The Leadership SHIFT: why most AI and data-focused organisations are only half-built
A framework for leaders redesigning both the machine and the human system around it, before the culture invoices them.
The shift is not about deploying more AI. It is about designing the human system to accelerate with it. The organisations that will separate from the field are not simply those that build the best AI architecture. They are those that close the integration gap: designing both the machine and the human system around it with equal rigour, before the culture invoices them.
Where is efficiency rising in your world, but human significance quietly falling?
The shift most leaders have not made
Leadership is shifting, and most organisations are only halfway through the shift they need to make. The gap is defined by two numbers. The upside is 810 times: Garry Tan at Y Combinator describes the productivity differential between a founder operating with a full AI agent workforce and one working alone. Josh Bersin calls what emerges the superworker: human judgment amplified by an agentic layer. The downside is minus 0.23: MIT's meta-analysis of 106 studies found that human and AI combinations perform measurably worse than the better of the two alone in decision-making contexts. The cause is overreliance. The gap between those two numbers is the human conditions of work. This paper is about closing it.
The structural case is this. The Founder model of leadership, which served brilliantly through the first two decades of the digital era, is becoming obsolete. AI does not merely automate tasks. It changes the architecture of how organisations learn, decide and scale. In the Architect model, the intelligence layer sits at the centre of organisational design rather than at the leader's desk or in the hierarchy's approval chain. This is the leadership form that accelerates with AI. As Ayman Al-Abdullah puts it, if your organisation chart looks the same as it did in 2024, you are already behind.
AI adoption is widely treated as a technology challenge: implement, train, manage resistance. That framing is incomplete. AI is a sociotechnical redesign event. It changes the relationship between people and their work, not just the tools they use. Leaders who treat it as a technology problem will be invoiced by their culture later. Speed, as the evidence makes clear, is not the same as significance.
The Architect model, as currently constructed, is half-built. The discourse on AI-led transformation focuses overwhelmingly on the machine side: data architecture, operating model redesign, AI deployment and structure. It treats the human system as an implementation variable that will self-correct through training, change communications and what most organisations quietly call natural assimilation. The evidence says otherwise.
Human and AI combinations performed significantly worse than the best of either humans or AI alone in decision-making tasks, not because the technology failed, but because the human layer was undesigned.Vaccaro et al., Nature Human Behaviour, October 2024: meta-analysis of 106 studies and 370 effect sizes
Klarna is the most vivid live case. In 2024 its AI assistant was credited with doing the work of 700 customer service agents; the company celebrated the efficiency and left assimilation to handle the rest. By 2025 its chief executive, Sebastian Siemiatkowski, acknowledged that cost had been too predominant an evaluation factor, and that the result was lower quality. Klarna began hiring human agents again. Gartner had predicted the pattern: by 2027, half of the organisations planning large cuts to customer service staff because of AI would abandon those plans.
The counter-model is DBS in Singapore. DBS built its AI integration around a deliberate governance framework, PURE, and reported more than S$1 billion of measured economic value from AI in 2025. It expects to create around 1,000 AI-related roles while some 4,000 temporary and contract roles phase out over three years. It did not deploy AI and wait. It integrated it.
This paper develops both arguments together, the structural and the human, and synthesises them into the SHIFT framework: five design disciplines that complete the Architect model by applying equal rigour to the leadership and human integration layer that most organisations leave to chance. It draws on advisory work in two live AI transformation contexts and three bodies of research: Orlikowski and Scott's 2025 work on AI in the making, Shneiderman's human-centred AI governance model, and Brynjolfsson, Li and Raymond's study of 5,179 customer support agents.
