IT Tech Pulse Exclusive Interview with Dr Thomas Di Giacomo, Chief Technology and Product Officer at SUSE

Stay updated with us

IT Tech Pulse Exclusive Interview with Dr Thomas Di Giacomo, Chief Technology and Product Officer at SUSE
🕧 20 min

Dr. Thomas Di Giacomo is Chief Technology and Product Officer at SUSE,, discusses agentic infrastructure, open source, AI sovereignty, identity and access management, enterprise portability, autonomous agent economics, and the foundations needed to keep AI infrastructure open and governable.


Thomas, you moved from networked multimedia research to hospitality IPTV, then SUSE’s CTO seat in 2016. Which technology best lesson from that path matters most now that infrastructure is becoming agentic?

Bet on the interfaces, not the components. In research, I worked on multimedia standards such as MPEG, and the lesson there was that the codecs changed constantly, but the standards that let systems talk to each other lasted for decades. At Swisscom we ran IPTV and connectivity for hotels, and that taught me the second half of the lesson: a technology is only as good as your ability to operate it at three in the morning when a guest’s WiFi or screen goes black.

Agentic infrastructure needs both lessons at once. The models and AI tech will keep changing, probably faster than any technology we have seen. Organisations that tie themselves to one model, one agent framework, one cloud or one system will have to rebuild every twelve months if not faster. Those that bet on open interfaces, composable architectures, such as Kubernetes for workloads and open protocols for how agents reach tools and data and each other, they can swap components as better ones arrive, they can adapt. And they need to run it all with the same discipline as any production system: observable, patched and secured, governed and supportable.

Your SUSE anniversary reflection pairs openness with pragmatism, citing the 2006 Microsoft interoperability agreement. How do you decide when compromise serves customers and when it erodes open source principles?

I ask two questions. Does this give customers more choice, and does it keep the code open? If both answers are yes, pragmatism is serving openness. If either is no, it is eroding it.

The 2006 agreement passed that test. Our customers were running mixed estates of Linux and Windows, and interoperability made their lives easier without closing a single line of our code. It was controversial at the time, but it put the user’s reality ahead of ideology, which is what open source is also meant to do.

Compromise erodes principles when it creates lock-in through the back door: when ‘open’ becomes a funnel into a proprietary layer, when licence changes pull the rug from under a community, or when a vendor contributes upstream in name only. Those moves may look pragmatic for a quarter, but they destroy the trust that makes open source work. After 34+ years, our view is simple. Customers should always be free to leave us. That is also why so many of them stay.

SUSE’s OpenShell integration runs agents inside enterprise policy. When AI acts rather than answers, what does sovereignty demand beyond data residency, and why is it no longer optional?

Data residency answers the question ‘where is my data?’ Agentic AI adds harder questions. Who is running this agent? Which model(s) is it using? Which systems can it touch? What did it just do, and can it be stopped? Sovereignty means control over execution, not just storage.

That is why we integrated the OpenShell runtime within SUSE AI Factory. Agents run in an isolated sandbox, inside the organisation’s own perimeter, with security and governance enforced between the agent and the underlying infrastructure. But containment is only half the answer—the other half is verifiable auditability. Enterprises need immutable telemetry showing every agent tool call, data access and state change.

It is no longer optional because an answer can be ignored, but an action creates liability. Regulators from the EU AI Act to DORA and NIS2 expect organisations to explain, prove and govern what their autonomous systems do. There is also a strategic dimension: if your agents can only run on one provider’s stack, your operational autonomy depends on that provider’s roadmap and pricing. Real sovereignty means you can choose where, and on what, your agents run.

In your Place de l’IT interview, you cast sysadmins as conductors of AI agents. Once agents reach enterprise tools through MCP, which permissions, skills and accountability must that conductor own?

A conductor does not play every instrument, but owns the performance. The same applies here.

First, permissions must move to least privilege: every agent needs an ephemeral, task-scoped identity – no shared service accounts, no standing admin rights, and MCP connections routed through a governed gateway rather than wired ad hoc. In SUSE AI Factory, that is exactly what we deliver. Second, skills shift from typing commands to system design: platform engineers must write policy as code, read agent execution traces like system logs, and pair deep Linux and Kubernetes fundamentals with the judgment of when to step in. Finally, accountability is non-negotiable. Every agent in production needs a named human owner, an immutable audit trail, and an instant kill switch. If nobody can answer ‘who is responsible for this agent?’, it belongs in a lab, not production. The sysadmin role isn’t disappearing – it is moving up a level.

At SUSECON you urged architecting for innovation through choice, not just survival. Which capabilities will decide whether enterprises can change direction as agents, regulation and hardware cycles accelerate?

Choice is a capability you build before you need it. By the time you need to change direction, it is too late to start. I would point to four capabilities.

First, workload portability. Containerised, Kubernetes-based platforms that run the same way in the data centre, any public cloud and at the edge, so moving a workload is a decision rather than a project. Second, model and hardware independence. GPUs, accelerators and models are changing every few months, so the platform must support different models and silicon without rewriting applications. Third, consistent policy and observability across the whole estate, because when a regulation changes you need to apply the new rule everywhere at once and prove you have done it. Fourth, a stable, long-lifecycle foundation underneath. Fast innovation at the top only works if the operating system and platform layer are secure, supported and predictable for years.

