Why Businesses Should Start Paying Attention to AI Agents Now

Blaxel CTO Christophe Ploujoux on building cloud infrastructure for an autonomous future

Most cloud platforms weren’t built with AI agents in mind, but that’s precisely what Christophe Ploujoux is focused on changing. As Co-founder and CTO of Blaxel, he’s building the infrastructure to support a world where autonomous AI doesn’t just assist with tasks, it manages and executes them. With a background in distributed systems and data platforms, Christophe has seen firsthand where today’s tools fall short and why that matters for businesses of all sizes. In this interview, he breaks down what “agentic computing” really means, why it’s more than just another AI buzzword, and how small teams and developers can start putting AI agents to work in practical, time-saving ways right now.

You’re working on cloud infrastructure for a world run by autonomous AI agents. What inspired this vision, and why now?

After working in cloud infrastructure for a while, you start to see the cracks. The issues weren’t just in the technology itself but in how the huge size of these companies makes it difficult for them to evolve as fast as  AI is moving. I was watching a video of Jensen Huang where he talked about a future with billions of AI agents communicating, and for me, that’s when it all clicked. It became clear what I needed to do. So now, I’m focused on building the infrastructure that this future will actually require.

The term “agentic computing” is gaining traction, but it can mean different things to different people. How do you define it, and what sets it apart from traditional AI workloads?

That’s the perfect way to frame it. The term is new and can be vague.

I see it as any computing that is driven by a goal, where the path to that goal isn’t pre-determined, and the time it takes can vary wildly.

The real distinction from standard AI workloads is the leap in autonomy. Traditional AI is great at executing a specific, well-defined task. An agent, however, is designed to pursue a broader objective. It has the freedom to decide its own tasks.

This is why the conversation is shifting away from constant human validation. We’re moving from a “do this, then wait” model to one where the agent can reason through multiple steps on its own. That’s the power of the new frontier models, they’ve bridged the gap from the early, clunkier LLMs to something that can tackle complex problems with a level of independence we just didn’t have before.

Traditional cloud platforms weren’t built for billions of independent agents. What are the biggest limitations of today’s infrastructure when it comes to supporting these kinds of workloads?

The infrastructure of today was built on the principle that workload size and time should be predetermined for you to build the perfect architecture that suits your use case. I said it in my previous answer: I believe this is shifting with AI Agents. We need a new architecture where we have:

• Low-latency computing and network
• Persistent state
• Sandboxed execution
• Tool access
• Robust networking

This is what I want to build.

The first one about low latency is for me one of the most important ones. Imagine billions of agents. In a world of agents like that, just a simple millisecond can have a huge impact on the energy it will consume. While that delay is nothing for a single agent, it adds up to a massive energy drain across the whole system. It makes me think that if we are not careful about that, the limitation we will have in the future will probably be physical instead of technological.

How do you see communication evolving between AI agents at scale? Are we looking at a whole new internet architecture or an adaptation of existing protocols?

It’s already happening. We’re seeing the first wave of this evolution with new standards emerging, like Google’s Agent-to-Agent (A2A) Protocol and Anthropic’s Model Context Protocol (MCP). This clearly shows we’re moving beyond the simple, stateless API calls of the past decade.

So, to answer your question directly, I see it happening in two stages. For the next year or two, we’ll likely stick to adapting existing protocols to meet these new demands. But in the longer term, I’m convinced major architectural changes to the internet itself will be necessary. The big challenge for all of us will be building the tools and infrastructure to integrate those major changes quickly and efficiently.

Security takes on new meaning when you’re dealing with trillions of autonomous entities. What are the key risks, and how should we start thinking about trust in agentic systems?

I can see a few key risks that we need to be very careful with when building an agentic system:

• Identity and access: use a temporary access system that you can revoke if necessary like OAuth. Use the Access Control List as much as possible. This is to prevent you in case credentials are stolen or another agent trying to impersonate as another
• Apply Zero Trust principles to all your workloads (agents, mcp servers…)
• Execution of malicious code by agent. I’d suggest to use air-tight short-live sandboxes to avoid any surprises
• Prompt injection/leakage: implement strong guardrails, sanitize and check any malicious input

There are many other risks in agent architecture, and we all need to be very concerned about those when building one. Your agent should never feel like a black box where you are uncertain of the outcome. Coming from a platform engineer background, I strongly suggest people to first be very worried about security and work with a strong observability system to avoid any surprises

Your company envisions cloud environments that agents can inhabit and evolve. Can you break down what that means in practical terms?

What that means in practical terms is that the developer’s role changes. Right now, a person has to design and set up the infrastructure for every single task. In the environment, I envision, an agent just does that work on its own.

For example, say an agent gets a goal like, ‘Process this dataset for financial anomalies.’ It won’t wait for a human to provision servers or databases. It will figure out what it needs, build a temporary, perfectly-sized architecture for that one job, and then tear it all down when it’s finished.

This vision is the entire reason we made our products easily pilotable by an agent. For a future like this to work, the cloud’s building blocks can’t be complex systems designed only for humans to click through. They need to have clean, simple APIs that an agent can understand and use to build with.

For businesses just starting to explore agent-based AI, what kinds of problems or use cases are best suited to this model today?

For businesses just starting out, I’d say that the guiding principle should be to find where you can free up the most time for your team.

Some examples that come to mind:

• Customer support. Don’t try to automate everything. Start by letting an AI agent be the first point of contact. It can handle simple, repetitive questions and issues, which provides your customers with 24/7 support. This lets your expert human team focus exclusively on the difficult cases that require deep knowledge and empathy, which dramatically increases the quality of your support.
• Complex, data-intensive process. If you have tasks that require someone to manually gather, clean, or analyze data on a regular basis, an agent can probably take that over completely.

There are so many things to explore with AI Agents it can be hard to figure out where to begin. That’s why I try to follow a principle: first, by thinking about my own work, and then expanding that to all the tasks being done in my company.

Another thing I find interesting to do is, that whenever we’re about to recruit for a new position, we always try to think of an AI alternative first. Most of the time, we still end up recruiting the person, but we’ll probably have an AI doing part of the mission we had in that original job description.

As someone with a background in distributed systems and large-scale data pipelines, what lessons from those domains apply or don’t apply when designing for autonomous AI?

Everything about my previous experience was about optimizing and building a great scalable architecture that would handle petabytes of data. I’m convinced this applies to autonomous systems. The amount of data that will be ingested, processed, and analyzed is unheard of. That is why everything, meaning every bit of the infrastructure, needs to be optimized. The tiniest error could have a huge impact.

That’s why I decided to build with all of my co-founders an infrastructure where optimization, reliability, observability and distribution are not an option but a key feature of our product.

Looking ahead, what excites you most about the potential for agentic computing and what keeps you up at night?

What truly excites me about the future of agentic computing is the sheer scale at which agents and large language models (LLMs) will need to operate. We’re already seeing discussions around the energy demands this will create, and frankly, the immense size of it amazes me. My drive is to be a part of this by building the most efficient systems possible.

I’ve always felt like my entire career has been leading up to this moment. I missed the birth of the web because I was born too late, I’m not going to miss the AI wave. More broadly, I want to contribute to building a system that genuinely makes our world a better place. While many people fear AI’s potential to surpass human capabilities, I’m a strong believer in a future where AI works in harmony with us, elevating humanity as a whole.

The post Why Businesses Should Start Paying Attention to AI Agents Now appeared first on SiteProNews.

Source: https://www.sitepronews.com/2025/07/10/why-businesses-should-start-paying-attention-to-ai-agents-now/