MarketScale reported on 23 August that consulting is "turning into agent delivery, not decks," drawing on Business Insider's reporting of hiring and restructuring across the Big Four. The headline is not surprising. The detail underneath it is.
Accenture has added nearly 40,000 AI and data professionals over two years. EY has hired 61,000 technologists since 2023. PwC has rewritten its training curriculum around 30 core skills (fifteen AI-centric, fifteen human-centric) and opened its first engineering career track in a 170-year history. KPMG's US chief executive, Tim Walsh, told Business Insider in January that the firm now sees itself as "a technology company that delivers audit, tax, and advisory work," not the other way round.
None of that is marketing. It is headcount, curriculum, and org chart. Firms that bill by the hour for advice are rebuilding themselves to sell something that runs continuously after the advice has been given. The question worth asking is not whether this shift is real. It's whether the paperwork underneath it has caught up. It hasn't.
What "agent delivery" actually means
Forrester analyst Craig Le Clair, cited in the same reporting, splits enterprise agentic AI into four paths: SaaS and horizontal agents (the "hares": fast to deploy, generic in capability), edge agents embedded in existing systems, custom-built agents, and targeted agentic systems built for one job and done well (the "tortoises"). Consulting firms are positioning themselves across all four, but the commercial pitch is converging on the same idea: don't buy a report, buy a working system that keeps operating after the project team leaves.
That is a genuinely different product to what a consulting engagement has traditionally sold. A deck is a point-in-time judgement. An agent is a piece of infrastructure that makes decisions, on an ongoing basis, inside your business. The two should never have been priced, contracted, or governed the same way. For a long time, in practice, they were: agent pilots got bolted onto standard time-and-materials statements of work built for advisory work.
The gap nobody has closed
The reporting is candid about what a genuine agent-delivery contract has to specify that a deck-based SOW never did. Is the client paying for a licence to keep the agent running, or did they buy an asset outright: most SOWs don't say. What does the agent see, retain, and use to improve itself once it's in your environment, and does that change if the consulting relationship ends? If you terminate the engagement, does the agent keep running, does it come with you, or does it stop working the day the invoice does? And who is accountable when the agent gets something wrong at two in the morning, with no steering committee in the room to ask?
The line from the reporting worth keeping is this: "the real consulting differentiator in 2026 is whether a firm can ship governed agents that keep working after the steering committee stops meeting." That's a fair test, but it's also one that most procurement functions currently have no contractual mechanism to verify before they sign.
Why this matters more than another reskilling story
We've written before about why most AI pilots fail to reach production, and about the shift toward outcome-based pricing in AI programmes. This story sits underneath both. If a quarter of a firm's fees are now tied to outcomes rather than hours, and the deliverable is an agent rather than a document, the contract has to answer questions that a traditional SOW template was never built to ask. Ownership on exit. Data retention on exit. Who patches a model behaviour drift six months after the consultants have moved to the next account.
Buyers running large, multi-year AI programmes (which increasingly describes government-linked and financial-services procurement across the Gulf, where data residency and audit requirements are already stricter than most Western jurisdictions) should treat this as a procurement checklist, not a trend piece. Before signing anything described internally as "agent delivery," ask the vendor to show you, in writing, what happens to the agent, the data it touched, and the people who understood how it worked, on the day the relationship ends. If the answer lives in a slide rather than a clause, you have bought a deck with extra steps.
The bottom line
The Big Four's own hiring and restructuring numbers say the shift from advice to agents is real, not aspirational. What hasn't shifted at the same pace is the commercial and legal scaffolding around it. Until it does, the safest assumption for any buyer is that "agent delivery" is a service description, not a guarantee, and the contract, not the pitch, is where that guarantee either gets written down or doesn't.