New AI agents now run renewals end to end, from the first read to the e signature. See how it works →
Cluster 4 · Agentless discovery

Agentless discovery: licensing run by AI, with no human operator.

Two small read only scanners, or one agentless collector, report what every machine has installed. From then on AI names each title, holds the installs against what you pay for and ranks what to fix, on the schedule you set. Nothing is installed that changes a machine, nothing is written to a tenant, and every change stays a person's press.

FFFredrik Filipsson, Cofounder1 November 20269 min read
2 scannersOne for Windows, one for Linux. Both only read.
1 collectorOn one machine. Nothing stays on the machines it reads.
179 publishersCovered by the scan, each with the metric its license is sold on.
No operatorMachines report on schedule. AI names, counts and ranks.

A license position built by hand, for an audit letter or a renewal, starts to age the day it is finished. Agentless discovery keeps it current instead: the machines report on a schedule, and AI does the counting that nobody on the team has time to do.

This guide covers the apps in Vera AI by VendorBenchmark that do licensing with no human operator: what each one reads, what it finds, and the decisions that always come back to you. Each section links to a full article. One word first. Agentless refers to your machines: there is no agent to install and maintain on them. The agents in this guide are AI, and they work in your workspace, not on your estate.

The apps that do the licensing

Four articles cover the machines, from the scanners to the weekly report. Two cover the subscriptions no scanner can see, Microsoft 365 and Salesforce, read through a read only connection or the exports you already pull.

How the licensing runs

You set it up once. After that the machines report on their schedule and the agents carry the work until a decision needs a person.

1

You roll it out once

The scanners through Intune, Configuration Manager, Group Policy, Ansible, cron or SSH, or the collector on one machine.

2

Machines report

Read only, every six hours, daily or weekly. The collector reads its whole list once a day.

3

Named and counted

Every install named, then held against the lines you pay for, with the money a year.

4

Ranked and sent

The biggest money and the audit risks first, and a report per publisher by email.

5

You decide

Removals, reclaims and renewal cuts are a person's press. Nothing changes on its own.

What the agents do, and what stays yours

No human operator does not mean no human. It means nobody has to collect the data, match it to the contracts or count the gaps. The table shows where the line sits.

StageThe agents do on their ownWaits for you
CollectMachines report on schedule; the collector reads every machine on its list once a day and names the ones it could not reachRolling out the scanners or the collector once; an owner creates the scanner key
NameEach install named by the catalog, a pattern or the component the scanner read, and kept named on every later scanA title nothing recognizes waits for one press of Name it
CountPaid against installed for every line: over, under or matched, with the money a yearAdding the lines you pay for, and how a per person line counts
RankEvery finding in one queue, biggest money first, the audit risks priced at listMarking a finding handled or a saving captured
ActRemoval and reclaim lists, a PowerShell script, the renewal sheet and drafted emailsRunning the change, sending the email and deciding the renewal
app.vendorbenchmark.com/discovery
Software Discovery Overview with the software map, where the money is and an Ask bar, on the sample estate

The estate on one screen. Software Discovery's Overview on the sample estate: 59 products on 590 machines as a map sized by machines; where the money is, as audit exposure, paid and not used, and paid with no contract on file; and an Ask bar that answers from the machines that reported.

Why the position has to run itself

License exposure rarely sits where a spreadsheet looks for it. Oracle prices Java per employee, so one install that needs a license puts the whole company in scope. On VMware, Oracle counts the cores of every host a database could move to, not the ones it runs on. A paid seat on a laptop nobody has signed in to for months costs the same as one in daily use.

A position built for an audit letter is a snapshot, and the estate keeps moving. A Java auto updater adds a new install. A team installs a tool nobody bought. A renewal arrives before anyone counted the true down. When the machines report on a schedule and every install is named and held against what you pay for, those changes show up in the next run instead of the next audit.

