Large organisations have become stymied by the ceremony of organisational decision-making – agentic AI is about to change all that
Throughout time, hierarchy in some form or other has existed in societies, social groups and business. You can argue whether it is a good thing or a bad thing, but it has survived this long because it has solved fundamental issues around co-ordination, particularly in large organisational structures trying to make decisions at scale. A major concern is that over time it has become increasingly complex. Layers of decision-making being added to the point where much of a hierarchical structure is ceremonial and only making marginal contributions to the quality of decisions made. It also adds layers of abstraction between intention and implementation of tasks which can create misunderstanding, prompt delays and dilute the impact of a project.
What if the arrival of agentic AI could change things by removing unproductive layers in decision-making and encourage new organisational structures?
Let’s consider what hierarchy enables organisations to do. There are five drivers of hierarchy. Three are operational:
- Serialisation: manager passes on instructions to the doer
- Risk distribution: at each level someone owns the outcome if it goes wrong
- Information compression: at each level an individual summarises information for the next level up
And two are social and institutional:
- Co-ordination under genuine uncertainty: an individual provides direction when a situation is unclear
- Political legitimacy: a recognised organisational structure means decisions are given legitimacy
If your team cannot self-align around these drivers to solve problems, then hierarchy provides formal structure to enable decision-making. The problem is that over time layers have been added which border on the ceremonial creating unnecessary complexity.
It is questionable that the ceremony of passing decisions through a hierarchy produces better outcomes. Tasks may acquire authorisation as they travel through nodes of approval, but how much does each stage add to the quality of the decision-making?
What if you changed your approach and moved away from an over-reliance on hierarchy to get things done? In an organisation infused with AI Agents co-ordination between teams can be designed by the specification of these Agents, particularly what information they share with each other and how they communicate at points of connection, almost like a contract defining the scope of their interactions. However, for this approach to work there are several considerations.
1. The rise of the roadblock remover
Unit4’s engineering team conducted an experiment which could be just as relevant for other functions and lines of business examining how decisions might be made if agentic AI assumed responsibility for parts of the decision-making in current hierarchical structures.
We created small teams of three to five engineers and gave them tasks. They did not have a manager, but a designated roadblock remover if the teams got stuck.
As teams looked to solve problems they stalled, not because they lacked authority to complete tasks, but because they couldn’t find the information they needed for solutions. It was a problem of indexing, a role that managers have clearly played acting as a navigation function or human index. Teams struggled to self-navigate to find information because the organisation had never required them to. The muscle memory of the organisation was to rely on a manager to find the answer.
What the engineers found is that the teams who could overcome this issue got beyond this blockage.
The roadblock remover function helps to solve this indexing issue, and also another one, judgement and legitimacy. Those working in these small teams found that when they had tensions with other teams or had a decision that required the approval of multiple teams across the organisation having a roadblock remover was useful.
2. What responsibilities could AI Agents take on?
There are three distinct roles that this roadblock remover function is playing: memory retrieval, operational clearing, and judgment at the boundary. What if AI could take on some or all of these responsibilities?
If you think about the role that AI can play, it can codify and learn memory retrieval holding institutional memory, such as deployment rules, exit gate criteria and process descriptions. All of it can be encoded, and once encoded, retrieved without a human in the path.
The second function, operational clearing, can also be transformed with AI but there is still a need for some oversight. Operational clearing or programme management performs tasks such as tracking what teams are building, sequencing the work, surfacing dependencies, compressing status upward. Many of these tasks are specification problems, but not all of them. Some coordination overhead exists because information is missing or badly encoded and goals may conflict, priorities shift, or resources are constrained in ways no specification resolves in advance.
