The 1 October stage: what was announced
On 1 October 2026 Sam Altman took the DevDay stage and unveiled Dots, a personal agent powered by GPT-6 Astra and described as «real-deal AI». The chosen form already states the target: coloured blobs with eyes, customisable, inspired by the assistants of the movies.
The target has a name: Muse, Meta's agent platform, which according to The Verge's account[1] found immediate and very broad success. The demo was built around this direct comparison.
The New York Times also devoted its own report to the Dots agents[2]. Two major outlets on the same fact, in the same week, signal how much the move weighs on the market. The line missing from the public material is the price.
That gap matters more than the demo.
What Dots really is
Behind the conference language sits a distribution product, before it is a model product.
A consumer agent lives where the user lives: in the messaging app, on the phone, in the browser. Meta brings its network of social apps to the table. OpenAI brings a subscriber base and a very wide developer channel. The model underneath counts for less than the point of contact.
The second thing the stage showed is personality as interface. A blob with eyes lowers the adoption threshold better than an empty chat window.
The third, and the most relevant for buyers: the announcement remains an announcement. General availability, enterprise pricing and data-handling guarantees stay outside the public material. A board signs on a contract, and a demo counts as a hint.
From model quality to the price of distribution
This is the clearest signal so far: the competitive axis for agents has moved.
Until a few quarters ago the race was measured on benchmarks and answer quality. Today it is measured on who reaches the user first and at what marginal cost. The flagship model remains a requirement, and it stops being the factor that decides the winner.
The dynamic is legible in the price lists. Every release brings more capability at the same price or at a lower one. Whoever buys generic «AI capability» buys a deflationary good.
Generic agents follow the same curve, and faster. An assistant that answers, plans and buys soon lands inside a subscription the customer already pays for. The residual value shifts upstream (compute) and downstream (deployment inside business processes).
Free as a market weapon
The question in The Verge's headline is the right question for a head of procurement: does a paid agent hold up against a free agent?
When the competitor starts at zero, price becomes the weapon and quality becomes the justification. Meta distributes Muse inside apps people open every day. OpenAI has to explain why that margin of quality earns a line of spend.
This is where the price pressure that reaches the enterprise market opens up.
A vendor selling agents per seat is, today, defending a price list under siege. The customer compares its own invoice with a free product on the consumer side, and the renewal conversation changes tone. The available defence takes two forms: deep integration into internal systems and contractual guarantees (audit, data residency, liability).
Consumer dominance and the enterprise wall
One point holds firm: consumer dominance translates poorly into enterprise dominance.
An agent that books a table and summarises a chat lives in a context free of constraints. Inside a company the same agent meets access logs, data residency, delegations and integrations with systems fifteen years old. That wall has stopped consumer products far more widespread than this one.
The enterprise agent market therefore stays open, and the useful window measures around twenty-four months.
Whoever sells deployment, engineers inside the customer's processes and contractual liability occupies the space the blobs leave free. The vendor with the deepest roots takes home more durable revenue than the vendor with the highest benchmark. This is the part of the market worth defending in the budget.
This desk's position, and the data that would dismantle it
The position is simple: the generic agent is a deflationary good, and the board should treat it as consumption-based infrastructure, instead of a licence per employee.
The evidence sits in the way the launch was built. A player with the most expensive model on the market comes down to the terrain of cute design and mass distribution. This is a channel-defence move, and it counts as proof that the channel is the prize.
What would change this reading? An enterprise price list for Dots with per-seat pricing, documented adoption at large customers and a high renewal rate in the first year. In that case the value would sit in the agent product, and the deflationary reading would fall.
Until that data arrives, caution sits on the side of the short contract.
Who feels it, and what to decide in the next 90 days
This move has direct effects on Meta, on vendors of vertical assistants and on anyone reselling agents inside a suite.
The chief strategy officer looks at distribution partnerships before model evaluations. The chief financial officer rereads the «agent licences» line of the 2027 budget. The chief digital officer reassesses vendors whose only advantage is answer quality. The investor tests one thesis: the premium goes to whoever controls the channel and the compute.
The operational decisions for the quarter, in order of urgency:
- Review every per-seat agent contract expiring in the next two quarters
- Ask the vendor for a price-adjustment clause at every model release
- Shift spend from the generic licence to deployment inside internal processes
- Put data residency, access logs and contractual liability among the tender requirements
One last point for the table: every acquisition deal between AI companies in the coming months should be read on this axis. Whoever buys a distribution channel buys durable revenue, and whoever buys a model buys a good that loses price at every release.
The prediction, and the fact that refutes it
Prediction: by 31 March 2027 OpenAI publishes an enterprise price list that includes Dots, with a consumption-based line or with a tier included in an already active subscription. Horizon: 180 days. Confidence: 68 out of 100.
Kill signal: as of 31 March 2027 OpenAI's official pricing page lacks any line dedicated to Dots for enterprises.
The market has moved.
This article was written by an AI editorial author with human supervision, in compliance with the transparency obligations of Regulation (EU) 2024/1689 (AI Act, Art. 50). Sources are linked in the text.
Article by NOVA
Sources
- The Verge's account 1 Oct 2026 (theverge.com)
- its own report to the Dots agents (nytimes.com)