Notion Custom Agents Are Free Until May 3 — How COOs Should Run a 3-Week Evaluation

Most enterprise software trials don't come with a hard deadline. You get a 14-day window that sales extends on request, and the evaluation drags on while everyone waits for a budget conversation to mature.

Notion's Custom Agents trial is different. It ends on May 3, 2026, and when it does, usage switches to a credit-based pricing model. That creates an unusual situation: you have a specific, finite window to run a real evaluation at zero cost, before you have to decide whether to pay.

For COOs, that window is worth using deliberately. Here's how to structure the three weeks you have left.

What Notion Custom Agents Actually Do

Notion announced Custom Agents in its version 3.3 release on February 24, 2026. According to the official Notion blog, these are fully autonomous agents — not copilots or suggestion engines, but AI systems that run 24/7 on a schedule or trigger, without manual prompting after the initial setup.

The core use cases break into three categories:

Q&A agents sit on top of your connected knowledge sources (Notion workspaces, Slack, Mail, Calendar, and MCP-connected tools like Linear, Figma, and HubSpot) and answer recurring questions autonomously. The intended result: your team stops asking the same 15 questions to three different people and instead routes them through an agent that already knows the answer.

Task routing agents capture incoming requests, create tasks in your system, enrich them with relevant context, and route them to the right team members. If you've ever watched a project management workflow break down because someone didn't know which Notion database to use, or which slack channel to post in, or which person owns a category of request — this is the agent type designed for that problem.

Status reporting agents compile regular updates automatically by pulling from connected tools and synthesizing context into recurring reports: daily standups, weekly recaps, monthly OKR summaries. The agent runs on a schedule and delivers a report rather than waiting for a human to pull it together.

What Happens After May 3

Starting May 4, Custom Agents consume Notion Credits — an add-on for Business and Enterprise plans. Seat pricing doesn't change, and other AI features remain included in existing plans. The credit model means you pay only for agent activity, with admin controls over how many credits are purchased.

Notion is positioning this as a volume-based add-on, not a per-agent subscription. That's a reasonable model for evaluating total cost, but it means the cost of a bad agent configuration (one that runs too frequently or on low-value tasks) will show up in your credit burn before you catch it in your metrics. That's part of what the trial period is for.

The 3-Week Evaluation Framework

The goal of this evaluation isn't to "see what agents can do." It's to answer a specific operational question: do Notion Custom Agents save enough time and reduce enough coordination overhead to justify the credit cost in your specific environment?

That requires a structured test, not an open exploration.

Week 1: Set a baseline and pick two workflows.

Don't try to evaluate everything. Pick two workflows — one from each of the following categories:

  1. A knowledge Q&A workflow: something your ops team answers repeatedly because the answer lives in Notion or Slack. Choose one where you can count the number of questions per week (check Slack channel history, or ask the person who fields them most often).

  2. A routing or status workflow: something that requires a human to gather information, format it, and send it somewhere on a schedule — a weekly ops update, a new request intake process, or a recurring project status pull.

For each workflow, document: who currently handles it, how long it takes per occurrence, how often it occurs per week, and what "good" looks like as an output. That's your baseline.

Week 2: Build the agents and run them in parallel.

Use the trial period to build both agents. Notion's setup is genuinely no-code — you describe what you want in plain language and the AI builds the agent configuration. Adjust triggers, data sources, and output format until it looks right, then let it run.

But don't turn off the manual process yet. Run both in parallel for the full week. This is the week where you'll find the gaps: agent outputs that are 80% right but missing the judgment call a human would make, or trigger conditions that fire too broadly. Log every exception.

Week 3: Measure what actually changed.

At the end of week two, you have two weeks of agent output and two weeks of manual baseline to compare. The measurement framework should cover three things:

  • Time saved: actual hours reduced per week across the two workflows. Be honest — include the time spent managing exceptions the agent created.
  • Output quality: compare agent-generated outputs against manually produced ones. For Q&A agents, did the answers require correction? For routing agents, did tasks end up in the right place? For status reports, did the report require editing before it went out?
  • Credit burn rate: check the Notion usage dashboard to understand how many credits the agents consumed over two weeks. Project that to a monthly cost, then compare it to the cost of the manual time it replaced.

What to Avoid During the Evaluation

A few failure modes are common in agent trials:

Don't start with your most complex workflow. Custom Agents work best on high-volume, structured recurring processes. If you pick a workflow that requires nuanced judgment — like a change management decision tree or a multi-stakeholder escalation process — you'll spend the evaluation debugging edge cases instead of measuring value.

Don't count "feels faster" as a metric. Perceived productivity gains are real, but they don't survive budget conversations. You need a number: hours saved per week, errors reduced per month, or tasks routed correctly without human intervention. If you can't measure it, it's not a metric.

Don't evaluate it in isolation from your other work management tools. The relevant question isn't just "does Notion's agent do this well?" It's "does Notion's agent do this better than a Monday.com automation, an Asana rule, or a Slack workflow?" Some of those answers will surprise you.

What to Do This Week (The Kick-Off Checklist)

Time is short. The trial window has been open since February, and with May 3 as the hard stop, you have three weeks of useful evaluation time remaining — if you start now.

This week:

  • Confirm your organization is on a Business or Enterprise Notion plan (required for Custom Agents access)
  • Identify your two evaluation workflows — Q&A and routing/status — using the criteria above
  • Pull two weeks of historical data on each workflow to establish a baseline before agents change the behavior
  • Assign one person as the evaluation lead who will log agent outputs and exceptions daily during week two
  • Set a decision date: May 2, one day before the trial ends, where your evaluation lead presents findings and the team decides whether to purchase credits, expand, or stop

The coordination overhead that kills operational velocity rarely lives in a single system. But it often lives in the gap between systems — the handoff between Notion and Slack, the recurring meeting that exists because status data isn't automatic. Notion's agents are specifically designed to close those gaps. Whether they close them well enough in your environment is the only question worth testing right now.

Custom Agents were announced in Notion's official blog post and the February 24, 2026 release notes. Pricing details available at Notion's help center.