By being clear on specification criteria, AI Agents can be used to ensure information is not missing or badly encoded significantly reducing the need for human oversight. If the interfaces between teams are clearly defined in advance, AI Agents can help to reduce conflicts, manage shifting priorities or deal with constrained resources, but they do not eliminate the need for human oversight. Equally, organisations should guard against over-reliance on AI Agents as there is a danger they automate away the connectivity between teams. So just as previous technology inflection points had seismic (positive and negative) impacts on organisational structures and economic opportunities, so too could a world with agentic AI operating in the background.
3. Getting the dependencies between agents right
The priority becomes ensuring that the AI Agents are set up correctly, which requires the skills of the builder and increases the importance of inquiry or critical thinking. The interfaces must be identified before execution begins. If two teams have a dependency that is not in the specification, no agent coordination resolves it. Agentic AI-infused organisational dynamics will only succeed if organisations understand that this is not a simple planning exercise in the sense of Gantt charts and resource allocation. It must be treated as an exercise in defining the architectural specification of how teams work together: where the boundaries fall, what the contracts say, which boundaries are owned by whom.
It is not an exercise in programme management. This approach to specification design decides what is possible for AI Agents to do in support of teams and will reduce or even remove some layers of the ceremony involved in traditional organisational decision making.
4. Don’t lose sight of the need for human judgement
Human involvement will change as AI Agents take on more responsibility. In an organisation supported by agentic AI, management takes on a new form because AI Agents will require a pool of human judgement to navigate situations that have not been defined by preset specifications. It could be a corner case the specification did not anticipate or an impasse between two teams that requires a human to establish a new truth that is added to the specification. This will still require a level of political legitimacy to ensure consensus on any changes or decisions made, but it should reduce the volume of such decisions if specifications are set out accurately.
What is clear is that agentic AI is dismantling organisational structures designed to enable decisions to be routed between teams. It does not mean there will be less organisation, just that it will be expressed differently. The organisations that win in this era will break down unnecessary ceremony and co-ordination to build more streamlined, dynamic operations making decisions at speed. And for individual employees this will mean they are no longer abstracted from the tasks they want to complete. They can be hands on, leveraging AI Agents to complete tasks. This will give them a far greater appreciation of the problems they are looking to solve. It will also speed up the feedback loop between idea and creation, which will reduce the cost of experimentation encouraging more teams to explore and play more. Fundamentally, it collapses the distance between the expert with the most relevant knowledge and the solution for a particular task or problem. That can only be a good thing for the future dynamics of organisations.
—-
Claus Jepsen, Chief Technology Officer, Unit4
Claus Jepsen is a technology expert who has been fascinated by the micro-computer revolution ever since he received a Tandy TRS model 1 at the age of 14. Since then, Claus has spent the last few decades developing and architecting software solutions, most recently at Unit4, where he is the Chief Technology Officer leading the ERP vendor’s focus on enabling the post-modern enterprise. At Unit4, Claus is building cloud-based, super-scalable solutions and bringing innovative technologies such as AI, chatbots, and predictive analytics to ERP. Claus is a strong believer that having access to vast amounts of data allows us to construct better, non-invasive and pervasive solutions to improve our experiences, relieve us from tedious chores, and allow us focus on what we as individuals really love doing.
About Unit4
Unit4’s next-generation enterprise resource planning (ERP) solutions power many of the world’s mid-market organisations, bringing together the capabilities of Financials, Procurement, Project Management, HR, and FP&A to share real-time information, and deliver greater insights to help organisations become more effective. By combining our mid-market expertise with a relentless focus on people, we’ve built flexible solutions to meet customers’ unique and changing needs. Unit4 serves more than 4,700 customers globally across a number of sectors including professional services, nonprofit and public sector, with customers including Southampton City Council, Metro Vancouver, Buro Happold, Devoteam, Norwegian Refugee Council, Global Green Growth Institute and Oxfam America. For further information visit www.unit4.com.
For more information, please visit https://www.unit4.com/, follow us on, Facebook: Unit4 Business Software, Instagram: @unit4global or visit our YouTube: Unit4 and LinkedIn page
