that last feature answers half the question by itself.
Pick by shape, not by fashion: one control room, or many cockpits. Both fly.
The Cloud Vendors Stopped Waiting
For years the managed-cloud answer to GitOps was a shrug and a blog post. That changed recently. AWS shipped an EKS capability for Argo CD in November 2025, meaning the reconciler is now a first-class part of the managed offering rather than a community add-on you weld on yourself. Microsoft’s AKS carries a managed Flux extension, and GitLab ships its own agent for pulling deployments straight from its repositories. The significance is less about features and more about defaults: when the platform console offers reconciliation as a checkbox, the operating model stops being an enthusiast position and becomes the way things are simply done. For smaller teams the arithmetic has changed accordingly — the “we cannot afford to run a control plane” objection has quietly retired, because someone else now runs it for you, patched and monitored, inside the bill you were paying anyway.
Below the Flight Deck: The Wider Fleet
A quick tour of the supporting squadrons, because the state of the ecosystem is bigger than two flagships:
- Cloud resources themselves. Crossplane graduated in November 2025, which tells you declaring and reconciling infrastructure — databases, queues, the lot — is no longer experimental. The tofu-controller covers the OpenTofu side of the house.
- Helm 4. Arrived November 2025 with server-side apply, which tidies one of the messier corners of chart upgrades. Charts remain the packaging most teams reach for first.
- Progressive delivery. Flagger and Argo Rollouts handle canary and staged releases on top of reconciliation — the difference between a change being applied and a change being proven.
- Promotion. Flux’s image automation keeps image references current, and Kargo has emerged as the dedicated promoter for moving tested artifacts between environments.
- Secrets. SOPS with age keys, Sealed Secrets, and the External Secrets Operator remain the three answers you will actually encounter in production.
The pattern across all five: consolidation. Categories that were three startups deep two years ago have settled onto one or two credible defaults each. If you have been waiting for the market to shake out before standardising, it has. That settles the “which tool” anxiety and moves the real work to where it always lived — how your team writes, reviews and promotes changes, which no version bump does for you.
What Still Has Not Changed
The tooling got dramatically better and the fundamentals did not move, which is the surest sign of a maturing practice. A merge is still the door through which changes pass. Reviewing those merges is still where your team’s time goes — more on that in a moment. And an agent that reconciles automatically still needs an operator who knows how to pause it when an incident responder needs it to stop undoing things mid-firefight. Tools answer “how”; the operating model still answers “who decides”. Anyone who remembers the earlier waves of this technology will recognise the rhythm: releases arrive, interfaces improve, and the discipline underneath stays exactly as demanding as it ever was.
The Costs Nobody Puts on the Homepage
Two running costs deserve honest billing. The first is compute: rendering is not free, because every Helm chart or Kustomize overlay means a fresh fork inside Argo CD’s repo-server — which is why big installations throttle renders and spread the work across sharded controllers. High availability wants three or more nodes before it deserves the name. Flux’s appetite scales with objects multiplied by interval; a common production pattern checks sources every minute but hunts drift only hourly. The second cost is human: every change waits for a reviewer, and that queue is the true throughput limit of the whole model. Neither cost is a reason to avoid the practice. Both are reasons to size it honestly, the way you would size fuel rather than hope.
GitOps Going Into 2027: Three Threads to Watch
Three threads seem likely to thicken. Pre-rendered sources should mainstream — reviewing finished output instead of templates is one of those changes that seems small until you live with it, after which going back feels unthinkable. Managed offerings will keep absorbing installation work, shrinking the “run the control plane yourself” constituency further. And promotion tooling will keep consolidating around artifact digests rather than loose tags, because promoting “the exact bytes we tested” is the only version of promotion that survives a serious audit.
Meanwhile the operating habits matter more than any version number. The teams getting the most from GitOps in late 2026 are not running exotic tooling; they are running ordinary tooling with unusual discipline — small reviewed changes, pinned versions, deliberate drift policies, and a rehearsed way to pause the automation. That was true in 2023 and it will be true after whatever ships next year. Treat the release notes as weather and the discipline as climate, and the whole ecosystem becomes much easier to follow.
Why We Say This on an Aviation Blog
Because our industry runs on exactly this bargain: heavy process on the ground so that the vehicle can behave simply in the air. The business side of charter flying teaches the same lesson from another angle — the teams who thrive are the ones with disciplined operations behind a simple customer promise, whether that is how empty leg jet flights work or the factors behind what actually drives private jet charter pricing. Under every smooth experience sits unglamorous machinery, checked constantly, and nobody photographs it for the brochure.
GitOps is that machinery for software. The 2026 vintage is good — mature flagships, vendor-managed installs, a settled supporting cast — and the operators who invested in the discipline rather than the brand names are the ones finding upgrades boring this year. Ask them whether they miss the excitement of earlier years and watch the answer. In this line of work, boring is the compliment that matters.