That is the case for agents here. Not cleverness: consistency. The run that would have been skipped in a busy quarter happens anyway, and the findings wait in one queue, biggest money first, for the person who decides.

The machines report. The agents count. The decisions stay with you.

The subscriptions no scanner can see

A scanner sees what is installed on a machine. It cannot see who holds a Microsoft 365 license or a Salesforce seat. For those, the agents read the system itself, read only.

  • Microsoft 365. Microsoft 365 Optimizer reads the tenant through Microsoft Graph, or the admin center exports, and runs seven rules on every user.
  • Salesforce. Salesforce Optimizer reads the org through a read only connected app, finds the seats to take back by seven rules of its own and sizes Data Cloud and Flex Credits to use.
  • Every other SaaS app. SaaS Manager joins sign ins, bills and contracts into one list of apps; it is covered in the spend guide.

None of them writes anything back. The Microsoft 365 PowerShell script is a list of license changes for your admin to read and run, and in Salesforce an Approve marks the seat on the renewal sheet; nothing changes until an admin makes the change there.

app.vendorbenchmark.com/m365
Microsoft 365 Optimizer Results on a sample tenant, with the spend to plan chart and the findings ranked by savings

A tenant, read only. Microsoft 365 Optimizer's Results on a sample tenant: what it pays a year, the savings walked finding by finding from today's spend to the plan, and seven kinds of finding ranked by savings, with the PowerShell script and Approve the plan at the top.

When you want a checked position instead

Agentless discovery states what the scan can see and says what it cannot. For every major publisher, Software Discovery+ lists what a true position needs beyond the scan: cores and clusters, the agreement and the license statement, named users and the vendor's own consoles, each with where it comes from and why the position is wrong without it.

Some moments need that gap closed by people: an audit letter, a true up, a renewal you are about to sign, a deal. For those, Same Day ELP takes your paper and your deployment data. AI drafts the position, an analyst reviews every line and a second analyst checks it, delivered within 24 hours of a complete submission. It covers 41 products and is part of the Enterprise plan.

Two paths, one estateAgentless discovery for a position that runs itself every week. Same Day ELP before an audit, a renewal or a deal, checked by two analysts. Which path fits sets them side by side.

What to do first

  1. Load the sample estate. In Software Discovery, Get started, See it working first: 590 machines across six sites, with every step filled.
  2. Scan one machine. An owner creates the scanner key. Download the scanner for Windows or Linux, run the one line with the key in it and press Listen.
  3. Roll out, or install the collector. Copy the line for the tool you already use, or put the collector on one machine in Software Discovery+ and let it read the rest.
  4. Connect the subscriptions, then choose the reports. Microsoft 365 and Salesforce through read only connections; then pick the publishers to manage and press Send me the reports.

Questions buyers ask

What does agentless mean here?

Nothing stays installed on the machines you scan. With the agentless collector, one machine reads the others over WinRM or SSH; on each, a read only scanner runs once from a temporary folder, and the folder is deleted. With the scanners, your distribution tool runs a small read only script on a schedule, so there is no agent to install and maintain.

Does anything change on our machines?

No. The scanners only read and need no administrator rights. They never read file contents, user names, email or IP addresses, and the machine identifier and serial number are sent only as hashes.

Who runs it once it is set up?

Nobody has to. Machines report on the schedule you set, every six hours, daily or weekly, AI names each title and builds the position, and a workspace owner can have one licensing report per publisher mailed weekly or monthly. Removals, reclaims and renewal cuts stay a person's decision.

When should we use Same Day ELP instead?

When the position has to stand up in front of a vendor: an audit letter, a true up, a renewal or a deal. AI drafts it, an analyst reviews every line and a second analyst checks it, delivered within 24 hours of a complete submission.

Agentless discovery

Let the machines report. Keep the decisions.

Software Discovery, Java Scanner and both optimizers carry sample data in the Free Trial, so you see every screen before a single scanner runs.