The three eras and the turn we are making
Ayman Al-Abdullah's essay Founder Mode Is Dead extends Paul Graham's 2024 insight in a direction Graham did not anticipate. Graham argued that the best founders stayed close to the real work, and that conventional wisdom about delegating to professional managers was wrong. Al-Abdullah's provocation is sharper: AI is making that advantage obsolete. The founders who work out what comes next will pull away from everyone else, including other founders still running on the old model.
| Mode | Structure | How decisions travel | The problem |
|---|---|---|---|
| Manager mode | Pyramid | A customer signal passes through four management layers before it reaches someone who can act. Three people have already filtered it. | Slow, distorted, diluted |
| Founder mode | Flat | Decisions concentrate around a single leader instead of a chain of approvers. Faster and more direct. | Does not scale; the founder becomes the bottleneck |
| Architect mode | Systems-led | The leader builds frameworks that distribute authority. Good decisions happen without the leader needing to be present. | The emerging model, pulling away fastest |
In Architect mode, the intelligence layer sits at the centre. Every functional unit connects to it directly. A signal from a customer churn event, a failed interaction or a rejected application goes straight to the centre and is reflected back to whoever needs to act. No relay. No decay. Every signal that enters makes the system smarter for the next one.
The chief executive's remaining jobs in Architect mode are distilled as Aim, setting the direction AI cannot choose; Army, hiring for the multiplier; and Assets, allocating every finite resource as one portfolio. To these three, this paper adds a fourth.
The three eras of leadershipManager, Founder and Architect modes, after Al-Abdullah (2025). Step through the shift from relay to intelligence layer.Open the interactive modelThe Architect thesis is strategically compelling, but it is almost entirely about system design: how to build the intelligence layer and redesign the operating model around it. It says almost nothing about how to integrate AI agents into the human teams that must work alongside them. That silence is where this paper begins.
The integration gap
When a new senior leader joins a complex organisation, serious companies invest heavily in their integration: structured onboarding, stakeholder mapping, cultural orientation, explicit conversations about norms and expectations, and a managed period before full authority. The investment is organisational design. A poorly integrated executive, however talented, damages the system they join.
When an AI agent joins a team, most organisations do none of this. They configure it, deploy it, announce it in a team meeting, and trust that people will work it out. This is natural assimilation, and it is the design failure at the heart of most AI programmes that underperform their potential.
What people contribute, how their judgment shows up, and what makes them feel valuable or replaceable are redesigned every time AI enters a workflow. Organisations are responding in three main ways, all necessary and none sufficient. The first is technology investment, where spending on AI runs well ahead of readiness: in McKinsey's research, most leaders say their organisations are not ready for the people and culture shifts an agentic future requires. The second is training, which leads the talent response ahead of role or workflow redesign. The third is communication and change management: BCG's 2025 AI at Work survey found that the share of employees who feel positive about generative AI rises from 15% to 55% with strong leadership support, yet only about a quarter of frontline employees say they receive it.
What almost no organisation is doing is the fourth move: redesigning the work itself. The human conditions of work, meaning who decides what, how contribution stays visible, and how people make sense of change together, need redesigning in parallel with the AI. This is not a new insight. Trist and Emery at the Tavistock Institute established the principle in the 1950s: the technical subsystem cannot be optimised without co-designing the social subsystem. AI is simply the largest test of that principle in seventy years. MIT Sloan Management Review named it for this moment: a formal and ongoing process of human and AI organisational design is a must.
Change management manages transition. Sociotechnical design redesigns the work itself. One is reactive; the other is anticipatory. The organisations pulling ahead are not those with the most AI tools. They are those that have redesigned their operating models and leadership culture around what AI requires from people.
The upside. Garry Tan of Y Combinator on the productivity gap between a founder with a full AI agent workforce and one working alone, 2026 against 2013. Josh Bersin's term for what emerges is the superworker.
A directional illustration from practitioner sources, not a controlled study. The principle holds whatever the precise multiplier.
The downside. MIT meta-analysis in Nature Human Behaviour (2024): 106 studies, 370 effect sizes. Human and AI combinations performed worse than the better of the two alone. The cause: overreliance, as people anchor on AI output and stop thinking independently.
Hedges' g = -0.23, Vaccaro et al. (2024).
The 810 times is available to every organisation. The minus 0.23 is what happens when leaders do not redesign the human conditions of work alongside the technology.
The MIT evidence is precise. In decision-support contexts, which is exactly where most organisations are deploying AI, human and AI combinations underperform AI alone. People anchor to the machine's output, stop processing information independently, and the judgment the system needs from them atrophies. What looks like AI adoption can be human capability erosion.
Klarna's case maps this failure at organisational scale. The company did not lack AI capability. It lacked integration design. When the AI handled straightforward queries it performed well. When interactions needed judgment, empathy or context, the absence of a designed hand-off between human and AI produced failure. Klarna had replaced human capacity without redesigning human capability.
