OpenAI Presence isn't a product launch, it's a consultancy with no juniors

For the past two years, any company that wanted to put AI to work seriously ended up signing two separate contracts with two vendors who never sat at the same table. One went to a lab — OpenAI, Anthropic, Google — and bought access to a model: tokens, an API, a bill that moved up or down with usage. The other went to a consulting firm, big or small depending on the budget, and bought something quite different: people. People who understood the internal systems, who knew how to negotiate with compliance, who sat in a conference room for weeks until the project actually shipped, and who then invoiced by the hour. The lab sold intelligence. The consultancy sold the patience of turning that intelligence into something that actually worked inside a real organization, with its legacy systems, its hierarchies, and its people who didn’t want to change how they did things. Nobody confused the two businesses. They weren’t even competing for the same budget.

That quiet division of labor had been cracking for months. On July 22, it broke for good.

That day, OpenAI unveiled Presence, a product for deploying voice and chat agents inside enterprises: customer support, outbound sales, internal IT help desks. None of that is surprising in 2026. What matters is how it’s sold: no pricing page, no self-serve signup, no published rate card. Deployments are led by OpenAI’s own Forward Deployed Engineers — the title Palantir made famous — alongside a small, hand-picked list of systems integrators, with every contract scoped and negotiated per customer and use case. That is, quite literally, a consulting firm’s business model. OpenAI could have shipped Presence as an API with a price per token. It chose embedded engineers, negotiated scope, and per-customer commercial terms instead. That choice isn’t a launch-day footnote. It’s the whole thesis.

Presence has no pricing page because it isn’t selling a product

Behind Presence sits a structure OpenAI had been quietly assembling for months. In May it launched the OpenAI Deployment Company with more than $4 billion in funding, led by TPG, Advent, Bain Capital, and Brookfield among nineteen partners spanning investment firms, consultancies, and other systems integrators — Bain & Company among them, not as a client but as a founding partner. The new company’s first acquisition was Tomoro, an applied-AI consultancy based in London with offices in Edinburgh, Singapore, Sydney, and Melbourne, which brought along roughly 150 Forward Deployed Engineers already fluent in wiring models into real business processes. Presence is the product layer sitting on top of that machinery: agents wrapped in policies, guardrails, simulations, and a continuous-improvement loop, branded and packaged to look like a platform, when what a customer is really buying is the integration work underneath.

The early customer list gives away what kind of engagement this actually is: BBVA piloting voice banking support in Mexico, SoftBank testing Japanese-language conversations, the Australian insurer IAG preparing for peak-demand periods. OpenAI is also happy to point out that its own support line now resolves 75% of inbound calls without a human, with a Codex-driven improvement loop that cut handoffs to people by fifteen percentage points in ten days. Numbers OpenAI is grading on its own homework, worth remembering — but they point in a clear direction: this isn’t a demo, it’s an operation built to scale.

Anthropic signed the same deal through a different notary

None of this should be read as an OpenAI-only move. Back in May, Anthropic announced a $1.5 billion joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman to do essentially the same thing: embed around a hundred of its own engineers directly inside client offices. They call it Ode, they pitch it as a “scaled boutique,” it runs Claude-first without being exclusive to it, and it’s led by Chris Taylor, formerly CEO of Fractional AI, the consultancy Anthropic absorbed to get the whole thing off the ground. Its chief technologist summed it up in a line that should unsettle any traditional consultancy paying attention: model choice matters, but that isn’t where most of the effort actually goes. The competitive edge has stopped being which model you use. It’s who does the integration.

What matters here isn’t that two rival labs happened to have the same idea at the same time. It’s that both, independently, reached the same conclusion about where the money actually is: not in selling intelligence by the token, but in charging to install it. Ode names its competitors explicitly, and they aren’t other labs — they’re OpenAI’s own Deployment Company and, more awkwardly still, Deloitte and Accenture, both of which have had to stand up their own embedded-engineer practices just to stay in a conversation that, a year ago, was entirely theirs to have.

The market had been pricing this in for months

And here’s where the euphemisms need to stop, because the numbers were already sitting there before anyone gave them a name. In June, Accenture posted quarterly results that in any earlier year would have counted as good — earnings per share up 9%, revenue up 6% — and the market responded by wiping almost 20% off the stock in a single session, its worst day in years. It wasn’t the quarter. It was the guidance: full-year revenue growth cut from a 3-5% range down to 3-4%, new bookings down from $19.7 billion to$19.3 billion, and consulting revenue — the part billed by the hour, not the software — up just 1%. The stock now trades near a multi-year low, at a price-to-earnings ratio around eleven, a level nobody had seen in a long time. That same morning, not by coincidence, Accenture announced $4.18 billion in cybersecurity acquisitions: a fairly blunt way of telling the market it wants to stop depending so heavily on hours billed for implementing someone else’s AI.

The shockwave didn’t stop at Accenture. Capgemini fell 8.4% the same day, hitting a 52-week low and bringing its twelve-month decline to 38%. Infosys is down more than 30% for the year, after large deal bookings dropped by a third in April and the company trimmed its growth outlook to a bare 1.5-3.5% range. Accenture’s market value has gone from $260 billion at its late-2021 peak to just over$108 billion. PwC has cut 5,600 jobs worldwide. The Big Four as a group is shrinking graduate intake in the UK. Anyone who’s ever defended a utilization rate in front of a partners’ committee recognizes the smell of these numbers: this isn’t a market correction. It’s an entire business model announcing, out loud, that it no longer knows what it’s worth.

