One general AI coworker or a team hired by function?

The slower thinking

The product market now offers both shapes. Lindy presents a broad personal assistant across inbox, calendar, meetings, follow-ups, and connected apps. Microsoft Scout presents a broad desktop worker that can use files, websites, Microsoft 365, schedules, and approval points, although Scout remains in Frontier preview.

Sintra and Marblism package several ready helpers or employees by common business function. Relevance AI gives a technical operator the parts to build and run a custom workforce. Asana AI Teammates place specialized work inside an existing project system.

These products do not fit on one ladder from weak to strong. They move the buyer’s work to different places.

A general coworker keeps the front door simple

One broad worker reduces routing at the start. The owner asks in one place and can change subjects without finding another employee.

That is useful when:

• The work is irregular.
• The owner gives detailed direction each time.
• The same person reviews every result.
• The sources are easy to gather.
• A mistake is easy to reverse.

The general coworker can also discover a future role. If the owner asks for the same weekly channel review every Monday, that repeated request may deserve a defined marketing responsibility later.

The cost is hidden coordination. The owner may still need to say which source is current, which standard applies, what “done” means, which decision needs approval, and when the work should happen again. The product has one front door, but the owner remains the dispatcher.

A specialist makes the review standard stable

A specialist begins with a narrower responsibility. The employee knows which sources belong to the function, what work product is due, where to report it, and which person reviews it.

This is useful when:

• The work repeats on a schedule or after a known business event.
• The result has a stable acceptance checklist.
• The function uses distinct data or systems.
• The reviewer has specific authority.
• Errors have a different consequence from errors in another function.

A podcast producer should not silently become a finance employee because both jobs involve files. A marketing employee should not approve a payroll exception because both jobs involve analysis. The role boundary makes refusals easier to understand.

The plain test for an AI employee describes the five parts of that boundary: responsibility, start condition, work product, reporting surface, and human approval.

A specialist label still proves nothing by itself. A company can put a job title on a chat personality without giving it role-specific evidence, a schedule, or a review standard. The buyer should inspect the actual work record.

A team creates handoffs that must be designed

Several specialists are useful only when their handoffs are clear. A marketing employee can prepare approved launch copy. A commerce employee can publish it. The handoff should identify the approved artifact, the receiving owner, the stop conditions, and the result expected from the next role.

Without that structure, a specialist team creates a new problem. Work can bounce between roles, sources can be copied badly, and the owner can spend more time deciding who should act.

New Street Studios is a narrow public demonstration of separate AI workers handing ordinary commercial artifacts to one another in Slack. One worker designs cards. Another prepares approved art for print. Another runs the shop. The owner opens the work and approves consequential decisions. The project is not evidence that every business should use a team, and it has not removed the human owner.

The field report on one card handoff shows the useful moment: the print worker rejected the design worker’s image because a crop could not repair a hidden focal object. Specialization mattered because the receiving job had a different acceptance standard and could say no.

The honest cost is management design

A general coworker asks the owner to keep more context in her head. A specialist asks the vendor or buyer to define a role. A team asks for role boundaries and handoffs. A builder asks the buyer to design and maintain the workers themselves.

None of these costs disappear because the software is capable.

The small-business owner should choose the form of management work she is willing to keep:

• Direct varied tasks to one broad coworker.
• Review one recurring function through a specialist.
• Manage handoffs across a team by function.
• Design the system through a builder.

The monthly platform comparison separates those product shapes and gives conditional recommendations instead of one winner.

A decision aid

Irregular research and personal tasks: start with one general coworker. The work changes each time, and the owner is already the common reviewer.

One recurring function with a clear weekly result: start with one role-specific employee. The source set, schedule, and acceptance standard can remain stable.

Several functions with different data and owners: use a team hired by function only when the handoffs are explicit. Each role needs its own sources, work product, and approval path.

A technical operator who wants control over the construction: choose a builder. The operator wants to define the workers and accepts the maintenance work. Zapier Agents, n8n AI agents, and Relevance AI are relevant starting points.

A small-business owner who wants the work done: choose a ready role that matches the function. Setup work is not the result she is trying to buy.

One worked week

Imagine one general coworker serving a six-person services company. On Monday, the owner asks it to prepare a campaign plan. On Tuesday, she asks it to inspect an overdue invoice. On Wednesday, she asks it to draft support replies. On Thursday, she asks it to update a contractor compliance file.

The coworker may be capable of each task. The review burden changes each day.

For the campaign, the owner checks positioning, channel data, and budget. For the invoice, finance checks payment records and customer terms. For support, the account owner checks the relationship and promised remedy. For compliance, a safety owner checks the source program and attestation.

If the owner remains the only reviewer, one coworker can still be simpler. She knows the company and routes each decision herself.

If the company already has separate owners, the one-coworker history becomes harder to read. A finance correction can sit beside campaign direction. Permissions can be broader than one function needs. A general instruction can conflict with a specialist standard. The team begins to depend on the owner’s memory of which rule applied to which task.

A function-based team can reduce that ambiguity. It also replaces some human coordination and production work. Fewer people may be hired for first drafts, routine checks, status records, and recurring preparation. Human owners remain for standards, relationships, legal responsibility, regulated attestations, and decisions whose effects are difficult to reverse.

The honest dividing line is not intelligence. It is whether the right person can check the work. Can that person review the source, result, and consequence without reconstructing the worker’s context from unrelated jobs?