The counter-models share one characteristic: they treat AI integration as a continuous design discipline rather than a one-off implementation. DBS's PURE framework (Purposeful, Unsurprising, Explainable, Respectful) is a governance architecture, updated as AI use evolves. Nubank's founder David Velez is reported to have hired the first employees in each new market himself before the product team arrived, building the culture before the technology. Itau Unibanco's chief executive puts the philosophy directly: humans and technology must evolve side by side.
The new organisation: three roles that will matter
Al-Abdullah's model of the AI-era organisation identifies three roles. It is more useful than a job-title taxonomy because it describes function: what each role does, and why the organisation cannot perform without all three operating at once.
The Architect layer sets the vision for what the system needs to become. Architects know what to ask, how to frame it and what good looks like; they define what gets built and sign off when output matches intent. The chief executive is the chief architect, but the role exists wherever system design and judgment meet. The Assembly layer builds, refines and validates: builders, domain experts, engineers and compliance teams working with the architects until output matches intent. This is where much of the human-in-the-loop judgment lives. The Amplifier layer makes sure the work lands: sales, marketing, customer experience and distribution, feeding market signal back into the intelligence layer so the system learns.
The Integration FunctionArchitect, Assembly and Amplifier layers (Al-Abdullah, 2025) with the Integration Function and human-in-the-loop points (Ritter, 2026).Open the interactive modelWhat Al-Abdullah's model does not address is the connective tissue that holds the three layers together. Each layer is described independently; the model is silent on who manages the junctions. That connective tissue is the Integration Function: the ongoing practice of designing the social and technical systems to strengthen each other at every junction. In organisational terms it is the work of strategic people and culture leadership, the sociotechnical organisation design function the Architect model requires but does not name.
The chief executive's four irreducible responsibilities
In Manager and Founder modes, the chief executive personally owned five things: vision, culture, hiring, capital and cash. In Architect mode, Al-Abdullah argues these compress to three: Aim, Army and Assets. The compression is elegant and largely defensible, but something critical is lost. Culture disappears, on the assumption that it is embedded in the system design. It is not. Culture is either designed deliberately or it emerges by accident, and in an AI-augmented organisation, accidents compound quickly.
The fourth irreducible responsibility, the one that completes the model, is Atmosphere.
The four responsibilitiesAim, Army and Assets (Al-Abdullah, 2025), completed by Atmosphere (Ritter, 2026).Open the interactive modelAim is choosing the mountain. AI can improve the map, model the routes and identify obstacles; it cannot choose the direction. As intelligence becomes more abundant, conviction becomes scarcer. Army is hiring for the multiplier: one high-agency operator with AI fluency and sound judgment outperforms ten average performers running on the old model. Assets is allocating the portfolio of capital, technology, attention and conviction as a whole.
Atmosphere is creating the conditions in which both people and AI agents perform at their best: culture, trust, psychological safety, meaning and belonging, the felt experience of working in the organisation. It is what Klarna lost when it optimised headcount, and what the strongest counter-models protect by building culture before technology. Atmosphere is the one responsibility AI cannot touch, and it decides whether Aim is believed, whether Army is retained, and whether Assets are well spent.
The chief executive is the organisation's Chief Energy Officer. Attunement is the wave that carries through the atmosphere. Resonance, where wave meets readiness, is where human and AI performance compound.Ritter, 2026
The mechanism by which a chief executive creates and sustains Atmosphere is Attunement: the capacity to sense and respond to the human experience of the organisation as carefully as they read the intelligence layer. A well-attuned leader generates a signal that carries through all three layers. Where it meets readiness, it creates resonance. Where it meets anxiety, misalignment or unresolved threat, it fades, and the culture invoices them later.
Atmosphere and AttunementHigh attunement sustains the wave through every layer; low attunement collapses into noise. After Rock (2008); Ritter (2026).Open the interactive modelThe social brain inside the machine
The Architect argument is strategic and structural. It does not address what happens to the people inside the architecture while it is being built: how they experience the transition, and what the arrival of AI does to their sense of status, certainty, agency, belonging and fairness. These are neurological concerns as much as cultural ones, and they are exactly the signals Atmosphere and Attunement exist to address.
