Notion's agent stack is live, not slideware, and it only changes one of your decisions

Notion shipped three distinct things in 2026 that now get talked about as a single product: Custom Agents in February, a developer platform built on a primitive called Workers on May 13, and release 3.6 with External Agents on July 1 ^[2][5][7]. For an operator who already runs a dedicated automation platform, only one of those three changes where a process should live.

Start with the sequencing, because vendor blogs blur "announced" and "available" until they sound like the same word ^[16]. Custom Agents arrived first, aimed at repetitive knowledge-work tasks inside a workspace ^[2]. According to Notion's own product messaging and reporting on the launch, customers built more than a million of them in the months that followed ^[3]. That number is a measure of curiosity, not capability: most of that first wave could not connect to external data sources and could not run custom logic ^[4]. An agent that can only read what is already in your workspace is a writing assistant with a task list.

The gap closed roughly three months later ^[17]. Workers is a hosted, sandboxed runtime where a team deploys custom code that syncs data, exposes new tools to agents, and responds to webhook triggers, without standing up a server ^[5]. Database Sync, built on Workers, pulls live records out of Salesforce, Zendesk, or a Postgres database into a Notion database ^[6]. That is the release that matters architecturally, because it is the first one that puts operational data under the agent rather than beside it.

Release 3.6, about five months after the February launch, is broader but shallower in consequence ^[18]. It added External Agents, so agents from Claude, Cursor, Codex, and Decagon can be assigned tasks on a shared board, mentioned like a teammate, and watched executing inside the same workspace ^[7]. The same release brought interactive HTML blocks agents can build inside documents, speaker-labeled meeting notes, native reading and writing of Microsoft Office files, and direct connections to Outlook mail and calendar ^[8]. Model choice is exposed too: higher-reasoning options for hard work, cheaper and faster ones for routine tasks ^[9]. All of it is live, and free to try on Business and Enterprise plans while Workers remains in beta ^[10].

The distinction worth holding onto is about attendance. An automation platform's job is moving data between systems on a schedule or a trigger, with retries, error handling, and observability built for unattended operation ^[11]. A knowledge-work orchestration layer's job is giving people and agents a shared surface to see state, decide, and act on data that has already been assembled ^[12]. The dev.to analysis of the rollout argues Notion is now attempting both and does the second considerably better than the first ^[13], and the use cases it finds durable are consistent with that: cross-functional visibility, judgment-in-the-loop tasks, and light integration without engineering time ^[14]. Founders keep asking whether Notion, Cursor, or Linear replaces n8n or a CRM implementation; the same analysis answers partly, and only for certain jobs ^[15].

Two things to watch. First, Workers leaving beta, because free-during-beta is a pricing decision deferred, not made ^[10]. Second, whether anyone publishes failure behaviour for Database Sync, since a sync you cannot observe is a sync you will eventually discover through a wrong number in a review meeting ^[11][6].