Techbanq

Cloud · platform decision

Run Kubernetes when the team has outgrown everything simpler

The orchestrator genuinely delivers scheduling, self-healing and portability — and genuinely charges for them in upgrades, security patching and engineers who understand its failure modes. This guide climbs the ladder of simpler alternatives first, then gives the criteria by which a team and a workload earn the cluster.

Enterprise delivery credentials

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

The ladder

Four rungs between a PaaS and a cluster of your own

Each rung hands your team more control and more operational load. The mistake is not choosing any particular rung; it is skipping to the top one by reflex.

OptionWhat you hand overWhat you keep control ofThe operational load on your teamWhere it stops fittingBest for
A platform as a serviceThe application code and little else — build, run and scale are the platform’s jobYour code and its configuration; little of the runtime beneath itClose to none beyond the application itselfUnusual runtimes, strict data-placement rules, or its pricing at scaleSmall teams shipping standard web workloads fast
Containers as a serviceA container image; the provider schedules and scales itThe image and its resources; not the machines or the schedulerLow — there is no cluster to upgrade or patchService meshes, custom scheduling, operators and cluster-level controlTeams that want containers without owning an orchestrator
Virtual machines done wellNothing — you run machines, with images, automation and monitoring of your ownEverything, which is both the point and the burdenReal, but built from familiar, widely held skillsLarge fleets of many services deployed many times a daySteady workloads and teams with strong operations habits
KubernetesThe scheduling of your containers, to a control plane you operate or rentNearly everything, expressed as declarative configurationThe heaviest on this ladder: upgrades, networking, policy, and a thin skills marketIt does not stop fitting; teams stop affording what it asksMany services, frequent deploys, and a staffed platform capability

The trade

What the orchestrator buys, and the currency it charges in

The benefits are real, and it matters to say so precisely. Scheduling packs workloads onto machines better than people do. Self-healing restarts what crashes and reschedules what a dying node was carrying, without a pager going off. Declarative rollouts make deployments repeatable and reviewable. And portability is genuine — the same manifests can run on any provider or on your own hardware — for workloads that stay disciplined about it. None of this is marketing; it is what the platform was built for, and at a certain scale of estate it replaces a full-time job’s worth of manual coordination.

The price is charged in a different currency, which is why it gets underestimated. There are no licence fees; there is a control plane whose upgrade cadence is set by the project, not by your roadmap. There are networking and policy layers with failure modes of their own, which someone on your team has to understand at three in the morning. And there is a skills market that is thin and expensive, so the cluster’s continuity rests on your ability to hire and keep people who can operate it. The cost lands on the team whether or not the benefits are being used: a three-service estate deployed weekly pays the same operational bill as one that redeems every capability daily.

The criteria

Let team size and workload shape cast the deciding vote

The technology comparison ends in a tie often enough. These situations do not — each points at a rung, and honestly naming your own row is most of the decision.

A handful of services and no platform engineer

Every hour on cluster operations is an hour off the product, and on-call for a control plane is a heavy promise for a small rota.

A defensible default: A PaaS or containers-as-a-service, revisited at the next growth stage.

Dozens of services, daily deploys, cluster experience in the team

The scheduling and rollout machinery replaces coordination you are already paying for by hand, and the skills bill is largely paid.

A defensible default: Kubernetes, managed where possible, with the platform staffed as a product.

One steady monolith with predictable traffic

There is little for a scheduler to schedule; the wins available are in automation and monitoring quality, not orchestration.

A defensible default: Virtual machines done well — images, automation, observability — until the architecture changes.

A hard requirement to run the same stack on-premise and in cloud

Portability is the one benefit the simpler rungs cannot offer, so the operational cost is buying something no alternative sells.

A defensible default: Kubernetes chosen for the portability itself, with its operations budgeted from day one.

The shelf life

A platform decision made once is a platform decision going stale

The right answer at ten engineers is often the wrong one at forty. The honest way to hold the decision is with named trigger points for re-running it: the team doubling, the service count crossing into dozens, the first dedicated platform hire, an acquisition arriving with its own estate, an on-premise or data-residency requirement appearing. At each trigger, the same criteria get read again with the new numbers. Re-running the decision is cheap; the migration it might recommend is not — which is precisely the argument for revisiting on a schedule, before pain forces the question in the middle of an incident.

The ladder runs in both directions, and saying so out loud matters. A cluster adopted early that never earned its keep can be stepped back to a simpler rung, and the step back is usually accounting rather than retreat: the workloads keep running, the team stops paying an operational bill nothing was redeeming. What makes either move possible cheaply is discipline kept in the meantime — containerised workloads, configuration outside the code, dependencies declared — because those habits, not the platform underneath them, are what actually transfer.

Our role

What we assess about the platform choice and what stays your call

  1. 01

    The whole ladder, considered

    We assess your workload shape and team capacity against every rung, not against Kubernetes alone — because an assessment that can only conclude in a cluster is a sales call wearing a methodology.

  2. 02

    The chosen rung, built properly

    Where the cluster is earned, we design it, harden it and set its upgrade path so your own team can run it. Where it is not, we build the simpler rung with the same seriousness — a PaaS pipeline or a VM estate is not a lesser deliverable.

  3. 03

    The revisit, scheduled

    The recommendation ships with its reasons in writing and the trigger points at which they should be read again, so the decision ages as a document to re-open rather than a belief to defend.

The risk appetite, the hiring plan and the growth bet are yours; we make sure the platform choice is derived from them, written down with its reasons, and cheap to revisit when they change.

Common questions

What buyers ask

How many engineers do we need before running Kubernetes makes sense?+

There is no magic headcount, but there is an honest test with three parts. Can you staff on-call for the platform separately from the product, so a cluster incident does not pull the same people as an application incident? Can you absorb the upgrade cadence — a control plane and node fleet that must be moved forward on the project’s schedule, not yours? And can you keep at least two people who genuinely understand the cluster’s failure modes, so its continuity does not hang on one resignation? If platform work would be one person’s part-time job, the answer is to rent the platform — a managed containers service or a PaaS — and to spend the difference on the product. Teams that pass the test tend to know it without asking.

If we start on a PaaS, how painful is a later move to Kubernetes?+

Smaller than the fear of it, provided the workloads keep good habits in the meantime. What actually transfers between platforms is discipline: applications packaged as containers, configuration read from the environment rather than baked in, state kept out of the process, health endpoints that report honestly. Workloads built that way move from a PaaS to a cluster as a re-plumbing exercise; workloads that leaned on platform-specific conveniences pay the real migration bill. That is why starting simple is rarely a trap — the cost of moving later is set by the habits, not by the starting platform, and carrying an unearned cluster for years while waiting to need it is usually the more expensive path.

Does managed Kubernetes remove the operational cost of the platform?+

It removes one layer of it — the care and feeding of the control plane — and that layer is worth removing. What remains is everything above it, and that is most of the bill: upgrades still land on your nodes and workloads on the provider’s schedule, networking and ingress are still yours to design and debug, policy and access control are still yours to get right, and the skills to do all of that are still the ones the market prices highest. Managed Kubernetes moves the cluster one rung down the ladder; it does not take it off the ladder. The teams that are disappointed by it expected a PaaS; the teams that are happy with it wanted Kubernetes and were glad to stop babysitting the control plane.

Test the decision against your estate

An assessment that walks your workload and team down the ladder and ends with a named rung, the reasons in writing, and the trigger points for revisiting it.

Start the conversation