The end of the pyramid: how artificial intelligence challenges the agency business model

Published 14 October 20266 min read

Written by

Fahad IshtiaqFounder, Vela

What happens to an agency’s business model when work that once required a substantial production team can be completed with less human effort?

The question reaches beyond adopting another tool. It concerns how the firm staffs projects, develops expertise, prices its service and accepts responsibility for the result. AI puts pressure on a familiar relationship between revenue and billable labour. Firms organised around efficient delivery have an opportunity to challenge competitors whose economics depend on preserving hours.

AI-native agencies can build around that shift from the outset. Established firms face the same choice: redesign how value is delivered, or defend an increasingly costly way to produce it.

Why the pyramid made commercial sense

Coase’s account of the firm explains why coordinating work internally can be less costly than arranging every task through the market (Coase, 1937). In professional services, that coordination often takes the form of a hierarchy: senior staff win and guide work, managers organise it, and junior staff undertake substantial parts of production.

Maister describes the economics of this leverage model, in which the mix of senior and junior time influences profitability (Maister, 2004). Its logic is straightforward: senior expertise supports work delivered by a larger team, and the firm earns a margin on that team’s labour.

Baumol’s analysis of unbalanced productivity growth explains the cost pressure on services where output remains closely tied to labour (Baumol, 1967). For an agency, improvements in how work is produced can change that calculation.

AI becomes commercially consequential where it changes the human effort required for a meaningful part of the work. That can alter staffing needs and create tension between efficient production and billing for time.

Task efficiency changes the calculation

Workplace research shows how the effort required for particular tasks is changing. AI assistance increased issues resolved per hour by about 15% in a customer-support field study, with larger benefits among less-experienced workers (Brynjolfsson et al., 2025). In the BCG consulting experiment, AI users completed more tasks, more quickly and at higher quality within the model’s capabilities. Performance worsened on a task outside those capabilities (Dell’Acqua et al., 2026).

Reported examples are more dramatic but less controlled. Harvard’s Center on the Legal Profession describes a complaint-response application reducing associate time from sixteen hours to minutes in a pilot (Couture, 2025). Reuters reports Kimberly-Clark’s account of content production becoming substantially faster through an AI platform (Soni & Suresh, 2026). Reporting on McKinsey also describes AI use in proposal and presentation work (El Chmouri, 2025).

The reported gains concern particular production steps. An agency calculating its costs must also count review, integration, exceptions and client service.

What “AI-native” means here

As used in this essay, an AI-native agency designs its delivery process around AI capabilities from the outset. People define the problem, direct the systems, evaluate the work and remain responsible to the client. Its advantage must be demonstrated in delivery cost, quality and reliability.

Investors have articulated versions of this model. Y Combinator’s requests for startups describe full-stack AI companies and AI-native agencies (Epstein, 2026; Friedman, 2025). Sequoia’s Julien Bek distinguishes selling a tool from selling the completed work, while Foundation Capital calls the latter service-as-software (Bek, 2026; Foundation Capital, n.d.). The investment thesis is clear: sell completed work, with software doing more of the production.

The important distinction is how the organisation delivers and earns. Adding AI to an existing workflow can improve a firm; redesigning that workflow can change its cost structure more substantially. Either approach can occur inside a new business or an incumbent.

Fixed fees and outcome-linked arrangements also predate generative AI. Their relevance here is the incentive they can create to improve delivery. An hourly firm can also reward efficiency through higher rates, more valuable work or greater capacity; the effect depends on its contracts and client demand. Thomson Reuters discusses this pressure in law-firm economics, while reporting on McKinsey describes its use of outcome-based fees (Ramanathan, 2025; Thompson, 2025).

The industry evidence is mixed

Established agencies are changing at different speeds. Their results show both pressure and adaptation.

WPP reported a 4.7% like-for-like decline in revenue less pass-through costs in the first half of 2026 (WPP plc, 2026). Reporting on WPP, Omnicom’s IPG integration and Dentsu describes restructuring and job reductions (Shah, 2026; Singh, 2025; Social Samosa, 2025). Mergers, account losses and demand also shaped these results.

Other incumbents grew. Publicis reported 4.8% organic net revenue growth in the second quarter of 2026 (Publicis Groupe, 2026). Accenture reported full-year fiscal 2026 revenue growth of 6% in US dollars and 5% in local currency (Accenture, 2026). These results challenge any simple prediction that incumbent scale has become a disadvantage.

Clients have alternatives as well. An ANA survey found that 82% of its responding members had an in-house agency in 2023, up from 42% in 2008 (Association of National Advertisers, 2023). The trend therefore predates the present wave of generative AI, although Reuters’ reporting describes AI supporting further in-housing (Soni & Suresh, 2026).

An AI-native agency must compete with capable internal teams and adapting incumbents, as well as firms that remain attached to traditional processes.

Selling an outcome requires defining it

A hypothetical agency might use AI to prepare several campaign concepts more quickly. That creates options: lower the price, improve the margin, explore more alternatives or spend additional time understanding the client’s audience. The commercial result depends on which option it chooses and whether the final work improves.

A promise to increase sales introduces a different issue. Sales may depend on the client’s product, distribution, budget and execution as well as the agency’s contribution. An outcome-based arrangement needs an agreed baseline, a measurement period and a clear account of responsibility.

Hinterhuber’s research describes obstacles to adopting customer-value-based pricing (Hinterhuber, 2008). The commercial challenge includes agreeing what the customer values and how the price reflects it.

An agency proposing a different commercial model should therefore be able to explain what it delivers, how acceptance is assessed and how changes in scope are handled. Speed makes the offer attractive when the result remains valuable and the terms remain workable.

A thinner pyramid still needs expertise

Changes to junior work raise a further organisational problem. Stanford researchers found a relative employment gap among young workers in AI-exposed occupations, largely associated with reduced hiring, while emphasising that their findings are descriptive rather than causal (Brynjolfsson et al., 2026). Reporting on PwC’s reduced entry-level intake also identifies economic conditions alongside AI (Fore, 2025).

If junior production work shrinks, where will the next generation learn the judgement that senior work requires?

A small senior team can look efficient on an organisation chart. It still needs a way to recruit, train and replace expertise. New entrants face that responsibility alongside incumbents.

The Susskinds describe parallel futures in which technology both improves traditional professional work and enables different ways to perform it (Susskind & Susskind, 2018). Both paths are open to firms willing to change how they work.

The model must earn its advantage

AI changes the economic possibilities of agency work. Firms can reconsider which tasks require people, how production is coordinated and how clients should pay. New entrants can build those decisions into their operations from the start; established firms can redesign around them.

The strongest case for an AI-native agency is an offer it can demonstrate: useful work delivered with a clear standard, efficient production and accountable judgement. That requires more than access to models or a smaller headcount.

AI puts pressure on the pyramid’s economics. Agencies that turn faster production into better work and clear value have a strong offer. The advantage belongs to the firm that can deliver it and stand behind it.

Fahad Ishtiaq is the founder of Vela, an AI-native systems agency in Ontario, Canada. He has a commercial interest in the model this essay describes.

Sources

Every figure in this article was checked against these sources.

All posts