Should You Keep or Kill Your Notion Custom Agents? A May 4 Audit for Solo Webflow Practices

Notion stopped letting Custom Agents run for free on May 4, 2026. Every agent run now consumes Notion Credits, priced at 10 dollars per 1,000 credits, with each run estimated at 30 to 60 credits depending on how many tools it touches and how much it reads or writes. For a solo Webflow practice running three or four always-on agents, that is a real monthly line item nobody costed when the agents were free. This piece walks through the five-minute audit I ran on my own agents this week, the math I used to decide which to keep, and the rule I now apply to every agent before letting it run. The discipline is simpler than it sounds. The cost saving is meaningful from week one.

What Did Notion Actually Change on May 4?

Notion 3.3 introduced Custom Agents on February 24, 2026, and gave every Business and Enterprise workspace free access during the trial window. The May 1 release notes flagged that Custom Agents would remain free until May 3, with metering switching on May 4. The pricing model is straightforward. Notion Credits cost 10 dollars per 1,000 credits. Each Custom Agent run consumes 30 to 60 credits depending on how many tools the agent calls, how much workspace context it reads, and how much it writes back into your databases.

The practical implication is that a Custom Agent running on a daily schedule, hitting the high end of the credit range, costs roughly 18 dollars per month just for itself. Three or four agents on the same cadence quickly become a 50 to 70 dollar monthly add to your Business plan, which already costs 20 dollars per seat. The total cost of running automation in Notion changed materially overnight. The instinct to keep all your agents because they were useful when free is the trap. They were useful at zero marginal cost. They are not all useful at 18 dollars per month each.

What Does the Per-Month Cost Formula Actually Look Like?

The minimum viable cost formula has three inputs. The credits per run, the runs per month, and the credit price. Credits per run lands somewhere between 30 and 60 for most agents, with simpler read-only agents at the low end and tool-heavy agents that write back to multiple databases at the high end. Runs per month depends on the schedule, with daily agents at 30 runs per month and weekly agents at 4 runs. Credit price is fixed at 10 dollars per 1,000.

Plugging in the numbers tells you the cost per agent quickly. A daily agent at 60 credits per run costs 18 dollars per month. A weekly agent at 30 credits per run costs roughly 1 dollar 20 cents per month. The 18-fold cost ratio between daily-heavy and weekly-light agents is the single most important number in the audit. Most solo practices have at least one agent that is structured as daily but produces value that is genuinely weekly. Demoting that agent saves 17 dollars per month per agent without losing meaningful output.

Which Agents Did I Keep After Running the Audit?

I went into the audit with five Custom Agents. A daily research drafter that pulls AI tooling news into a workspace database. A weekly client status compiler that builds a snapshot per retainer client from Slack and email. An IT triage agent that watches my inbox for license renewal alerts. A daily competitive intelligence agent that monitors three named competitors. A weekly content brief generator that pulls the next blog topics from my editorial calendar.

Two passed the audit. The weekly client status compiler stays, because the four monthly runs at roughly 50 credits each cost about 2 dollars per month and the output is the foundation of every Friday client update. The weekly content brief generator stays for the same reason. The other three failed and were either killed or downgraded. The daily research drafter became a Notion AI inline question I run manually each morning, which costs nothing extra and forces me to read the news rather than skim a database row. The competitive intelligence agent dropped to weekly. The IT triage agent was killed because email rules in my client already do most of the same job. I covered the upstream cost lens in my monthly AI tooling cost piece.

What Is the Hard Rule I Now Apply to New Agents?

The rule is two questions. Does this agent need to run on a schedule, or could a human triggering it on demand produce the same value? If on-demand works, the agent should be a Notion AI inline question, not a Custom Agent. Notion AI inline questions are included in the workspace plan and do not consume per-run credits the same way agents do. The cost difference between scheduled and inline is the entire economics of the audit.

The second question is whether the agent's output is read within 24 hours of running. If an agent runs daily but you only read its output during the Friday client review, the agent should run weekly to match the read cadence. The mismatched cadence is the most common waste pattern, and the fix is a one-line schedule change that drops the credit cost by 80 percent without changing the value. I covered the broader rhythm in my AI daily workflow changes piece.

How Does This Connect to the Broader AI Cost Picture?

Notion's metering switch is not the only AI cost flip in the market this quarter. Anthropic's Wall Street venture announced May 4 signals that enterprise AI services will keep rising in price as demand outpaces delivery capacity. Cursor's plugin marketplace controls let studios distribute toolchains, but the marketplace itself charges for Required-mode plugin distribution at scale. The combined picture is that the era of free or near-free AI tooling is closing, and 2026 is the year solo practices have to start budgeting for AI the way they budget for hosting or design software.

