There has been a lot of talk about the transformation recently announced by Moderna, which has led to the merger of HR and IT, but many questions remain unanswered regarding the operational implementation of this change, as the main idea is to reinvent work around the human/AI duo.
To sum up:
- Moderna has merged its HR and IT teams to create unified governance, led by a Chief People and Digital Technology Officer, in order to integrate people and technology at the heart of the organization (HR and IT merger: Moderna redesigns its organization for and with AI).
- HR is becoming the architect of skills and transformation, while IT is developing AI platforms and tools (more than 3,000 GPT agents) to automate and enrich all internal processes.
- Work is no longer organized into business silos but into flows: each task is orchestrated in real time between humans and AI, for an agile, fluid, and adaptable organization, where the best combination of man and machine takes precedence over traditional structures, without it being clear how this will materialize and what operations will become in all of this (HR/IT and the reality of working at Moderna: the unspoken truths of a reorganization).
In this article, I would like to return in more detail to the idea of thinking about work as a flow rather than as a sum of activities or, in other words, as a logic that cuts across silos rather than a sum of activities carried out within silos.
I had already noticed that Moderna’s inspiration, as evidenced by the wording used, is very industrial (People Centric Operations 2.0: how AI is reinventing knowledge work at scale), and this notion of flow is yet another example of this.
This is a real paradigm shift, but a paradigm shift means changes in indicators, and I find it very difficult to understand how Moderna will measure the performance and effectiveness of its new organization.
When I read “flow”, I immediately thought of Eliyahu Goldratt, who, in his theory of constraints, redefines business performance around a simple idea: what matters is not what you produce or your costs, but the speed at which you transform inputs into value delivered to the customer. It’s called throughput (a word that will test your pronunciation in English).
To say that costs don’t matter may seem totally unrealistic, but read “The Goal” and you’ll come away convinced.
A system can be perfectly economical and totally unproductive, and this is one of the biases of traditional management: we optimize costs, we pool resources, we eliminate redundancies, and sometimes we forget the essential, which is to know whether we really have the capacity to deliver what matters. In the interest of local efficiency, we create global friction and slow down a flow to reduce a load, without realizing that we are weakening the entire system.
What he tells us, and what is easy to verify, is that local optimization never guarantees global optimization. In other words, maximizing the performance of one link (or one team, one department, one tool) can harm the performance of the entire system.
Speaking of the increase in employees due to AI, I had already suggested that just because we were gaining productivity at the individual level did not mean that we were gaining at the end-to-end process level (AI in the workplace: going beyond augmentation to actually transform) what seems to be supported by a study conducted at Procter & Gamble, which shows a time saving of 16.4% for individuals compared to 12.7% for teams ([FR]How to revolutionize your teamwork with IAGen).
In other words, having people working at 100% capacity does not mean that you are efficient across the business or that you are making money. I’ll leave you to think about that, as I think we’ll be talking about it again in the future.
In Goldratt’s thinking, it is not a question of ignoring costs, but of viewing them not as objectives, but as means. What matters is the system’s ability to transform a need into delivered value, quickly and with as little resistance as possible. If an expense improves fluidity, it is useful; if a saving slows down the flow, it is harmful. The criterion is no longer budgetary, it is systemic.
I think you understand better what I mean when I talk about indicators.
In industry, this has led to profound changes: gone are local optimization, machine overload, and obsession with utilization rates. Instead, we now see global orchestration, bottleneck identification, and fluidity of flow. Performance is a systemic issue.
But that’s all well and good in a factory, but does it work in an office and for knowledge workers? Because that is the hidden challenge behind Moderna’s good intentions.
Too often, we still only measure what is visible: time spent, tasks completed, meetings held, deliverables provided. In this context, throughput is the ability to efficiently circulate information, decisions, and contributions until they have their final impact. In other words, what matters is not what you do, but what your system enables you to accomplish.
Moderna’s idea of thinking of work as a flow is very appealing when it comes to physical flows, but won’t it clash with the reality of intangible flows, leading to the consequences I have just outlined?
In short:
- Moderna has merged its HR and IT teams to establish unified governance focused on integrating AI and people, with the aim of reorganizing work from silos into dynamic flows.
- The flow logic, derived in particular from constraint theory, redefines performance at the system level by valuing fluidity (throughput) rather than isolated efforts or costs.
- Applying this logic to knowledge work is counterintuitive, as it requires abandoning traditional benchmarks (job roles, tasks, individual productivity) in favor of systemic coordination that is difficult to visualize.
