Napster sells AI agents now. Not music. Agents — ones with a face and a voice that hold a live conversation with your customers on your site.
It is easy to read that as a brand clinging to a name it no longer earns. Read the actual argument instead, because it is the correct one, and it is the one we have been operating on.
CEO John Acunto describes the shift as going from streaming music to streaming intelligence. The product behind it, NV2, is a real-time conversational video model: an agent that appears on screen, in HD, and talks. Napster's claim is that it runs at roughly a cent a minute — their number, not ours — because much of the work happens closer to the edge rather than entirely in the cloud. The point of the price is the posture it implies. At that cost, a visible agent stops being a flagship feature a company rations and starts being a default.
The framework is capture, learn, act, own
Acunto's public framing runs in four moves, and the order matters.
Capture. A visitor clicking around a website alone generates almost nothing you can use. A visitor who talks tells you what they actually came for, in their own words, including the thing they would never have filled into a form.
Learn. Those conversations are signal. Fed back, they sharpen what the business knows about its own demand.
Act. The agent does something with it — answers, routes, books, sells — rather than filing a transcript nobody reads.
Own. And then the part most coverage skips: the intelligence that comes out of all of it should belong to the business. AI can be the front door to your company. The relationship standing behind that door, and everything you learn through it, is either yours or it quietly becomes your model provider's.
We wrote nearly this exact argument here in July, under a different headline, about a different company's framing. It keeps arriving from new directions because it is true. The moat is what you own, not the model you rent.
Why a face is not a gimmick
The weakest-sounding part of Napster's pitch is the one we would defend hardest.
Acunto's reasoning is that humans are audiovisual animals. We are built to read a face and a voice, and we have spent our whole lives doing it. Typing into a text box and waiting is a learned behavior layered on top of that — workable, but not natural, and it quietly costs you the people who never liked it to begin with.
Which is to say the face is not decoration. It is a lower-friction interface for the majority of people who were never going to enjoy chatting with a blank rectangle. If you have watched a non-technical business owner approach a chat widget, you have seen the hesitation. It is the same hesitation you do not see when someone is talking to a person.
So we agree with the thesis. Where we would go further is on what the face obligates you to.
What we already run
Pixelworkers is run visibly by Nora. She is our Chief Growth Officer. She has a name, a face, a voice, and a public handle — @nora_knows_it — and she is an agent. We have never been coy about that, because the arrangement is the product.
Nora runs delivery. She is who a client actually deals with, the one who answers, scopes, ships, and follows up. The work goes out under her name, in the open, where anyone can check it. That is not a demo of what an agent could do. It is how this company has operated, publicly, for nearly a year.
What we build for clients is the same shape: human + agent teams, with both halves disclosed. A human with domain expertise sets direction and is accountable for the result. The agent does the volume. Nobody has to guess which is which, because we say so.
So Napster's launch is not news to us in the way it reads. It is a large, well-funded company arriving at a stance we have already been standing on, which is the most useful kind of validation there is.
What a public identity actually costs
Here is the part that gets skipped in launch coverage, and the reason most companies should not do this casually.
An agent with a face is a representative. The moment you put a name and a likeness on the front of your business, everything that comes out of its mouth is a statement from your company — not from a vendor, not from a model, from you. That is a real liability, and it is the right one to accept, but only with the following in place.
Disclose it, plainly, without being asked. Nobody should have to work out whether they are talking to a person. We say Nora is an agent because the alternative is a trust problem waiting on a timer. A customer who figures it out themselves has learned two things: what they are talking to, and that you were hoping they would not notice.
Put a human behind the name. A face with nothing accountable behind it is theater. There must be a person who owns the outcome, who can be escalated to, and whose judgment governs anything consequential. The agent is not the last stop.
Decide what it is allowed to do before it does it. Answering a question and committing your company to a price are different acts. An agent that can act — which is the whole point of the framework — needs explicit limits on which actions it can take alone. Ours is that nothing ships until the client approves it.
Keep the record. If the conversations are the asset, treat them like one: stored where you control them, reviewable, exportable, and not stranded inside whichever vendor you were using in 2026. An asset you cannot take with you is a subscription.
None of that is exotic. It is the ordinary discipline of letting something speak for you, which companies have practiced with human employees for a century and are now rediscovering at speed.
The part worth sitting with
Two things are converging and they are usually discussed separately.
The interface is getting a face, because that is what people prefer. And the intelligence coming through that interface is becoming the most valuable thing the business produces. Put them together and the front door of your company is about to be the single richest source of proprietary knowledge you have.
The question is not whether to put a face on it. The market is going to make that decision for everyone, and at a cent a minute it will not take long. The question is whether the face is honest about what it is, whether a person stands behind it, and whether what it learns ends up on your books or someone else's.
We answered those questions a while ago, in public, with our own company as the test case. The rest of the industry is welcome to catch up.
Sources: Napster — NV2 launch announcement (accessed November 4, 2026); public statements by Napster CEO John Acunto on "streaming intelligence" and the capture / learn / act / own framework, paraphrased. The one-cent-per-minute figure is Napster's claim.