When AI changes the work faster than people can make sense of it, the first response is the social brain's threat detection. David Rock's SCARF model lets leaders diagnose which domains are under pressure and act before the costs show up in the data. In the AI rollouts this work draws on, Autonomy and Fairness come under pressure first: agency moves to the system before people have redefined their contribution, and the accountability gap for AI decisions raises acute questions of fairness.
SCARF applied to AI integrationFive threat domains to design for before disengagement shows in the data. After Rock, NeuroLeadership Journal (2008).Open the interactive modelStatus: when an agent handles tasks that once demonstrated someone's expertise, do they still know where their contribution matters? Certainty: do people know what is changing, what is not, and what their role looks like next year? Autonomy: when AI automates decision pathways, do people keep meaningful agency over their work? Relatedness: as work becomes more interface-heavy, are people losing connection with colleagues? Fairness: is AI introduced consistently, and who captures the gains while others carry the risk?
- Where is AI now making decisions that people used to make? (Autonomy: agency is being quietly transferred to the system.)
- Where is contribution becoming invisible? (Status and Fairness: when AI produces the output, who gets credit for the judgment?)
- Where are people making sense of change together, and where are they processing it alone? (Relatedness and Certainty: isolated sense-making compounds the threat.)
These three questions need no survey and no consultant. They need a conversation, and the willingness to hear an honest answer.
Where AI augments, and where it does not
Brynjolfsson, Li and Raymond's study of 5,179 customer support agents gives leaders the most precise evidence available on when AI augments human performance. The headline gain of 14% understates the strategic insight. Novice and lower-skilled workers gained 34%; experienced and highly skilled workers gained very little. AI spreads the tacit knowledge of the most capable workers to those with less experience, compressing the experience curve. But it only works when the organisation's design supports that transfer. Without complementary investment in role design, skills and workflow, the gains do not materialise.
This is the evidence behind the Architect insight that organisations should hire not for skills the AI can replicate, but for the human qualities that make AI more powerful: taste, conviction, judgment and the ability to specify intent clearly. The design challenge is to build the conditions where those qualities flourish alongside the AI rather than atrophy because of it.
The SHIFT framework
Five design disciplines complete the Architect model by giving equal rigour to the human integration layer most organisations leave to chance. Each addresses a specific failure in current AI transformation practice.
The SHIFT frameworkFive design disciplines for the human integration layer (Ritter, 2026).Open the interactive modelS: Structure the intelligence layer, not the hierarchy. Redesign roles around proximity to signal rather than position in the hierarchy. The intelligence layer, where data, models, customer inputs and operational feedback converge, sits at the centre. The new organisation chart is about signal flow, not reporting lines.
H: Handle AI agents as workforce participants, not tools. An agent joining a team is a workforce integration event. It needs a defined role with scope, accountability and escalation; explicit communication to the human team about what it does and where it stops; and performance measures that are reviewed and adjusted. Treat new agents like interns: supervised, monitored, graduated.
I: Invest in the conditions where human and AI outperform either alone. Human and AI combinations add value in creative, generative and judgment-intensive work; in decision support, the risk is overreliance. Design for complementarity: a clear line between what AI executes and what people decide, workflows that keep human judgment visible and consequential, and regular work without AI assistance to keep independent capability.
F: Frame the social brain impact before the culture invoices you. Use SCARF as a diagnostic to anticipate the threat responses AI integration triggers before they surface as disengagement, resistance or withdrawal. The chief executive's Atmosphere and Attunement work at organisational level; SCARF is the team-level diagnostic.
T: Trust through governance, not reassurance. Telling people AI is here to help rather than replace is not a governance strategy. Trust is built structurally: transparency about how models are used and what they decide, accountability that keeps people responsible for AI outcomes, and frameworks such as DBS's PURE model. Who holds the pen on how AI is configured is a question of power. Leaders who answer it explicitly will have organisations that reflect their values.
Eight recommendations for leaders
Four on building the Architect organisation. Four on integrating the human system. Together, they close the integration gap.