- The management of this model is hampered by the lack of suitable native indicators: new measurement tools need to be invented, such as decision-making lead time, friction rate, and resolution throughput, which are still not widely used or proven.
- AI can facilitate this real-time orchestration, but its effectiveness depends on clear governance, human support, and constant vigilance against the risks of dehumanization, rigidity, or loss of meaning.
Applying throughput to knowledge work
Transposing this logic to knowledge work means moving away from a business-centric approach, where individuals are managed individually and by task, to a flow-oriented vision.
This involves a number of things, including:
- Identifying invisible bottlenecks (Eliyahu Goldratt’s fictional interview on infobesity and bottlenecks in knowledge work): slow validation, poorly integrated tools, unclear prioritization, overworked or cognitively overwhelmed people, and bodies whose meeting and decision-making pace is slower than the pace of other people’s work.
- Implement work orchestration rather than rigid planning or overlapping reporting. Rigid planning ignores the vagaries of everyday life and freezes responsibilities, while reporting gives the illusion of control without improving the ability to act. Orchestration, on the other hand, allows actions to be adjusted to the reality of workflows.
- Manage not the effort but the speed and quality of action.
Does this make sense? Certainly. Does it seem intuitively obvious to implement? Perhaps for you, but not for me, even though I would like to see this type of practice adopted.
In this context, AI can help and play a structuring role in accelerating throughput. It does not just distribute tasks: it learns from existing flows, anticipates slowdowns, and suggests adjustments. It also enables scalability that would otherwise be impossible: when information becomes too abundant, when decisions are too complex or too fast to be made by humans, it helps filter, prioritize, and route what matters to the right place.
This is exactly what Moderna does, or what we guess they have in mind: rather than optimizing what already exists, the company is reprogramming its operations around flows. It is not digitizing what already exists, but rebuilding a system.
What indicators should be used to manage knowledge work?
But managing workflows based on throughput requires changing the indicators, and that’s where things get complicated. Here are a few ideas that might make sense.
Decision lead time: the time between identifying a need and making a decision. This is not intended to judge an individual’s speed, but rather the fluidity of the system.
Cycle time for “useful” deliverables: the time between the start of an action and its actual delivery, use, with a measurable effect. I emphasize “useful” because I remain convinced that we produce a huge amount of deliverables and documents that serve absolutely no purpose (Is creating documents really work?).
Reuse rate: the frequency with which a decision, deliverable, or solution is reused elsewhere. This is a good indicator of its real value and ties in with the point I made earlier.
Number of autonomous decisions: an indicator of teams’ real ability to make decisions without having to escalate or wait for validation. This is a sign of organizational agility and a good way to measure how well work orchestration promotes responsiveness and local accountability.
Friction rate: interruptions, follow-ups, misinterpretations, time spent searching for information or waiting for a decision, conflicting priorities between two people or two departments, poor coordination, overlapping tasks or roles… anything that slows down the flow of information and the smooth running of work.
Resolution throughput: problems closed with real impact, versus total volume of problems raised… it measures an organization’s ability to solve problems or finalize useful issues over a given period.
These indicators do not measure activity but the system’s ability to deliver, not effort but impact.
I would add the concept of stock and work in progress, which also plays a role in the original vision of the concept and which, in knowledge work, translates into everything related to information and cognitive overload, i.e., information that we don’t have time to process or that clutters our brains, to-do lists that pile up, emails to be dealt with, upcoming meetings, decisions to be made, etc. (Digital Infobesity: When Collaboration Tools Degrade Productivity, QWL and Amplify Mental Workload and Hyperconnectivity in the workplace: digital becomes a burden).
Does this sound appealing to you? Does it make sense? I think so, and I believe that lean management and even agility enthusiasts will find something to relate to here.
But do these indicators exist today? For the most part, no. Are they easy to implement? In my opinion, not at all, even if it should be possible in the modern work environment (The quantified organization: Grail or Big Brother?).
I don’t think we should underestimate the cultural and even mental gap that will need to be bridged. We have always measured quantities and stocks, but now we will have to measure movements, which means a starting point, a destination, and the time spent at intermediate points, as well as analyzing the route taken.
Can AI really improve flow work?
In Moderna’s vision, AI is truly central, and if we start from the premise that thinking about flow work is sensible but very complicated given the complexity of measuring relevant signals to manage it, we must ask ourselves what role AI can play in making this vision operational.
