Your software plans now include AI agents. How do you avoid paying for the same work twice?

Where agent capacity now sits in plans you already pay for

Agent capacity now arrives inside plans many companies already hold, so the first cost question is what you have, not what to buy. Between April and September 2026, four large vendors added agents to existing plans or changed how agent work is priced and governed.

• Salesforce announced new Core, Advanced and Max editions on 3 September 2026. They include 500,000, 1 million and 2.75 million Flex Credits respectively, the unit Salesforce uses to meter agent work.¹
• OpenAI said at DevDay on 29 September 2026 that the first dot, its always-on agent, comes with Pro and Business Premium plans at no extra cost, with an allowance for deeper work. More dots and more output per dot will be sold later, at prices not yet announced, and the launch covers eligible markets only.²
• Microsoft made Agent 365, its layer for registering and governing agents, generally available on 1 May 2026. It comes in Microsoft 365 E7 or standalone at $15 per user per month.³
• HubSpot moved its Customer Agent to $0.50 per resolved conversation from 14 April 2026, down from $1.00 per conversation, and charges $1.00 per lead its Prospecting Agent recommends for outreach.⁴

Each change makes sense on its own. The problem is the sum. A sales and service team on all four platforms could have three vendors' agents able to draft the same follow-up email, each metered in a different unit. Availability also varies by country, so a multinational may hold different entitlements in different markets.

Our summary of the vendor announcements cited above, read on 7 October 2026. Check your own order forms, which can differ.

Where does the double payment happen?

Double payment happens in three places: two agents doing one job, included capacity left idle while you pay for the same work elsewhere, and governance priced per user on top of agent fees.

Two agents, one job

Take follow-up after a sales call. The CRM agent can draft it from the opportunity record, a general assistant can draft it from the meeting notes, and the office suite can draft it in the email client. Each one consumes an allowance or bills a task. The rep sends one draft, and the work behind the others is paid for and thrown away.

Allowances you do not use

An included allowance is prepaid. If the team does that work in another tool, you pay twice: once for the idle allowance and again for the tool they prefer. Whether unused credits roll over is a contract term, and the announcements do not settle it, so check before the next renewal.

Governance on top

Governance priced per user follows headcount, not the number of agents doing work. Agent counts can mislead too: Microsoft said on 29 July 2026 that nearly 40 million agents were registered on its platform, a count of registered agents, not active ones.⁵ Our note on registered and active agents explains the gap. Count both your users and your working agents before you price a governance layer.

A four-step method to give each job one agent

Inventory what your plans include, map each workflow to one owning agent, price the next task on each platform, and only then decide where a custom agent earns its cost. Two people and a spreadsheet can do the first pass in a week.

One job, one agent: the four steps

1. Inventory entitlements

   Every plan, its included agent, allowance, unit, renewal date and owner.

2. Map workflows to one owner

   The agent on the system of record owns the job; switch it off elsewhere.

3. Price the next task

   The cost once the allowance runs out, plus your people's exception time.

4. Decide on custom builds

   Build only where no bundled agent covers the whole job and its controls.

Inventory entitlements by system

List every plan that now includes agent capacity: the included agent, the allowance, the unit it is measured in, the renewal date and the person who owns the contract. Pull the figures from order forms and admin consoles, not from launch pages, because negotiated terms differ.

Map each workflow to one owning agent

Give each workflow to the agent that sits on its system of record. Case follow-up belongs to the CRM agent, because the case lives there. Document drafting belongs to the office suite. Then write down where the same job is switched off, or the overlap returns at the next product update. Our scoring model for a first agent helps rank the workflows worth mapping first.

Price the next task, not the first

An included allowance makes the first tasks look free. What matters is the cost of the next finished task once the allowance runs out, plus your people's time on the items the agent hands back. Our comparison of per-resolution, per-seat and per-credit pricing shows how to put different units on one scale.

Running costs also include monitoring, prompt changes and exception handling. Our breakdown of monthly agent running costs lists the lines a vendor quote leaves out.

Decide where a custom agent still earns its cost

Build a custom agent only where no bundled agent covers the whole job with the controls you need. Three cases come up most often: a workflow that crosses systems, such as matching an ERP invoice against an email thread; a job that needs an approval step the vendor's agent lacks; and data the bundled agent cannot read.

A disclosure: Sigzen AI builds custom agents, so we have an interest in this step. Our view is still that bundled agents come first wherever they cover the job, because you already pay for them and the vendor maintains them.

What to ask each vendor before you renew

Ask each vendor what its included agents cost once the allowance runs out, what happens to unused capacity, and how its agents can be switched off for jobs another platform owns.

Large buyers already ask these questions. In KPMG's survey of 314 US leaders at firms with $1bn or more in revenue (24 July–25 August 2026), 74% included cost reviews in their AI approvals.⁶ A mid-sized company can run the same review with a shorter list.

• What exactly is included, in which unit, and does the allowance reset monthly or annually?
• Do unused credits roll over, and are they pooled across teams or tied to users?
• What is the price per unit beyond the allowance, and can it change during the term?
• Can the agent be disabled for one workflow while it runs others?
• Does it log each action with the user, the data touched and the cost, so you can audit overlap?
• Which countries does the plan cover, and does availability differ by market?

Our view: put the answers in the same inventory sheet as the entitlements, and review it each quarter. The announcements above show how often agent terms changed in 2026, and a sheet that is a quarter old is already out of date. If the mapping shows a gap no bundled agent covers, our AI automation page sets out how we scope a single workflow. At our published price, the AI readiness assessment is from $6,000 (₹2.5L) over three weeks and starts with a process inventory, where this mapping fits.