OpenAI Answers Its Own Phones Now. That Is a Consulting Line Item Disappearing.
Three weeks of records show model vendors shipping the service layer themselves while Asian integrators fund the other end of it. Here is the one sentence in an AI services pitch that tells you which side a firm is standing on.
When OpenAI launched Presence in late July, the detail worth stopping on was not the product page. It was the reference customer. Presence already runs OpenAI's own English-language phone support, resolving most inbound issues without a human touching them.
That is a demo. It is also a services contract that no longer needs to be signed with anyone.
A fortnight later the Financial Times asked the obvious follow-up question: if generative AI can partially automate the analysis, what exactly is the client paying the high fee for? The column's answer is the interesting part. Implementation, change management and political capital are becoming the scarce inputs. Slide decks are not.
Read those two records together and you get the shape of the enterprise AI services market for the next two years. The advisory half is being commoditised by the same vendors who sell the models. The delivery half, the ugly part where a workflow actually has to survive contact with a regulated business, is where the money moves.
## The squeeze is coming from two directions at once
From the top, the model vendors are selling the outcome directly. Presence is policy-bound agents for support, sales and internal workflows, in limited general availability, sold to the enterprise without an integrator standing in between.
From the side, the firms best positioned to absorb that are the ones who own a model of their own. Fujitsu is building the Uvance for Finance AI Transformation Platform on its own Takane large language model, with guardrail tooling and multi-agent systems aimed at loan screening and branch operations from August 2026. That is a very different position from reselling somebody else's API. Fujitsu owns the model, the guardrails and the workflow, in an industry where the compliance burden is the moat.
Then there is the layer being funded right now, and most of it is not in San Francisco. Singapore's Whale raised a $40 million Series C extension in July, taking its Series C to $100 million, to push an enterprise AI operating system across Asia-Pacific, North America, MENA and Europe. In Mumbai, Vishleshan AI joined Anthropic's Claude Partner Network, taking certifications and enablement material to deploy Claude-based systems for enterprise clients.
Those two are not the same bet, and the difference matters more than the size of either round.
## Certification is not a moat
Whale is building something it owns and charges for. Vishleshan is building a practice on top of a vendor programme. Partner-network membership is a certificate the vendor issues, and the vendor can issue it to a hundred more firms next quarter, in the same city, at the same price. It buys distribution, not defensibility.
That is the test worth applying to every AI services pitch you see this year. Find the sentence that says what the firm actually owns. If the answer is a partner tier, a certification and a methodology, then the model vendor sets the ceiling on the margin and can lower it whenever it wants to.
The incumbents can see this. Accenture's Iberia CEO spent early August co-writing an op-ed with Microsoft Spain's president urging Spanish firms to move from small experiments to deep integration across core operations. Strip the framing and that is a consultancy repositioning itself onto exactly the ground the FT column says is still scarce: the long, political, unglamorous work of getting a thing into production. It is the right move. It is also an admission about where the easy revenue went.
## What to do with this
**If you are buying:** ask your services vendor what share of the engagement is billed for advice versus for running something in production with an SLA attached. The first number is the part their own supplier is coming for. A firm that cannot answer has not priced its own risk.
**If you are watching the market:** track how many enterprise AI announcements name a model vendor as the direct counterparty rather than an integrator. Presence is the first big one we have logged. If that count climbs through the autumn, the services multiple that AI was supposed to lift is going the other way, and the firms holding a proprietary model plus a regulated workflow are the ones to own.