AI, as the orchestrator of knowledge work, could detect weak signals in activity flows, suggest sequence optimizations, link fragmented tasks, detect bottlenecks, allocate resources, and adjust pace and deadlines accordingly. We could therefore expect it to transform a fragmented and chaotic work environment into a more fluid mechanism, where each actor, human or machine, intervenes at the right time, in the right place, with the right information. It does not set a uniform pace but modulates the rhythm according to the needs and constraints of the moment.
In a flow-oriented system, AI can therefore play several key roles:
- Dynamic flow mapping: by analyzing interactions, delays, and validation loops, it helps visualize areas of slowdown.
- Friction reduction: by automating certain repetitive steps or facilitating access to contextual information, it frees up cognitive time.
- Adaptive orchestration: it can reallocate tasks or recommend adjustments based on workloads, deadlines, or dependencies in real time.
- Alert or prioritization signals: By monitoring flows, AI can identify when a task is deviating from its objective or slowing down the flow, suggest trade-offs, or alert managers before bottlenecks become blocking issues.
In this sense, AI is not a gadget but a mechanism for amplifying collective intelligence to optimize work, capable of providing the organization with an overview that no single player can have.
It does not automate knowledge work but optimizes its coordination, and this is where throughput acceleration comes into play.
Risks that should not be underestimated
But this flow-oriented, AI-driven transformation is not without its pitfalls. It can even produce the opposite of the desired effect if implemented blindly. Let’s not forget that if few businesses have successfully implemented Lean (we’re talking about similar issues here), it’s not because Lean doesn’t work, but because it’s not a generic approach that can be applied to any organization; it has to be adapted to each one.
Among the risks, I would mention the following:
- Automating inefficiency: without questioning processes, AI risks making what is already dysfunctional faster. When technology is applied to a dysfunctional organization, it will simply function faster and on a larger scale. And let’s not forget that if we don’t ensure that this is not the case, AI will only learn from the past…
- Reinforcing asymmetry in control: by concentrating the global vision in the machine or in the hands of a few decision-makers, we weaken local autonomy, which is the opposite of what we are seeking or of a People-Centric Operations approach (People Centric Operations 2.0: how AI is reinventing knowledge work at scale).
- Creating information overload: if poorly configured, AI can generate more alerts than it resolves, or make the system more difficult to understand.
- Making iteration more difficult: an overly automated system can become rigid, when it should be continuously adapting.
- Dehumanizing work: by fragmenting tasks and accelerating the pace of work, AI can make the individuals behind the flows invisible, reducing them to resources that are activated on demand.
- Subjugating employees to machines: when AI recommendations become prescriptions without discussion, we slide from decision support to delegation of judgment.
- Making work meaningless: by making contribution chains more opaque, automated orchestration can distance everyone from the real impact of their work.
- Losing sight of the ultimate goal: by breaking work down into micro-tasks and managing it through local signals, we risk obscuring the collective purpose, the “big picture” that gives coherence and motivation to individual action.
This is why organizational design, listening to the field, and co-constructing flows with teams remain essential. AI does not replace governance—quite the contrary—but it does force us to rethink it.
That said, if all the challenges are met, Moderna may be the laboratory that will test many hypotheses on improving knowledge work inspired by its half-tech, half-industrial DNA (Just because work is invisible, it doesn’t mean that it can’t be improved).
Bottom line
Orchestrating work is not the same as organizing it
Thinking about work in terms of throughput means stopping optimizing locally to improve globally, making the value chain more transparent, and rethinking what it means to “work well” in a complex environment. But this requires more than just changing indicators or tools: it requires a clear, shared vision of what we are trying to produce together.
The orchestration of work must not become a blind mechanism. To be successful, it must remain clear, human, and governed. AI can be a powerful lever, provided it is integrated as a decision-making support, not as an autonomous logic, and must neither obscure the collective goal nor fragment the meaning of the action.
Moderna shows that this path is possible, drawing inspiration, perhaps unconsciously, from industrial logic to structure the organization of knowledge. I am even deeply convinced that one day someone must have said, “We should be as efficient in offices and laboratories as in the factories where vaccines are produced,” and that everything may have started from there.
But this transformation is only sustainable if it remains focused on people, their capacity for action and judgment, and their understanding of why. Flow is a means, not an end, and AI is an ally, not an authority.
I dare not think that before talking about rethinking work in flow, Moderna did not consider all that this implied. Not only is it completely counterintuitive at first glance, but there are also no “native” indicators available to guide it. In any case, if they fail on the concept of flow, their AI project will lose all meaning.
