Agentic Workflows and the Death of the Approval Chain
The approval chain did not die because technology killed it — it died because it was already dead, and agentic workflows simply made the corpse visible.
Executive Summary
The sequential approval chain — the governance mechanism by which decisions move upward through layers of authority before action can proceed — has been the bedrock of organisational change management for decades. It exists because organisations need to control risk, ensure alignment, and maintain accountability when undertaking transformation. But a pattern is now emerging across sectors that calls the entire model into question: agentic workflows are completing work faster than approval chains can authorise it, and rather than adapting their governance, organisations are building elaborate workarounds that preserve the appearance of control while the substance of it erodes.
This essay explores why the approval chain persists despite mounting evidence of its dysfunction, what structural forces sustain it, and what the collision between agentic workflows and traditional governance reveals about the deeper gap between transformation intent and transformation reality.
The Architecture of Permission
Every large organisation runs on approval chains. A change request is raised, reviewed by a line manager, escalated to a change advisory board, assessed against risk criteria, approved or deferred, and finally released for implementation. The process exists for sound reasons: it creates checkpoints where informed judgement can be applied, it distributes accountability across the hierarchy, and it produces an audit trail that regulators, auditors, and senior leaders can inspect.
The problem is not that approval chains exist. The problem is that they were designed for a cadence of change that no longer reflects how work moves through an organisation.
When a transformation programme operated on monthly release cycles, a weekly change advisory board was a reasonable governance mechanism. When changes were large, discrete, and human-authored, sequential review by multiple stakeholders added genuine value. Each approver brought contextual knowledge that the requester might lack. The time spent waiting for approval was a reasonable trade against the risk of an uninformed decision.
But the cadence has shifted. In the organisations now deploying agentic workflows — systems in which AI agents initiate, execute, and in some cases complete work autonomously — the unit of change is no longer a monthly release or a weekly batch. It is a continuous stream. An agent identifying a process inefficiency, drafting a remediation, testing it against defined parameters, and flagging it for deployment does not wait for Thursday’s change board. It operates in minutes, sometimes seconds. And the approval chain, designed for a world measured in weeks, has no mechanism to keep pace.
The Workaround Economy
What happens when governance cannot keep pace with operations? In theory, either governance adapts or operations slow down. In practice, neither occurs. Instead, organisations develop workarounds — informal mechanisms that allow work to proceed while maintaining the formal appearance of control.
The pattern I have observed across multiple sectors is remarkably consistent. First, a category of “pre-approved” changes emerges — a class of actions that agents can take without entering the approval chain at all. The criteria for pre-approval are initially narrow and carefully defined. Over time, they widen, because every expansion of the pre-approved category reduces the bottleneck at the approval board. Within twelve months of initial deployment, organisations routinely find that seventy or eighty percent of agent-initiated changes are classified as pre-approved, flowing through the system without any governance touchpoint.
Second, the approval process itself becomes retrospective rather than prospective. Changes are implemented and then reviewed, rather than reviewed and then implemented. The language shifts subtly: “approval” becomes “ratification,” and the change advisory board’s role shifts from gatekeeper to auditor. This is not inherently wrong — retrospective review can be a valid governance model — but it is rarely acknowledged as the fundamental shift it represents. The organisation continues to describe its governance as approval-based while operating a ratification model.
Third, and most consequentially, the human reviewers in the chain begin to rely on agent-generated summaries of the very changes they are supposedly governing. The agent raises a change, generates a risk assessment, produces an impact analysis, and presents a recommendation. The human approver reviews the agent’s output rather than the underlying change. The approval chain has not adapted to the agentic world — it has been absorbed by it.
The approval chain did not die because technology killed it — it died because it was already dead, and agentic workflows simply made the corpse visible.
Why the Chain Persists
If the approval chain is demonstrably failing to govern the work it was designed to control, why does it persist? The answer lies in a set of structural forces that operate independently of whether the mechanism is effective.
Regulatory expectation. In financial services, healthcare, and public sector organisations, regulators expect to see evidence of human decision-making in change governance. The approval chain produces this evidence — names, dates, sign-offs — regardless of whether the underlying decision was genuinely human-directed. Dismantling the chain would require a conversation with regulators about alternative assurance models, and few organisations are willing to initiate that conversation while the existing model, however hollow, satisfies the audit requirement.
