Your AI is Only as Fast as Your Slowest System
If you’ve been thinking about adding an AI agent to your business, you might be worried about picking the “right” model. But according to three of the largest tech companies in the world, you’re worrying about the wrong thing.
At VB Transform 2026, senior leaders from LinkedIn, Walmart, and Zendesk shared a common story: when they moved AI agents from pilot to production, the bottleneck wasn’t the intelligence—it was the infrastructure. And the fix wasn’t more complex AI; it was better plumbing.
Here’s what they learned, and why it’s probably the most important thing you’ll hear about AI this year for your Malaysian SME.
What Happened
On July 17, 2026, three infrastructure leaders—Animesh Singh (LinkedIn), Desiree Gosby (Walmart), and Sami Ghoche (Zendesk)—shared the stage at VB Transform 2026 and described what went wrong when they tried to scale AI agents. Each hit a different wall, but the conclusion was the same: none of the bottlenecks were model problems. Source
The core tension was simple: most enterprise infrastructure was built for how humans work, not for how agents work. Agents think in milliseconds, but legacy systems take seconds to respond. That gap is where the real engineering happens.
“We built our own harness, our own control flow, and pushed the LLMs to the leaf instead of them orchestrating the loop,” Singh said. Roughly 80% of LinkedIn’s workflow is now scripted, deterministic code, with LLMs used only where reasoning is required. Source
LinkedIn’s first bottleneck wasn’t a model—it was Kubernetes. Containers that spin up on demand take seconds, which is too slow for agents. The fix was pre-provisioned pools of containers that swap workloads in real time. A second challenge was handling hallucination when LLMs evaluated each other. Singh’s team built a custom harness that commits evidence to disk before moving on. Source
Walmart’s bottleneck came from success. An internal agent harness went viral among employees, and “citizen developers” started building their own agents. The problem wasn’t innovation—it was duplication. The fix was governance: spotting overlap, promoting the best versions, and getting them into production without engineering becoming a chokepoint. Source
Zendesk hit its bottleneck on the data side. With roughly 20 billion customer conversations in their repository, Ghoche said you can’t just feed that into a large language model with a big context window. Instead, you have to invest in underlying data pipelines and infrastructure. Source
Why This Matters for Your Business
You might be thinking, “Those are huge companies. I’m a small business owner. How is this relevant?”
It’s more relevant than you think. The same principle applies no matter the size: your AI agent is only as good as the data and processes you feed it. If your inventory system is on a spreadsheet that updates daily, an AI agent that tries to give real-time stock updates will fail. If your customer service tickets live in a messy email inbox, an AI agent won’t magically sort them out.
The panel’s advice is surprisingly practical for SMEs:
- Invest in evals first. Ghoche called evals the thing common to every use case. They force you to break problems down, and with robust evals, you can move faster. For you, this means testing your AI with real scenarios before expecting it to work perfectly. Source
- Own your agent harness from day one. Gosby suggests putting the tool directly in employees’ hands early, but with monitoring. This could be as simple as letting your team use a chatbot while tracking what it does. It’s likely a fast way to see what works and what doesn’t. Source
- Build for flexibility. Singh advised building for independence from model providers. For an SME, this might mean choosing AI tools that are easy to switch if a better one comes along. Don’t lock yourself into one vendor. Source
The core message: before you spend on AI, make sure your infrastructure—your data, your processes, your workflows—is ready for it. Otherwise, you’ll be blaming the AI when the real problem is your old systems.
The Bigger Picture
This panel is a sign of a larger trend. As agents get faster and smarter, the weak link will shift from the AI to everything around it. For large companies, that means rethinking decades-old infrastructure. For small businesses, it means a chance to leapfrog: if you haven’t built up a lot of legacy systems, you can start with a clean slate that’s built for speed.
We’re likely to see a growing focus on “agent infrastructure” as a category. Tools that help with memory, orchestration, and routing will become as important as the models themselves. The companies that get ahead won’t be the ones with the best AI, but the ones that build the best bridges between their business operations and AI.
For you, the long-term bet is on data hygiene and process automation. If you keep your data clean and your workflows documented, you’ll be ready to plug in better AI as it comes along—without hitting a bottleneck.
The future isn’t about “which AI model to choose.” It’s about “is my business ready for AI?”
Book a free 15-min call to see how infrastructure modernization applies to your business →