Consulting firms are redesigning their job ladders around AI agents

A title ladder is a pay scale with names on it. Retiring analyst, consultant and manager in favour of role-specific titles organised by job family changes how roughly 181,500 Deloitte employees in the United States are compared with each other, and therefore how they are ranked and paid ^[2]. Accenture arrived at a similar place from the other side. By February, according to Lenin Gali's account, the firm was telling senior staff that failing to use AI tools would cost them promotions ^[4]. That sits inside an $865 million optimisation programme whose stated logic was to exit people for whom reskilling on AI was not viable ^[3].

Bob Sternfels gave CES a ratio in January: 25,000 AI agents alongside 40,000 humans, with rough parity expected by year's end ^[5]. Reaching parity means adding 15,000 agents inside a year, a 60 percent increase on the January count ^[17]. In January the firm was running about 0.63 agents per person ^[18]. McKinsey has also committed to moving about a quarter of its global fees to outcome-based pricing ^[13]. Gali's argument for why that follows: a firm still selling by the hour while running part of delivery on agents is charging for a cost structure it no longer carries ^[14].

"I've watched a few of these shifts before, from the offshoring wave of the 2000s to the cloud and DevOps era," Gali wrote ^[16]. His case for why tool rollouts alone stall is specific. One person finishing their work in half the time does not compress a process that runs through six people; the gain is absorbed by the next bottleneck downstream ^[11]. Roles go, on that reading, when a rebuilt process no longer needs a step that existed only to move work between two other steps ^[12].

BCG reported $14.4 billion in revenue in April, a 22nd consecutive year of growth, with headcount up and the additions tilted toward AI engineers, data scientists and IT architects ^[9]. Gali's point is that the firms cutting roles and the firms hiring aggressively are, in several cases, the same firms ^[10].

The whole record here is one Forbes Tech Council column written by the chief digital officer of Atomicwork, a former CIO ^[1][19]. It does not include measured productivity results for any of the firms named, so the claim that organisational output lags individual output rests on reasoning.

The detail Gali dwells on is an identity question. Inside OpenAI, he writes, employees work from templates the firm calls playbooks, and every employee deploys agents under their own identity ^[7]. If an agent carries an identity, he argues, it has to be audited, validated and accountable for what it produces ^[8]. Applied to a buyer, that changes the pitch from whose resumes are on the staffing plan to who is accountable for the agents, what their access and guardrails look like, and how their output is audited ^[20]. The third shift he names is the forward-deployed engineer, who sits with the customer and turns the fix into a product capability. He attributes the role to agentic deployments that rarely work out of the box ^[15].