If you’ve been letting Notion’s Custom Agents auto-fill your project tracker, draft your weekly updates, and prep your meeting briefs for free since the beta opened, today is the day that changes. Starting Monday, May 4, 2026, every Custom Agent run consumes Notion Credits at $10 per 1,000 credits, with each run estimated at 30 to 60 credits. A workspace running three or four agents on a normal cadence is looking at an extra $20 to $50 a month on top of whatever you already pay for Business or Enterprise.
This is not a “should I learn AI” question anymore. You already use it. The real question is the same one you ask about any tool when its pricing flips: which of these is paying for itself, and which one is just running because you forgot to turn it off?
Here’s the 5-minute audit, structured the way a project manager actually thinks about it.
What actually changed today
Three things flipped at once, per Notion’s pricing help page and the Connex Digital May 2026 explainer:
- Credits are now metered. $10 per 1,000 credits. Each Custom Agent run consumes 30 to 60 credits depending on how many tools it uses, how much it reads, and how much it writes back into your workspace.
- Free and Plus plans don’t get Custom Agents at all. Custom Agents are now Business + Enterprise only. If you’re on Plus, the agents you set up during the trial period stop running.
- You can set caps. Workspace-wide caps and per-agent caps are now in admin settings. This is the safety rail you want on before you go through your audit.
If you’re on Free or Plus, your decision is made for you — the agents stop, and your audit is “do I upgrade to Business for this.” Skip to the section below on whether the upgrade math works.
If you’re on Business or Enterprise, here’s the audit.
The 5-minute “keep or kill” audit
Open Notion, click your workspace name top-left, then Settings → AI & Agents → Custom Agents. You’ll see every agent currently running in your workspace plus their last 30 days of credit consumption.
For each agent, ask four questions in order. Stop at the first “no.”
Question 1 — When was the last time this agent’s output actually changed a decision?
Not “when did it run.” When did its output cause you, or someone on your team, to do something different than you would have done without it?
A meeting-brief agent that you read every Monday and use to reorder your day passes. A weekly-update drafter where you delete 80% of what it wrote and start over fails. A risk-register auto-triage that flagged the one issue your VP brought up at last quarter’s review passes.
If the answer is “I don’t actually read it” or “I read it but ignore it,” kill the agent. Don’t optimize the prompt. Don’t lower the cap. Kill it. You’re paying credits for a thing nobody uses.
Question 2 — How often does it run, and is that the right frequency?
Most agents people set up during the free beta run on autopilot — every workday, or every hour, or on every database update. That was free. It isn’t anymore.
A Monday-morning status-doc drafter that runs once a week is fine. A Slack follow-through agent that runs every time someone updates the project tracker is going to consume 4× to 6× more credits than you think.
For each agent that survived Question 1, change the trigger to the lowest frequency that still makes the output useful. Weekly beats daily for status reports. Monthly beats weekly for client recaps. Your meeting-brief agent should run the morning of meetings, not every morning.
Question 3 — Could this be a Notion AI question, not a Custom Agent?
Custom Agents are the heavy version. They have access to your workspace tools, run on a schedule, and read/write to multiple databases. They cost 30 to 60 credits per run.
A lot of what people set up as Custom Agents during the trial is actually Notion AI inline questions in disguise. “Summarize this page.” “Pull the action items from this meeting note.” “Translate this paragraph.” All of those are Notion AI features that come with your existing seat license — they don’t burn credits.
If your agent’s job is “answer one question about one page on demand,” that’s not an agent. That’s a hotkey. Convert it.
Question 4 — Does the credit-cost math actually justify the time saved?
Now do the per-month math. For each surviving agent:
(runs per month) × (estimated credits per run, conservatively 50) × ($0.01 per credit) = monthly cost
A weekly status-doc drafter (4 runs × 50 credits × $0.01) costs $2/mo. Keep it.
A daily Slack follow-through agent on a 30-person team (30 runs × 60 credits × $0.01) costs $18/mo. That’s two hours of admin work a month. Probably keep it, but tighten the cap.
An every-update autofill agent on a busy project tracker (200 runs × 50 credits × $0.01) costs $100/mo. That is a salary line item now. Either kill it, downgrade it to weekly, or split it into a manual-trigger agent that you fire when you actually need it.