Accountability distribution. The approval chain serves a social function as much as a governance one. It distributes accountability across multiple individuals, so that when a change goes wrong, no single person bears the full weight. This is not cynicism — it is a rational response to the way organisations treat failure. In cultures where individual accountability for a bad outcome is career-threatening, the approval chain provides a collective shield. Replacing it with a model that concentrates accountability on fewer individuals — or on the designers of agent parameters — meets deep institutional resistance.
The illusion of control. Senior leaders who sit atop approval chains derive comfort from the belief that nothing significant happens without their knowledge and consent. This was always partly illusory — the volume of changes in any large organisation has long exceeded any individual’s capacity for meaningful review — but the illusion served a purpose. It provided a narrative of control that satisfied boards, investors, and regulators alike. Acknowledging that agentic workflows have rendered this narrative untenable requires a degree of institutional honesty that most leadership teams have not yet summoned.
The Transformation Gap
What makes this pattern particularly revealing is what it tells us about the gap between transformation intent and transformation reality. Organisations investing in agentic workflows are, by definition, committed to a particular vision of the future: one in which AI augments human capability, accelerates delivery, and reduces the friction of organisational processes. The approval chain is precisely the kind of friction these investments are designed to eliminate.
Yet the same organisations that are investing millions in agentic capability are simultaneously preserving — and in some cases reinforcing — the governance mechanisms that agentic workflows render obsolete. They are transforming their operations while leaving their governance untouched, as though the two could evolve independently.
This is the transformation gap in its most concentrated form: the distance between what an organisation says it is becoming and what its governance structures reveal it still is. An organisation that deploys agents to execute work at machine speed while governing that work through a weekly committee operating at human speed has not transformed. It has automated one layer while preserving the bottleneck above it.
“The organisations that will navigate this transition successfully are not those with the most sophisticated agents. They are those willing to redesign governance at the same pace they are redesigning operations.”
What Would Genuine Adaptation Look Like
If the approval chain is dying — and it is — what should replace it? The honest answer is that the profession has not yet converged on a model, but the outlines are becoming visible.
Continuous assurance over sequential approval. Rather than requiring human sign-off before action, governance shifts to continuous monitoring of outcomes against defined parameters. The emphasis moves from “was this approved?” to “is this within tolerance?” This is a fundamentally different governance philosophy, and it demands different skills, different tooling, and different accountability structures.
Exception-based human intervention. Humans remain in the governance chain, but their role shifts from reviewing every change to investigating exceptions. When an agent’s action falls outside defined parameters, triggers an anomaly in downstream systems, or produces an outcome that deviates from expected patterns, a human reviews. This preserves human judgement for the moments where it adds genuine value — the ambiguous, the unprecedented, the politically sensitive — rather than consuming it on routine approvals.
Parameter governance as the new approval. If agents operate within parameters, then the act of setting, reviewing, and adjusting those parameters becomes the primary governance activity. This is a significant shift: it means that the most consequential governance decisions happen not at the point of action but at the point of configuration. The people who define agent parameters are, in effect, pre-approving every decision those parameters will produce. The governance framework must recognise this and ensure that parameter-setting is subject to the rigour, scrutiny, and accountability that currently attaches to individual approvals.
Retrospective accountability over prospective permission. Acceptance that in a world of continuous, agent-driven change, some governance will necessarily be retrospective. The critical design question is not whether retrospective review is acceptable — it already is, in practice, whether organisations acknowledge it or not — but how to make it effective. This means investing in the analytical capability to review agent decisions at scale, identify patterns of drift, and intervene before cumulative small deviations produce a systemic problem.
The Honest Conversation
The death of the approval chain is not a technology story. It is a governance story, an accountability story, and ultimately a story about whether the professions that manage organisational change are willing to change themselves.
The agentic workflow did not kill the approval chain. It revealed that the approval chain had already become a ritual — a set of gestures that organisations performed because the gestures were expected, not because they were effective. The question now is whether the profession can design governance mechanisms that are genuinely fit for a world in which much of the consequential work happens at machine speed, or whether it will continue to perform the rituals while the real governance — such as it is — happens elsewhere.
The organisations that get this right will not be those with the best technology. They will be those with the institutional courage to admit that their existing governance is theatre, and the professional discipline to design something better. That conversation is overdue. The approval chain is already dead. The only question is how long we will continue to pretend otherwise.