Enterprises with those four in place can treat each new wave as an opportunity. Those without them treat each wave as a risk and a need for migration.

Gartner warns two in five enterprises may roll back autonomous agents by 2027. Which assumption fails first: public cloud first, human approval everywhere, or metered API pricing? What replaces it?

Metered API pricing fails first. It suits experiments, where a few thousand calls cost little. Agents do not behave like experiments. They loop, retry, call tools and call other agents, and a single workflow can generate thousands of model calls. When the first invoice arrives for an agent running all month, many business cases fall apart. Gartner’s own analysis cites rising costs and unclear value as leading reasons projects get cancelled.

What replaces it is the predictable capacity the enterprise owns. That means running open models on private infrastructure – on-premises or in a sovereign cloud – where you pay for compute capacity rather than every looped reasoning step. You reserve metered commercial APIs for edge-case tasks or burst peaks. The economics shift from ‘per-token anxiety’ to traditional capacity planning, which IT and platform teams already know how to model and budget.

‘Human approval everywhere’ fails close behind, because it does not scale. If every action needs a click, you have built a slower workflow, not an agentic one. It should give way to risk-tiered autonomy: agents act freely on low-risk, reversible tasks, and humans approve the high-impact ones.

Public cloud first does not so much fail as mature into ‘right place first’, where each workload runs wherever cost, control and compliance are best balanced.

As a CNCF board member since 2017, what should CTOs and open source communities do now so agentic AI strengthens shared infrastructure rather than concentrating control among a few vendors?

Cloud native is a good precedent. Kubernetes could easily have become one company’s product. It became shared infrastructure because vendors and users agreed to build it together under neutral governance. Agentic AI is at the same fork in the road now.

For CTOs, the most powerful lever is procurement. Make openness and exit a buying criterion: open protocols for how agents connect to tools and data, the ability to run your choice of models, and no proprietary sandbox that holds your agents hostage. Fund the upstream projects you depend on, and let your engineers contribute on company time. Use is not the same as support.

For communities, the priority is the boring, essential layers: agent identity, security, observability and interoperability standards, plus reference architectures that anyone can deploy. That is where shared infrastructure is won or lost. We also still need refined clarity about what ‘open’ means in AI, because open weights, open source and open washing are not the same thing.

If we get this right, agents will run on a common, open foundation, just as cloud native workloads do today. If we do not, we will spend the next decade painfully negotiating with different fragmented gatekeepers.

Thank you so much, Thomas, for participating in our IT Tech Pulse Interview series. We look forward to having you and SUSE leaders again!

Write to us [editorial@demandmediaagency.com] to learn more about our exclusive editorial packages and programmes.

 

About Dr. Thomas Di GiacomoAbout SUSE

Dr. Thomas Di Giacomo is Chief Technology and Product Officer at SUSE, the global leader in enterprise open source solutions. He leads SUSE’s engineering, product and innovation organisations, shaping the company’s technology strategy across enterprise Linux, cloud native and Kubernetes, edge and AI infrastructure. His focus is helping enterprises build open, secure and sovereign platforms that give them genuine choice as technology, regulation and hardware evolve.

Thomas joined SUSE as Chief Technology Officer in 2016 and later served as President of Engineering, Product and Innovation before taking his current role. Under his technical leadership, SUSE has expanded from its enterprise Linux heritage into cloud native computing, including the integration of Rancher, and more recently into enterprise AI with SUSE AI and agentic AI capabilities.

Before SUSE, Thomas was Global CTO at Hoist Group and spent several years at Swisscom, rising from service manager to CTO of Swisscom Hospitality, where he led engineering, development, IT and product for IPTV and connectivity services. Earlier, he managed Linux server infrastructure for the City of Geneva and worked as a researcher, contributing to multimedia standards work including MPEG and to EU-funded industrial projects.

Thomas is a long-standing advocate for open source governance. He served on the governing board of the Cloud Native Computing Foundation (CNCF) for many years, and has previously served on the boards of the Cloud Foundry Foundation, LF Networking and OPNFV. He holds a PhD in computer science from the University of Geneva and master’s degrees in computer science from Université Joseph Fourier and Grenoble INP.

SUSE is a global leader of enterprise open source software. By transforming community innovations into secure, sovereign and AI-ready solutions, SUSE empowers customers to escape vendor lock-in and regain control of their IT destiny. Through industry-leading Linux, Kubernetes, Edge and AI infrastructure solutions, SUSE delivers the flexibility to innovate everywhere—from the data center to multi-cloud and out to the edge. Only SUSE also manages many Linux and Kubernetes distributions. At SUSE, Choice Happens because we prioritize community, interoperability and relentless innovation.

  • Raviraj Solanki

    Raviraj Solanki is a PR & Media Strategist and Growth Partner specializing in executive thought leadership, global PR programs, and B2B pipeline growth. At ITTech Pulse and Demand Media BPM, he partners with enterprise tech leaders to amplify their vision through structured Q&As, strategic editorial placements, and multi-channel campaign distribution.