The rule of thumb: if the agent costs more than 30 minutes of your hourly rate per month and isn’t saving you 30+ minutes of focus work, it’s not paying for itself.
The 3 Custom Agents most PMs should keep on
Across PMs we’ve talked to who’ve been in the beta since spring, three agent types have a consistent return on credits. If you’re rebuilding your stack from scratch this week, start here:
1. The Monday status-doc auto-drafter. Pulls from your project tracker, last week’s updates, and assigned tasks. Writes a draft you spend 10 minutes editing instead of an hour writing from scratch. Runs once weekly, Monday at 7 a.m. Cost: roughly $2 to $3 a month. The agent every PM should keep.
2. The Friday client-recap generator. Pulls from your week’s project notes, meeting recaps, and completed tasks. Drafts the client-facing summary you used to write yourself in the last hour of Friday afternoon. Runs once weekly. Cost: roughly $2 to $3 a month per client.
3. The risk-register auto-triage. Reads new entries in your risk-register database, scores them by impact, suggests an owner, flags any that look like duplicates of an existing risk. Runs daily on weekdays only. Cost: roughly $5 to $10 a month depending on how active the register is. The one that earns its credits when something actually breaks.
What we’d kill in most workspaces: the “every database update” auto-fillers, the daily-digest agents nobody reads, the “translate this Slack message” agents (use Notion AI inline instead), and any agent set up during the trial that you can’t immediately name what it does.
What this means for you, by role
If you’re a project manager at a 10 to 50 person company: Run the 4-question audit on every agent in your workspace this week. Set a workspace-wide cap at $30 a month while you’re testing — that’s an honest “I’m using these enough to keep them” signal without runaway risk. Move all agents to weekly cadence unless you can name a reason to run them more often.
If you’re an admin or HR coordinator running Notion for a recruiting team: The agents that pay for themselves are the candidate-tracker autofill (when people forward you a resume) and the weekly hiring-status digest. The agents that don’t are the ones that try to draft outreach emails — those are better as Notion AI inline prompts you fire on demand.
If you’re a freelance virtual assistant managing Notion for clients: This is now a line item on your monthly invoice, not silent overhead. If a client’s workspace is going to spend $40/mo in credits and your retainer is $800, that’s 5%. Surface it once, name it on the invoice, don’t apologize for it. Per-client credit budgets and a line item beat a surprise question in month three.
If you’re a fractional COO or operations consultant: The client conversation this week is “we’re paying for these now — here are the three I want to keep, here’s the cap, here’s the kill list.” Bring the spreadsheet, not the slide deck.
If you’re a recruiter or talent ops lead running 4 agents: The 30-day audit is your tool. Most teams find one of the four is doing 60% of the work and another is consuming 40% of the credits with no measurable output. Kill that one tonight.
What this audit can’t fix
A few honest limits before you run it:
- You can’t recover the Plus plan. If you’re on Plus, Custom Agents are gone regardless. The decision is “is upgrading to Business for $20/seat/mo worth the 3-4 agents I had running.” For a 5-person team that’s $100/mo seat cost + ~$25/mo credit cost = $125. Check if the agents are worth that before upgrading.
- The 30-60 credit estimate is a range, not a guarantee. Agents that touch many databases, do long writes, or use external connectors (Slack, Calendar) will land closer to 60 than 30. Watch the first month carefully.
- “Coming soon” cross-workspace queries are still coming soon. If you’ve been waiting for portfolio-level agents that can pull data across Workspaces, today’s pricing change doesn’t unlock that. It’s still on the roadmap.
- Caps don’t catch everything. A workspace cap stops new runs once you hit the limit, but in-flight runs that day finish. Set the cap conservatively and tighten over time.
- The “should I pay” question isn’t the same as “are agents worth using.” They’re worth using. Most PMs will end up paying $10 to $30 a month and getting back several hours of focus work. The audit is about which agents, not whether agents.
The bottom line
The Custom Agents pricing flip is the cleanest “should I keep using this” decision a project manager will make this quarter. Run the 4-question audit, kill the agents that fail Question 1, set the workspace cap, move the survivors to the lowest sensible cadence, and check the credit dashboard once a week for the next month.
If you want the longer version — building agents that justify their credits, prompt patterns that actually save you a working hour, and the cadence playbook for a PM running Notion as their daily driver — our AI for Product Managers course walks through it lesson by lesson. Free to start, Pro for the full path.
