Agentic AI is reshaping enterprise workflows by introducing autonomous, decision-making systems that collaborate like human colleagues. But rather than displacing jobs, these digital agents may be the catalyst for unlocking untapped human potential.
There is a quiet revolution taking place in the architecture of enterprise software. Not one driven by headline-grabbing demos or corporate vision statements, but in the complex practicalities of integrating legacy infrastructure, mismatched data formats and overworked employees. At the centre of this movement is agentic AI, intelligent systems that do not just automate, but reason, adapt and collaborate.
For Dr Jan Kunkler, Principal Data Scientist at Lobster, this shift is about far more than efficiency. It is a reimagining of how organisations design digital workflows and structure their teams. The goal is not simply to reduce human labour but to reallocate it, to give people back the time and space needed to think, improve and innovate.
Not just automation, but autonomy
Traditional automation, even in its so-called ‘intelligent’ form, has always been bound by rules. It is deterministic: a process mapped end-to-end by human operators with rigid pathways and pre-coded conditions. That rigidity becomes its weakness when complexity or variability enters the picture.
Agentic systems introduce a layer of non-determinism. Instead of blindly following instructions, these systems interpret goals, evaluate context and make choices. In Kunkler’s view, they are best understood not as tools but as collaborators. “What we now have,” he says, “is an intelligent operator that can detect edge cases, handle the unexpected and even decide when escalation is necessary. That ability to respond flexibly transforms how enterprises can structure their workflows, particularly in areas like logistics, where exceptions are the norm rather than the outlier.”
The case for digital headcount
It is no coincidence that much of the current discourse around AI centres on cost-cutting and job displacement. But Kunkler takes a contrarian stance. Rather than shrinking headcount, agentic AI has the potential to unlock new growth by compensating for the chronic shortfall in digital skills. “Many companies know they need to do more with data, but they simply do not have the people,” he says. “Agentic AI offers the chance to build out a team of digital collaborators.”
The value lies not in replacing existing roles, but in absorbing the low-value, high-friction work that prevents teams from innovating. Freed from the grind of documentation, repetitive integration tasks or process routing, employees can turn their attention to strategy and continuous improvement. This, in turn, opens the door to organisational growth, not despite AI, but because of it.
A striking illustration of this approach was presented at the Lobster Data Hero Summit in May, where Lobster and its Intelligence Partner, OneThousand AI, showcased a live deployment with HEROSE, a German manufacturing customer. The use case demonstrated how agentic workflows and Lobster’s orchestration platform could be combined to integrate legacy systems, interpret complex data, and automate tasks across the supply chain. HEROSE’s digital transformation was not about removing human workers, but enabling them to do more, better.
Architecture that scales with purpose
Beneath the agentic layer lies a modular infrastructure, purpose-built for scale. At Lobster, this takes the form of microservices, a flexible design pattern in which each function is containerised and independently deployable. It is the opposite of monolithic software: instead of rigid dependencies, every agent, tool or integration module is loosely coupled and orchestrated in real time.
This approach is not simply theoretical. One of the most heavily used services on Lobster’s platform today is a documentation agent that interprets XML integration profiles and automatically generates markdown-based technical documentation. It is a microservice, exposed via API, which can be scaled horizontally depending on demand. If usage spikes, additional containers are spun up automatically, ensuring performance and reliability. More importantly, the same infrastructure is used to compose agentic workflows, each agent operating with a clear scope, and tools exposed through a defined interface.
The ability to flex services based on real-time needs, whether driven by peak transaction loads or an AI orchestrator deciding on the best tool for the job, is essential in high-volume domains like logistics and manufacturing. The HEROSE example, built atop this microservices foundation, illustrates how AI services can be introduced in a modular fashion, without disrupting core operations.
Semantics unlock the data flywheel
Yet none of this works without data that is high-quality, structured, and semantically aligned. That final word is crucial. Most enterprise data is not inherently meaningful outside of its original context. Column names, abbreviations and data types may reflect decades of departmental habits or legacy constraints. What makes agentic systems work is not just format consistency, but shared understanding.
Kunkler is unambiguous on this point. “We see a lot of companies coming to us with the assumption that having a legacy system full of data is the same as being AI-ready,” he says. “But unless that data is made intelligible, you are feeding the system noise.”
At Lobster, that challenge is met by a semantic layer built over two decades of supply chain integrations. By mapping synonyms, variants and proprietary terminology onto a canonical data model, the company enables meaningful interoperability, for example, turning ‘FRA’, ‘Fraport’, and ’Frankfurt Airport’ into one shared concept. This foundation allows LLMs and agents to act with contextual intelligence rather than statistical guesswork.
It also fuels what Kunkler calls the data flywheel. Every interaction with an agent, every prompt, correction or escalation, creates a new source of signal. Capturing and reintegrating that feedback is essential for iteratively improving the system. Yet many enterprises fail to build the loop, treating AI as a static service rather than an evolving participant in the organisational knowledge graph.
The intelligence ecosystem emerges
What Lobster and partners like OneThousand AI are building goes beyond point solutions. It is an ecosystem of interoperable intelligence, where AI agents, integration services and third-party tools can be orchestrated in concert, layered onto legacy infrastructure and deployed incrementally.
This model solves one of the most persistent barriers to enterprise AI: accessibility. Many organisations lack the internal expertise to develop bespoke AI applications or the resources to rip and replace existing systems. An intelligence ecosystem offers a middle ground, abstracting away the complexity of integration and enabling plug-and-play functionality with trusted partners.
Kunkler’s keynote at the Lobster Data Hero Summit reinforced this vision. Data, he argued, must be treated as infrastructure: structured, standardised and governed. Only then can organisations take advantage of the intelligence layer that sits on top. The HEROSE deployment made this concrete. By leveraging the semantic data layer, deploying agentic tools from the Lobster Marketplace, and coordinating processes through orchestrated microservices, HEROSE successfully introduced AI capabilities without altering its core systems.
The evolving role of data scientists
As agentic systems grow in capability, one question looms: What happens to the role of the human expert? For Kunkler, the answer is a shift in focus. The future of data science is not in writing code or building models, but in orchestrating intelligence across systems, human and digital alike.
“It becomes a cross-functional role,” Kunkler says. “You are the architect, the translator, the educator. You need to understand the research, the engineering, the business objectives and how to communicate all of that up and down the hierarchy.”
In this model, the data scientist is no longer a technical specialist working in isolation. They are a bridge, curating the components of a digital workforce, choosing which agents to deploy, which data sources to tap and which guardrails to implement. It is a role grounded in systems thinking, not syntax.
And while LLMs and co-pilots can now write proof-of-concept code in seconds, the judgment required to decide what should be built, and why, remains human. “The current models are already powerful,” Kunkler concludes. “The bottleneck is not the technology. It is our ability to imagine how to use it.”
The trajectory is clear. Agentic AI is not an existential threat to enterprise teams, but a new set of capabilities waiting to be woven into the fabric of business. The job of leaders now is not to resist that integration, but to design for it, to build the infrastructure, governance and organisational culture that allows human and machine intelligence to co-evolve. The fundamental transformation lies not in the code, but in the conversations it enables.
