Dots versus Muse: a dev.to framework for testing paid and free consumer AI agents

What shipped

According to a post on dev.to's community feed, OpenAI has put Dots, an agent tier described as continuously available, behind its Pro plan at roughly $100 a month, while Meta has released Muse, a free assistant built into WhatsApp, Messenger and Instagram. The author treats this as the opening round of a genuine pricing conflict in consumer AI agents, with reviewers already lining up to declare which product is smarter.

That comparison misses the point for small businesses, the post argues. Rent gets paid by an agent that finishes tedious work overnight and unsupervised, not by one that writes elegant prose — and not by one that quietly hands a customer list to a data model the owner doesn't understand.

A three-job comparison

Rather than coding benchmarks or trivia, the dev.to piece proposes giving both agents the same three routine tasks that pile up in any business: untangling two weeks of calendar chaos, from double bookings to recurring meetings nobody attends; auditing every recurring charge on a company card and drafting cancellation emails for anything unused or duplicated; and chasing one overdue invoice or refund from start to finish, with scheduled follow-ups, polite escalation and a final report.

Scoring comes down to three columns. Actions completed: did the job actually finish, or did the agent produce a flattering summary of work it could have done? Guardrails: what required permission, what was touched without asking, and is there a visible record afterward? Data and access: where did the work run, what was read, and what happens to that context if the subscription lapses?

The author points to one reported Dots behavior as the right kind of feature: it noticed a recurring weekend meeting that nobody attended and offered to resolve it. Unglamorous upkeep, rather than autonomy for its own sake, is what makes an agent worth paying for.

What the price tags signal

The $100-versus-free framing is mostly noise, the post argues, because the two products are not selling the same thing. Muse costs nothing because it lives inside Meta's attention-driven ecosystem across WhatsApp, Messenger and Instagram; the payment is implicit, in where working context resides and how it may be used. For customer-facing busywork in apps people already use, that position is a real advantage rather than a gimmick.

Dots is paid because OpenAI is selling capability as the product itself. The bet is that a stronger tier finishes jobs a free assistant merely summarises. If testing shows Dots completing work Muse can only plan, the subscription earns its keep; if both finish the same jobs, the author suggests, the $100 amounts to a fee for brand enthusiasm.

The outcome nobody wants to name

The piece also raises a scenario it says most coverage avoids: both agents fail most of a specific business's work, because generic assistants don't know which leads are hot, how quotes are produced or which customers get exceptions. They handle assistant-level chores — inbox triage, calendar cleanup, drafting first versions — quite well, but the processes that define a business require context these products simply don't have. That is not a flaw in either product; it is the difference between an assistant and an employee.

The honest conclusion, per the post, is that consumer agents are a $0-to-$100 monthly add-on rather than a strategy. Test the free one first, upgrade only when the paid tier demonstrably finishes more, and keep core processes in a system whose context you actually control.

Running the experiment safely

The suggested safeguards: use disposable test data — a trial calendar or a low-stakes refund — rather than a real customer list; require human review before anything is sent externally until an agent has earned trust; and limit the trial to one week, three jobs, both agents. If neither finishes, the post concludes, that is a real answer about whether the market is ready for your use case — and far cheaper than a year of either subscription.

Why it matters

The split between paid capability from OpenAI and free, ecosystem-distributed assistance from Meta sets the template for how consumer agents will be priced and bundled in the coming cycle. The dev.to argument is a useful corrective for buyers: the sticker price is the smallest cost involved. The real cost is the work and data you hand over, and whether the agent's output is something a business owner would put their name to. Judged on completed actions and visible guardrails rather than benchmark scores, the winner of this pricing war gets decided job by job.