AI and monday.com.

What monday AI does natively (October 2026)

monday’s AI comes as monday sidekick (an assistant), monday agents, AI blocks, AI Notetaker, AI workflows and monday vibe, all paid for in AI credits. According to monday’s pricing article (last modified 3 October 2026), customers who joined the work platform on or after 6 May 2026 must buy credits with their seats: at least 1,000 a month on Basic, 2,000 on Standard and 3,000 on Pro.

Typical consumption: sidekick uses about 10 to 150+ credits per message, agents about 10 to 250+ per task (a run can include several tasks), AI blocks 8 per action, and AI Notetaker 120 per meeting hour. Admins can watch usage and set limits under AI governance, with alerts at 80% and 100%. Seat prices on monday’s pricing page are shown as “$” with no currency code and depend on the billing country, so confirm at checkout.

Integration routes and when to use each

Automation 1: a weekly portfolio status people can trust

Inputs

Items from the active project boards (name, status, owner, due date) and updates posted in the last seven days, read through the API with a read-only token.

Deterministic step

Code counts overdue items, status changes this week and items stuck for more than ten days. Numbers never come from the model.

Model step

One paragraph per project, labelled On track, At risk or Off track only from the computed facts, with links to the items behind each claim.

Human approval

Each project manager edits and approves their own paragraph.

Write-back

The approved text is posted as an update on that project’s row in a separate Portfolio board. No status column is changed.

Failure and recovery

If the daily API limit is hit, the run resumes from its saved position the next day, and the summary names any boards it could not read.

This follows the reporting approach in our AI reporting service and the lesson on preparing management reports with AI.

Automation 2: blocker notes when an item turns “Stuck”

1. Subscribe narrowly. Create a change_specific_column_value webhook for the Status column only; the payload carries the new and previous value.
2. Gather context. Fetch the item, its recent updates and its linked items.
3. Draft. The model writes a short blocker note: what is blocked, who probably owns the fix, and the question that would unblock it.
4. Approve. The project manager sees the draft in Teams or Slack and posts, edits or discards it.
5. Reconcile. A nightly query catches status changes whose webhook was lost.

Permissions and approval design

Run custom jobs through a dedicated app with only the scopes it needs; webhook management alone needs webhooks:read and webhooks:write. App tokens get separate read and write complexity budgets of 5M points a minute, while personal tokens share 10M. For MCP, each user authorises individually and sees only their own data, admins control connections under Admin, Permissions, AI Connectors, and monday’s help centre recommends reviewing AI-suggested changes before they apply. Our MCP guide covers putting your own approval layer in front.

Updates often mention clients and staff. Do not let a summary rate individuals: under the Privacy Act 2020, IPP 8 requires reasonable steps to check information is accurate before you use it.

Costs and limits (as listed 19 October 2026)

• Daily API calls: 1,000 on Free, Basic and Standard, 10,000 on Pro and 25,000 on Enterprise, reset at midnight UTC, which is 1 pm in New Zealand during daylight saving. MCP calls count.
• Complexity: a single query can cost at most 5M points.
• Worked example: an AI block run over 400 items every week uses about 12,800 credits a month (400 × 8 × 4), well above Standard’s 2,000 minimum. A custom weekly summary of ten projects at 8,000 input and 400 output tokens each on Claude Sonnet 5.5 costs about US$0.20 a run.

Where it goes wrong

• Short retry window. monday retries a failed webhook once a minute for 30 minutes. Longer outages lose events unless you reconcile.
• Inconsistent labels. “Done”, “Complete” and “Shipped” across boards need a mapping table before anything is counted.
• Column formats. Column values are JSON whose shape depends on the column type; check the type before writing.
• API exhaustion. An MCP agent exploring boards can use the whole Standard daily allowance in a busy morning.

Teams that run delivery in other tools should compare our Asana and Jira guides.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.