- Redesign roles around proximity to the intelligence layer, not position in the hierarchy. Audit how many roles are relay nodes that compress, filter and summarise information before it reaches a decision-maker, and how many generate direct signal. Collapse the relay architecture deliberately and invest in roles that produce insight.
- Write job descriptions for your AI agents. Every agent should have a defined role: what it does, what it explicitly does not decide, where it escalates to a person, and what good performance looks like. Start supervised and graduate autonomy as reliability is shown.
- Hire for the multiplier, and develop for the architect. Hire for taste, conviction, the ability to specify intent and the judgment to know when the machine is wrong. Build the capability to design systems, interpret signals and direct AI execution.
- Build your governance layer before you need it. Governance built after a failure is remediation; governance built before scale is architecture. DBS's PURE framework and Shneiderman's three levels (reliable systems, safety culture, independent oversight) are practical templates.
- Treat agent integration as a workforce onboarding event, not a technology rollout. Map the SCARF impact first, redesign the social architecture, then deploy. Before any significant deployment, hold a thirty-minute team conversation on two questions: what does this change about what each of us contributes, and what must we do together that this AI cannot?
- Diagnose SCARF before the culture invoices you. For each domain, ask how it is changing for your team and act before disengagement shows in the data. An organisation that gains AI efficiency while losing the trust of its people has not transformed; it has automated itself toward a different kind of fragility.
- Protect the conditions where human judgment is irreplaceable, and make them visible. As AI handles more of the what and the how, leaders must become more intentional custodians of the who and the why: meaning, belonging, trust and the experience of contributing something the machine cannot.
- Design the entanglement; do not try to resolve it. AI and organisations shape each other continuously. Build regular sense-making practices that help teams navigate that co-evolution, and decide explicitly who holds the pen on how AI is configured.
The shift is not about deploying more AI. It is about designing the human system to accelerate with it.Ritter, 2026
Sources
- Vaccaro, M., Almaatouq, A. and Malone, T. (2024). When combinations of humans and AI are useful: a systematic review and meta-analysis. Nature Human Behaviour, October 2024.
- Brynjolfsson, E., Li, D. and Raymond, L. (2025). Generative AI at work. Quarterly Journal of Economics, 140(2). NBER Working Paper 31161.
- Orlikowski, W. and Scott, S. (2025). Exploring AI-in-the-making. Information and Organization, 35(1).
- Shneiderman, B. (2022). Human-Centered AI. Oxford University Press.
- Rock, D. (2008). SCARF: a brain-based model for collaborating with and influencing others. NeuroLeadership Journal, 1(1).
- Trist, E. and Emery, F. Sociotechnical systems research, Tavistock Institute of Human Relations, 1950s and 1960s.
- MIT Sloan Management Review (2025). On human and AI organisational design.
- Al-Abdullah, A. (2025). Founder Mode Is Dead, and Architect Mode essays.
- Graham, P. (2024). Founder Mode. paulgraham.com.
- Boston Consulting Group (2025). AI at Work: momentum builds, but gaps remain.
- McKinsey & Company. From adoption to impact: three horizons of AI transformation.
- Bloomberg and other reporting (May 2025) on Klarna's customer service reversal and Sebastian Siemiatkowski's comments.
- Gartner (2025). Prediction on organisations abandoning plans to cut customer service staff because of AI.
- DBS Bank. PURE responsible AI framework; reporting on AI economic value (S$1 billion in 2025) and workforce plans.
- Harvard Business Review (2025 and 2026). Articles on agentic AI and the workforce, and on onboarding AI agents.
- World Economic Forum (January 2026). AI is becoming your new work colleague.
- Tan, G. and Bersin, J. (2026). The superworker thesis (practitioner sources).
About the author
Dennis Ritter is an organisational design and effectiveness consultant with a global practice across financial services, oil and gas, insurance and business process outsourcing. His work sits at the sociotechnical intersection of people, process and system design, with a live focus on AI and data-led enterprise transformation. He has led large-scale change before and after mergers and acquisitions, and holds executive qualifications from Stanford Graduate School of Business, INSEAD and the University of Michigan. Contact: dennis@sociotek.ai.
Where does your organisation sit?
The Sociotek Pulse gives a first reading of how ready your organisation is for the autonomy it plans to give AI, people side included. How the diagnostic works