A crack is opening in the systems integrator model

The pyramid arithmetic behind enterprise AI transformation is breaking. Weigh AI vs a systems integrator and the shift is showing up in contract renewals before it shows up in the trade press.

At one Fortune 100 manufacturer, a supply chain workstream that had sat with a large systems integrator moved to a small AI-native firm. The incumbent kept its other lines of business with the account; it lost the piece where AI agents plus named experts had shipped faster, through consecutive 12-week cycles, than a staffing plan built around junior analyst hours. Nothing about the change appeared in a press release. That absence is the story: this kind of shift happens inside renewal negotiations, quietly, one workstream at a time.

The arithmetic that built the systems integrator pyramid

The economics of a systems integrator are simple once someone draws them out, which is why so few clients ever ask to see them. A partner's compensation, and the firm's margin, is funded by leverage: a wide base of junior consultants billed at rates well above their cost, a thinner layer of managers billed a bit higher, and a small group of partners at the top billed the most while doing the least direct work. The junior layer is the model.

That pyramid also doubled as a training system. A first-year analyst spent two or three years building spreadsheets, writing first-draft memos, and cleaning data before earning a seat closer to the client. Firms tolerated the inefficiency of junior work because it was the only path to producing the senior people who would eventually run engagements and sell the next one.

Why the systems integrator arithmetic breaks when AI agents do the junior work

AI agents now do a meaningful share of that junior-tier work directly: pulling and reconciling data, drafting first-pass analysis, running steps that used to occupy a graduate class for a year. When that work moves to software, the billable hours that funded the pyramid do not just shrink; they stop existing as a line item at all. A firm cannot bill for hours nobody worked, and it cannot charge time-and-materials rates for output a machine produced in minutes.

The result shows up first in hiring, not in client-facing language. AI-native boutiques are growing roughly 50% a year, while several of the largest integrators have cut graduate and analyst intake over the same period, according to 2026 trade reporting on consulting hiring. Fewer juniors on staff also means less depth accumulating inside any one engagement: the person who used to learn a client's business over three years of staffed rotations is increasingly an agent with no institutional memory, supervised by a senior person spread across several accounts. Per-engagement experience thins out even as the firm's total volume of engagements grows.

The leading indicator shows up at contract renewal, not in a press release

The manufacturer's renewal is not an isolated event. Procurement teams at large enterprises increasingly run two vendors side by side, a legacy integrator and a smaller AI-native shop, and quietly shift workstreams to the smaller one at renewal. None of this generates a headline, because a renewal is a private negotiation, not an announcement.

The clearest public confirmation is coming from inside the incumbents themselves. BCG has said it will phase out the billable hour for its core strategy division, a retreat from the exact unit of measure that built its own partner economics. A firm does not walk away from its founding metric because business is good under it.

That is the tell.

Agentic AI consulting replaces the pyramid, but not yet at every scale

What is replacing the pyramid is a different shape entirely: a small pod of named senior people, each responsible for a workstream end to end, each orchestrating a fleet of AI agents to do the drafting, data wrangling, and first-pass analysis a graduate class used to handle. Fees are outcome-staked rather than billed by the hour, with a portion held back until the client's own finance team verifies a value target. Work runs in 12-week cycles, with the value baseline locked before build starts, not inside a multi-year statement of work. That is what an AI-native transformation engine brings: a different unit of delivery altogether.

Future Works, a three-year-old firm founded by Matt Leta, built its model around this structure, the operating partner approach to enterprise AI: AI agents plus named experts rather than a substitution for them, instead of a staffing pyramid. The approach is unproven at the scale of a five-year, multi-country ERP replatform; no AI-native firm has yet run a program that large end to end. That is precisely why the cycles are short and the fees are staked: each 12-week increment has to earn the next one on its own record, rather than asking a client to fund a multi-year roadmap on a promise.

For a CIO or COO auditing their own vendor stack, three checks say more than any pitch deck: whether the seniority of the people who scoped the engagement matches the seniority of the people still on it a year later, whether AI agents for enterprise workflows are actually running inside the largest workstreams rather than only a sandboxed pilot, and whether any share of the fee is contingent on a number the client's own finance team validates instead of one that is fully fixed regardless of outcome. A vendor stack built on pyramid staffing, sandbox-only AI, and fixed fees is still running the old arithmetic, whatever language the pitch deck uses. The renewal conversation, not the sales deck, is where that arithmetic gets tested.

If you are comparing an incumbent's renewal terms against an AI-native operating partner's, the commercial structure tells you which arithmetic the vendor runs: start by comparing the two models side by side.

Matt Leta, Founder and CEO, Future Works.