Skip to content

Three Ways to Run a Device Fleet: An EndpointOps Comparison

Published November 15, 2026 by Beena Yadav

People who move into DevOps from support or system administration often inherit a fleet of laptops without anyone describing how it is run. When they ask, they get three different answers depending on who they ask. The label that covers all of it is EndpointOps, an informal term for the whole working life of user hardware, buying through disposal. This comparison sets three common ways of working side by side, then compares two ways of treating personal phones, so you can see what each choice costs and which one fits the team you are joining.

Approach one: console only

The simplest operating model is to trust the management console. If a device shows there, it counts; if it is green, it is fine. It is cheap, it needs no extra tooling and it makes a pleasant dashboard. The trouble is structural. A console can only report on what has been registered with it, and it cannot flag machines that never arrived. In the lab exercise described by the Ops Atlas page on EndpointOps, a console claimed that about 89 percent of its devices were compliant, while a closer check found that only 40 percent of the actual estate was fully managed, compliant, encrypted and protected.

Best for: a team of a handful of laptops, all bought the same week. Weakness: blind to anything outside its own list.

Approach two: console plus protection agent

The next model adds the security agent’s view. Now two systems each report in, and gaps between them become visible: a laptop that is enrolled but whose agent is silent, or a laptop running the agent that was never enrolled. The Atlas treats these two layers as overlapping halves of one job, device management on one hand and detection-and-response on the other, and it says plainly that the latter does not replace the former. Many mid-sized teams stop here, and it is a real improvement. But the agent’s roster has its own blind spots, and two lists can agree with each other while both miss the same lost machine.

Best for: teams with a security tool already in place and a part-time administrator. Weakness: shared blind spots, and silence that nobody is assigned to chase.

Approach three: three-source reconciliation

This is the model the Atlas recommends once you want honest numbers. You compare the console, the agent’s roster and an independent record such as purchase data or an account directory. A device passes only if it shows up in every list and clears each test: under management, a check-in inside the last month, disk encryption backed by a saved key, and an agent heard from during the past week. It costs an afternoon a week at the start and gets cheaper as scripts take over. In return you obtain figures such as coverage, agent health, and the share of encrypted devices with a saved key (the source works through 1,164 encrypted and 1,131 escrowed in a fleet of 1,200, or 94.3 percent).

Best for: anyone whose access rules depend on device state. Weakness: it needs an owner, and it will produce uncomfortable numbers at first.

Side by side

  • Effort to start: console only is the lowest; reconciliation is the highest, but it is a one-off cost that declines.
  • Visibility of missing devices: none, partial, and full respectively.
  • Accuracy of the headline number: optimistic, better, and closest to truth.
  • Usefulness in an incident: the third model lets you say within minutes which machines are exposed, while the first leaves you counting by hand.

A second choice: how to handle personal phones

The other decision you will meet is whether personal phones may reach company data, and how. Two models dominate. In the first, the phone is fully enrolled, which gives the organisation wide control. In the second, the phone is left alone, but the work apps are protected, so company data lives in a managed container. Among the Atlas’s anti-patterns is insisting on full enrolment for staff-owned handsets, because people resist it and quietly route around it. It also notes that an enrolled phone is not an open window: the Intune service does not see email or text content, while Apple’s enrolment routes cannot open a user’s private inbox.

The trade-off: full enrolment gives stronger guarantees and a heavier privacy footprint; app-level protection gives lighter guarantees and a smaller footprint. For a contractor-heavy team, the lighter model is generally easier to defend.

Audit first or enforce first?

One more comparison worth making is about application control. You can switch on allow-lists straight away, or run them in audit mode and study what they would have blocked. The source recommends the second. A 2023 advisory from US cyber agencies about attackers using portable remote-management tools is its stated reason for taking this area seriously, and its stated reason for caution is that enforcing without an audit period breaks legitimate software. Auditing costs a few weeks. Enforcing blind can cost a morning of angry tickets.

How to choose for your own team

Start from the question your organisation needs answered. If it is only “do we know roughly how many laptops we own?”, approach one is enough. If you must show that unmanaged machines cannot reach production, you need approach three. A staged route exists: begin with the console, add the agent roster in the first month, and bring in the third list by the end of the quarter. The ninety-day plan in the Atlas follows a similar shape.

What this means for your career

Whichever model you inherit, being the person who can count devices honestly is a visible skill. It pairs well with the broader path described in the outline of how a junior engineer grows into a senior one in DevOps. If you are arriving from a help-desk role, the guide for sysadmins and support staff moving into DevOps will show how this kind of operational hygiene counts as relevant experience.

The verdict

No approach wins on every axis. Console-only is quick and fragile, console plus agent is balanced, and reconciliation is the most truthful. Choose the lightest model that can still answer the questions you will be asked, and plan to move up one level when those questions get harder.