Worth resisting the urge to stop at the apocalyptic headline, though. Gartner published a note the same day Presence launched warning that, by 2027, half of the organizations currently planning to move customer service to AI agents will have walked those plans back. Human contact remains genuinely irreplaceable in a non-trivial share of interactions, and the recent history of call-center automation is full of projects that quietly reversed course. Presence can fail in half its deployments and still permanently rewrite the rules of the business it’s competing in. Both things are true at once.

The pyramid was never built on intelligence, it was built on juniors

This is the mechanism that actually explains the market panic, and it has nothing to do with how good the models are. Large consultancies run on a pyramid: an army of junior analysts bills cheap, plentiful hours, those hours fund the partners’ margin, and knowledge moves upward as juniors survive long enough to become seniors. McKinsey already ties roughly a third of its work to outcome-based fees rather than hourly billing. BCG pulls 40% of revenue from AI- and tech-focused work. The model has been shifting for a while — but it still depended on a wide base of young people learning the trade through repetitive work that was well paid for what it actually required.

A lab doesn’t need that base. Its Forward Deployed Engineers aren’t juniors in training — they’re senior engineers who already knew how to solve the problem before they showed up, funded by a capital round that doesn’t demand the investment back this quarter. OpenAI and Anthropic can afford to sell implementation below cost for years, because what that negative margin is really buying is the direct relationship with the client, access to its data, and the habit of already knowing which model to reach for next time. No traditional consultancy can compete with a balance sheet like that. And no traditional consultancy can compete without the base of the pyramid, which happens to be exactly the part of the job an agent now does better and cheaper.

In the short term, whoever already holds the channel wins

Over the next few months, the party doing best out of this, oddly enough, is part of the Big Four itself: Deloitte and Accenture already show up as selected integrators inside OpenAI’s and Anthropic’s own deployment networks, billing above the table as preferred partners instead of competing under it. That’s real relief, with an uncomfortable footnote attached: pricing and terms aren’t set by the consultancy anymore, they’re set by whichever lab decides who makes the shortlist of approved integrators. You can keep billing while losing the ability to set the price. Accenture already reports $2.2 billion in AI-linked bookings, a number that sounds great on the earnings call and, project by project, hands the lab exactly the client data and access it needs to stop needing the integrator within a couple of years.

In the medium term, the mid-size firm has nowhere left to sit

The middle tier is the one that fits worst into this new map, and not by accident: it lacks the capital and channel muscle to make the Big Four’s shortlist of preferred integrators, and it lacks the deep specialization to sell itself as something a lab can’t just replicate with its own team. It’s the mid-size generalist consultancy — for years the firm that lived precisely in that in-between spot, big enough for serious projects, small enough to stay flexible — that today has neither OpenAI’s phone number nor a niche worth defending. What used to be a comfortable position has turned into the worst square on the board.

In the long run, there’s only room for the specialist or the subcontractor

Where there’s still room to survive, even to grow, is at the extremes. Boutique firms founded by former Big Four partners are already selling delivery teams built around 80% agent-driven work and just 20% human hours, aimed at clients who want the outcome, not the billing structure. Private capital has started betting on that model: more than €500 million poured into a single boutique tax-advisory firm with the stated goal of hiring a hundred partners in five years. Deep, vertical specialization — sector-specific regulation, technical judgment no agent can still handle unsupervised — keeps its value precisely because a lab with near-infinite capital would rather subcontract that niche knowledge than rebuild it in-house for every single client. What doesn’t survive is the mid-size generalist: not big enough to negotiate as an equal with a lab, not focused enough to be irreplaceable at anything in particular.

Who trains the next generation if nobody hires the bottom rung

There’s a question none of the press releases bother to ask, and it’s the most uncomfortable one of all. If the labs don’t need juniors because they hire seniors trained somewhere else, and traditional consultancies can no longer afford a wide junior base because those hours no longer sell, where does the next senior engineer capable of doing this work in ten years actually come from? A KPMG leader put it in almost personal terms while talking about her own firm’s future: she wants the organization to still exist, and said so precisely to underline how disruptive she thinks all of this is going to be. That’s not an optimistic sentence. It’s a Big Four executive admitting, in public, that her firm’s own continuity is no longer something to take for granted.

AI isn’t killing consulting, it just took away the alibi

None of this means consulting is going away. It means the fiction propping it up is gone: the idea that deploying AI inside a company necessarily required an army of people billing by the hour, trained rung by rung, insulated from direct competition with whoever built the model in the first place. That fiction held up as long as the labs were content to sell access. The moment they decided to sell the implementation too, half the industry discovered that its real product was never the intelligence — it was the patience of installing it, and that patience can now be bought straight from the manufacturer. Whoever holds the channel, or holds a specialty nobody else can genuinely do, still has a seat at this table. Everyone else just lost the alibi they’d been repeating for two years.