For a solo Webflow practice, the practical move is to do the same audit I did across every AI tool with metered usage. Notion Custom Agents. Claude API usage on long-running tasks. ChatGPT Team seat usage. Cursor token spend. Each tool's per-month cost should be visible on a single line item, and each line item should pass a value-per-dollar test. Tools that fail the test get killed or demoted. Tools that pass get budgeted into the practice's overhead and priced into client retainers accordingly. I covered the related Anthropic context in my Anthropic Wall Street venture piece.

What Are the Common Mistakes Solo Practices Make in This Audit?

Three patterns kill the audit's value. Estimating credit consumption from memory rather than from the actual usage data Notion exposes in the workspace settings. Running the audit once and never repeating it. Treating all agents as binary keep-or-kill without considering the demote-to-weekly option that captures most of the savings without losing the agent's value entirely.

The fix for the first is to look at the Notion credit usage dashboard before deciding. The dashboard shows actual credits consumed per agent in the trial period, which removes the guesswork. The fix for the second is to schedule the audit quarterly, ideally at the same cadence as the monthly tooling cost review. The fix for the third is to default to demote rather than kill when an agent's value is real but its frequency is wrong. I covered the parallel discipline in my per-client AI memory stack piece.

What Did I Change in My Own Practice This Week Because of This?

Three concrete changes this week. The daily research drafter became a 6 AM IST manual prompt, which forced me to actually read the AI tooling news instead of skimming a generated row. The competitive intelligence agent moved from daily to weekly, which dropped its monthly cost from roughly 18 dollars to 1 dollar 20 cents while keeping the strategic value. The IT triage agent was killed entirely, with its work moved to a Gmail filter I had not bothered to set up because the agent was technically free.

The deeper change was the discipline. Every new agent I create now needs to pass the two-question rule before it goes live. The default cadence for new agents is weekly, not daily, with promotion to daily requiring evidence that the output is actually read daily. The defaults matter more than the rules, because the defaults compound across every future agent I might create. I covered the related operational rhythm in my six AM Bengaluru routine piece.

How Does This Map to Notion AI Inline Questions Versus Custom Agents?

Notion AI inline questions are the cheaper alternative for most solo practice use cases. Inline questions run when you trigger them, consume credits per question rather than per scheduled run, and produce output immediately in context where you read it. Custom Agents are the more expensive alternative built for cases where the output needs to exist independently of when you ask for it, in a database row that other people or other agents can read.

For a solo practice with no team to read agent output, the case for Custom Agents narrows to two scenarios. First, where the agent's output is genuinely time-sensitive and waiting until you ask defeats the purpose, like a weekly competitive monitor that needs to capture content that disappears or changes. Second, where the agent is feeding another tool downstream, like an agent that pushes briefs into a Webflow CMS via the workspace integration. Outside those two scenarios, inline questions are the cleaner choice. I covered the related AI integration discipline in my AI as senior team member framework piece.

What Should I Watch For in Notion's Next Few Releases?

Two trajectories are worth tracking. Notion will probably introduce per-agent credit caps to give workspaces predictable spending without surprise overages. The cap structure will determine whether teams can run aggressive multi-agent setups without budget blowups. Watch for credit-cap controls in workspace settings over the next two quarters.

The second trajectory is whether Custom Agents extend beyond Business and Enterprise plans down to Plus, which would broaden the user base meaningfully. The current restriction to higher tiers makes Custom Agents an enterprise feature in pricing terms even though the work they do is often individual contributor work. If Notion opens Custom Agents on Plus, the audit calculus changes again, and solo practices currently on Plus get a useful new tool. For now, the practice has to be on Business or Enterprise to use them at all, which is itself a budget decision worth running through the same audit.

What Is the One Sentence I Take Away From This?

Free tools have hidden costs the moment they start charging, and the audit catches the cases where the value-per-dollar collapsed without the user noticing. The five-minute audit ran in the gap between client work and lunch on a Tuesday. The result was 35 dollars per month back in the practice's budget and a sharper sense of which automations actually earn their place. For solo Webflow Partners reading this, the recommendation is to run the audit on every AI tool that recently changed pricing, not just Notion. The 2026 AI tooling market is shifting from land-grab pricing to revenue-led pricing, and the studios that catch the shift early are the ones that protect their margins.

Notion's pricing change is one of several this quarter. The pattern is the signal, not the specific tool. The discipline is to keep the practice's monthly AI cost on a single line item, audit it quarterly, and demote or kill anything that fails the value-per-dollar test. The tools that survive the audit are the ones the practice should price into client retainers. The tools that fail are the ones the practice should never have been carrying for free. I covered the upstream proposal-stage discipline in my winning project proposal piece.

If you are running a Webflow practice and want to walk through your own Custom Agents audit this week, drop me a line and tell me which agent is most likely to fail the value-per-dollar test. Let's chat.