The collapsed pyramid: Redesigning the AI-enabled firm – Part 2

In Part 1 we examined the legitimacy deficit that arises when authority and reward concentrate at the top while responsibility and risk are displaced downwards.

We saw how Gen Z acts as a stress test rather than the cause of the problem, and how the traditional employment bargain is already fraying as artificial intelligence removes the junior work that once formed the foundation of professional development.

Here we turn to the practical question. The economic case for the “obelisk” – a leaner structure of client leaders, engagement architects, and AI facilitators – is strong. Productivity gains from generative AI are real and well-documented.

The danger is that leaders solve the economic problem while making the legitimacy problem worse.

The central choice is not automation versus no automation. It is whether AI is designed to augment human judgement or to remove human agency.

AI: democratiser or digital foreman?

Experimental research shows that generative AI can improve the speed and quality of professional work (Noy and Zhang, 2023). In a large field study of customer-support agents, AI assistance increased productivity on average, with particularly strong gains among less experienced and lower-skilled workers.

The technology appeared to diffuse the practices of more capable colleagues (Brynjolfsson, Li and Raymond, 2025).

Read: AI saves time but most companies waste the gain

Used well, AI could therefore democratise expertise. It can give younger employees access to methods, knowledge and analytical support previously controlled by experienced specialists.

It can reduce low-value administration, broaden participation in decision-making and allow small teams to produce work that once required several hierarchical layers.

This is the positive promise of the obelisk: judgement becomes more valuable than headcount, and competence matters more than proximity to power.

But technology does not arrive in an organisational vacuum. It enters an existing structure of incentives, status and control.

ADVERTISEMENT

CONTINUE READING BELOW

Research on algorithmic management shows that digital systems can intensify surveillance, narrow employee discretion and make managerial control harder to see or contest (Kellogg, Valentine and Christin, 2020).

The leader can then claim that “the system” allocated the work, rejected the candidate, calculated the performance rating or identified the redundancy. Authority remains at the top, but accountability disappears into the algorithm.

That is the collapsed pyramid with better software.

Read: The collapsed pyramid: When authority leads without legitimacy – Part 1

Jarrahi (2018) argues for human – AI symbiosis, combining machine strengths in computation with human strengths in contextual, intuitive and ethical judgement.

Raisch and Krakowski (2021) describe the automation – augmentation paradox: organisations must pursue efficiency without allowing automation to undermine the human capabilities on which future performance depends.

The apprenticeship paradox

The greatest strategic risk may not be job loss. It may be judgement loss.

Bainbridge’s (1983) classic “ironies of automation” warned that automation can remove routine practice while leaving humans responsible for the rare, difficult situations the system cannot handle.

The less people practise, the less prepared they become for precisely the moments when human competence is essential.

Professional judgement is not downloaded. It is accumulated through repeated exposure to imperfect information, client tension, mistakes, correction and reflection. If AI completes every first draft, young professionals may become excellent prompt operators without becoming excellent thinkers.

ADVERTISEMENT:

CONTINUE READING BELOW

Read/Listen: Investment strategies, AI and human governance

The answer is not to preserve meaningless junior work. It is to redesign apprenticeship deliberately.

Early-career employees should use AI, but they should also be required to challenge it, verify sources, explain reasoning, defend recommendations and observe the consequences of decisions.

AI facilitators must not become a new clerical underclass servicing senior “judgement owners”. They need progressive decision rights, client exposure and accountable opportunities to build judgement.

This is particularly important in professional services. In the traditional pyramid, junior employees learned by doing work that was sometimes repetitive but exposed them to patterns, clients and errors.

In the obelisk, that learning must be designed rather than assumed. Senior people will need to invest more, not less, time in coaching, reviewing reasoning and explaining context. A leaner structure does not remove leadership work; it makes genuine leadership more visible.

The new legitimacy bargain

Leaders moving from pyramid to obelisk must offer a new employment bargain.

First, productivity gains must be shared. If AI produces more output with fewer hours, the benefit cannot flow only into executive reward and headcount reduction.

It should also fund development, improved work design, greater flexibility and more meaningful roles. Employees will not regard transformation as legitimate when efficiency gains travel upwards while insecurity travels downwards.

Second, responsibility must remain attached to authority. An algorithm may inform a decision, but a named human leader must own it, explain it and provide a route of appeal.

No employee should be told that an opaque system made a consequential decision that nobody is prepared to defend.

Third, employees need voice before deployment, not consolation afterwards. Consultation strengthens procedural legitimacy because people can influence the systems that will influence them.

ADVERTISEMENT:

CONTINUE READING BELOW

It also improves implementation: those closest to the work often understand the unintended consequences of automation before senior leaders do.

Fourth, organisations must measure capability creation as seriously as cost reduction. A structure that improves today’s margin while destroying tomorrow’s leadership pipeline is not efficient. It is merely consuming its own future.

Fifth, leaders must submit themselves to the transparency they expect from others. If AI measures employee productivity, it should also illuminate decision delays, executive overrides, unequal benefit allocation and the quality of leadership judgement. Transparency that flows only downwards is surveillance, not governance.

The right to lead

AI is likely to make many organisations leaner, and Gen Z is likely to make their contradictions more visible. Neither development automatically creates a leadership crisis. The crisis arises when leaders mistake positional authority for earned legitimacy.

The obelisk may be the right structure for the AI-enabled professional firm. It is faster, more expert and potentially more empowering. But an obelisk is also narrow, concentrated and dependent on the strength of its foundations.

Read: ‘We’re becoming an AI-enabled organisation’ – Standard Bank’s Margaret Nienaber

If those foundations are fairness, competence, reciprocity and accountability, it may stand taller than the pyramid it replaces.

If they are not, it will simply become a more elegant monument to concentrated power.

Dr Chris Blair, group director of 21st Century.

#collapsed #pyramid #Redesigning #AIenabled #firm #Part