This is an opinion piece. It sets out the view of NerdBible's editors; the facts it relies on are sourced below.
Our position is simple. When a workflow tool sells AI by the credit, the pricing page should say what a typical task costs in credits, and it should say who pays when a shared agent runs. An allowance of 1,500 or 500 credits is a number without a unit that buyers can use. Notion's rollout this month is a timely example, and we want to be fair to it: its documentation is more open than many. The gap that remains is the one buyers most need filled.
What changed
On October 19, Notion announced upgraded agents, a smarter model router called Auto, and a new credit system that brings AI usage under one scheme. According to Notion's help centre, new Business and Enterprise workspaces moved to it on October 19, and existing workspaces move on November 2.
Each member gets two monthly pools. Standard credits, 1,500 per member, cover work where Notion picks the model. Select credits, 500 per member, cover work where you pick the model yourself, or run a Custom Agent you set up. The help page says the pools are separate, cannot be combined or moved between members, reset monthly and do not roll over. Once a pool runs out, the affected work pauses until the reset, unless an admin has allowed the workspace's purchased Notion credits to take over. Notion also notes that the included amount may vary by plan and purchase arrangement.
The unit problem
The pages we read give the allowances in credits. In the text, they do not say what a normal task costs in credits. They do say that Auto delivered similar results to frontier models at about 90% lower average cost across 20 or more common use cases, and that credits therefore go about ten times further. That is a relative claim. A relative saving tells you nothing about the starting price. If you do not know what a task cost before, a 90% cut leaves you unable to size the monthly pool.
Notion's own explanation shows why the number matters. In explaining why it kept two pools, the company says AI tasks consume credits at very different rates, and that a few complex tasks or frequently running agents could exhaust a shared allowance. We agree with that. But it is exactly the reason a buyer needs a range. A pool that can be drained at very different speeds is a pool you cannot budget against without examples.
Who pays when an agent is shared
The second gap is about people, not models. Notion's Custom Agents page says a Custom Agent uses its owner's Select credits by default, even when other people use it. The owner can set a spending cap per agent, and an admin can instead fund an agent with a monthly budget of Notion credits. If the owner's credits run out and no workspace credits are available, the agent pauses.
This is documented, and the controls are real. But think about how a small team behaves. One enthusiastic person builds a useful agent, shares it, and colleagues start using it. The cost lands on that one person's 500 credits. Nothing about the headline allowance, which reads as a per-seat benefit, hints at that. A pricing page that said "a shared agent draws on its builder's allowance unless an admin funds it" in one line would save small teams a surprise.
The strongest counter-argument
The fair objection is that a fixed credits-per-task figure would mislead. Models change, tasks vary wildly, and a vendor that publishes an average invites complaints whenever one task costs ten times it. Notion also says that, on its estimates, the vast majority of users are unlikely to run out of credits, and that admins can see usage, set limits for members and agents, and buy more. The announcement says members can see how many included credits they have used and what each of their agents spends. Seen that way, credits are a flexible metering system, and live usage dashboards are a better answer than a static table.
We accept most of that, and it is why we are not asking for a promise. We are asking for a range with a label. Show what a short summary, a long document analysis and an agent run of a stated length typically cost on the default model. Say plainly that these are examples, and date them. Dashboards help after you have committed. They do not help you choose a plan. The people who need the number most are small businesses deciding between tiers before they have any usage history, and for them the "most users will be fine" estimate is a vendor's average, not their own workload.
This is not only a Notion issue
Credits are becoming the standard way to price AI inside workflow software. One trade report on monday.com's move to a seat-plus-credits model, from SaaSRise, describes credit pricing with a stated dollar rate per credit, $0.01 on annual billing and $0.0125 monthly, and bundles of 1,000 to 3,000 credits. We have not checked that figure against monday.com's own page, so treat it as reported rather than confirmed. Even so, it illustrates a useful practice: a price per credit lets a buyer compare vendors on cost. It still does not say how many credits a task burns, so the same question remains.
What we think vendors should do
- State credits per typical task, as a range, for two or three named task types, with a date.
- Say which pool a shared or scheduled agent draws from, and who pays when it runs.
- Say in the plan summary, not only the help centre, that included credits reset and do not roll over.
- Show what happens at the limit: pause, overage, or switch to a cheaper mode.
What buyers can do now
Existing Notion customers move over on November 2. Notion says admins will get an email with the details, and that usage can be tracked under Settings, Notion AI, Usage. Admins on any workflow tool should check which agents exist, who owns them, and what caps are set before the new system starts. If a vendor cannot tell you what a typical task costs in credits, ask for it in writing and judge the plan on that answer. Credits can be a fair way to meter costs that really do vary. They are only fair to the buyer if the unit is explained.
