Why complexity is sabotaging AI ambitions

In my 11 years building enterprise systems — from Practo’s healthcare cloud infrastructure to VMware’s hybrid cloud solutions to now architecting F5’s BIG-IP management plane — I’ve witnessed many technology cycles that promised transformation. Most delivered complexity instead. But what I’m observing in 2025 is different, and frankly, more troubling.

The enterprise technology landscape presents a fascinating paradox: at the very moment when artificial intelligence promises to solve our most complex operational challenges, those same complexities are preventing organizations from effectively deploying AI solutions. When I analyzed F5’s latest State of Application Strategy report alongside A10 Networks’ State of Application Load Balancing study, the data revealed a troubling disconnect between AI aspirations and infrastructure reality that I recognize from my own work building these systems.

We’re facing the ultimate irony: Organizations can’t use AI to solve their IT problems because their IT problems prevent them from using AI.

The wake-up call in the data

The statistics in these reports immediately grabbed my attention because they mirror what I’ve been seeing in my work with enterprise customers. F5’s research shows that 96% of organizations are deploying AI models, and 73% want AI to optimize their application performance. On the surface, this looks like a massive AI adoption success.

But then I saw the other number: 60% of IT professionals are ‘mired in manual operational tasks.’ That statistic hit home. In my current role managing BIG-IP’s management plane for large enterprises, I regularly see teams that want AI-powered automation but can’t find the bandwidth to implement it because they’re constantly firefighting.

The shift in AI adoption barriers tells an even more revealing story. In 2024, data quality was the primary obstacle. By 2025, it’s shifted to human skillsets — 54% of organizations lack sufficient AI expertise. But having worked on both sides of this equation, I suspect the real issue isn’t lack of skills, it’s lack of time to develop and apply those skills.

When A10’s research showed that 58% of organizations consider API sprawl a ‘significant pain point,’ I knew we were looking at more than just operational inefficiency. We’re looking at the root cause preventing AI implementation: teams so overwhelmed with managing complexity that they can’t step back to implement the solutions that could manage that complexity for them.

What I learned on both sides of the complexity equation

My journey from VMware to F5 gave me a unique vantage point on this complexity crisis. At VMware, I was part of the team promising hybrid cloud simplification. At F5, I now help enterprises manage the complexity that those promises created.

During my four years at VMware, I worked on the development of their hybrid cloud solution that enabled seamless workload management across on-premises and cloud environments. We integrated core VMware technologies like vSphere and NSX with public cloud platforms. On paper, it was elegant: one management interface, consistent policies, workload mobility between environments. The solution was successful enough that Forrester named VMware a leader in hybrid cloud management.

But implementation taught me that ‘seamless’ and ‘simple’ aren’t the same thing. Each cloud provider had different APIs, security models and operational procedures. What we called ‘unified management’ still required teams to understand the nuances of multiple platforms. We solved the technical integration challenge, but we didn’t eliminate the operational complexity — we just centralized it.

Now, in my role as technical leader for F5’s BIG-IP management plane, I see the other side of that equation. I work with enterprises that have implemented these hybrid solutions and are struggling with exactly the operational burden we thought we had solved. They’re managing applications across multiple clouds, dealing with different load balancing requirements for each environment, and trying to maintain consistent security policies across platforms that were never designed to work the same way.

The F5 report’s finding that 94% of organizations deploy apps across multiple environments, with a median of four different public cloud vendors, isn’t just a statistic to me — it’s the daily reality of the enterprises I support. When I see that 79% have moved applications back from public clouds to on-premises, I recognize the disillusionment. The hybrid cloud flexibility we promised became hybrid cloud complexity that they couldn’t manage.

What strikes me most is how this complexity compounds. Each additional cloud provider doesn’t just add one more platform to manage — it adds exponential integration points, API relationships and potential failure modes. The very solutions designed to provide operational flexibility are consuming operational capacity.”

The AI implementation irony

The most frustrating part of this complexity isn’t just operational. It’s the missed opportunities. I regularly see enterprise teams that could benefit enormously from AI-powered infrastructure automation, but they’re trapped in a cycle of vendor API management that consumes their bandwidth.

Here’s what I’ve observed repeatedly: an organization will implement automation that works beautifully across their infrastructure stack—load balancers, cloud platforms, monitoring tools all talking to each other through APIs. Then a vendor pushes an API upgrade that breaks compatibility with existing integrations. Suddenly, the automation that was supposed to reduce manual work becomes the source of emergency manual work as teams scramble to update their integrations.

Making this worse is how differently each vendor approaches the same functionality. AWS’s load balancer API uses completely different authentication, data structures and error handling than Azure’s, which differs entirely from on-premises solutions. A team might spend months building AI-powered traffic optimization that works perfectly with one vendor’s API, only to discover they need to rebuild everything when they expand to another platform.

This is the AI implementation irony in its purest form: we want AI to manage our infrastructure complexity, but we can’t implement AI because we’re too busy managing infrastructure complexity. The F5 data showing that working with vendor APIs is the most time-consuming automation-related task isn’t just a statistic—it’s the daily reality preventing AI adoption.

The vendor relationship crisis

Working inside F5 while analyzing both our own research and A10’s findings gives me an uncomfortable but necessary perspective on vendor relationships. The data from A10’s study is particularly sobering: 47% of EMEA IT professionals and 55% of U.S. executives would change their ADC providers due to limited or poor vendor support.

As someone who works for one of the major players in this space, I have to acknowledge this reflects a real problem. The research shows that 44% of organizations have faced issues with recent vendor licensing changes, and 29% of executives cite rising licensing costs as their top complaint…ahead of even security concerns.

What troubles me most is the timing. These vendor relationship issues are surfacing precisely when enterprises need the most support navigating hybrid cloud complexity. When organizations are struggling with managing applications across four different cloud platforms, the last thing they need is uncertainty about their infrastructure vendor relationships.

From my position at F5, I see both sides of this dynamic. Vendors are under pressure to demonstrate value and growth in an increasingly competitive market. But enterprises are also under pressure — they’re drowning in operational complexity and need partners who reduce rather than increase their burden.

The A10 data shows that when organizations do consider changing vendors, they prioritize integrated security features (52%), lower infrastructure costs (45%) and superior vendor support (33% in EMEA). What’s telling is that ‘superior vendor support’ ranks so highly. It suggests that the current vendor ecosystem isn’t meeting basic support expectations during a critical transformation period.

What enterprises should do differently

After little more than a decade building and managing these systems, I’ve learned that the path out of this complexity trap isn’t more technology — it’s more discipline. The enterprises that successfully implement AI aren’t necessarily the ones with the biggest budgets or the most advanced infrastructure. They’re the ones that have simplified their operational foundation first.

The F5 research shows that 93% of organizations now generate revenue through digital applications, up from just 79% two years ago. This isn’t just a statistic, it’s a business imperative. When your revenue depends on digital infrastructure, complexity becomes a direct threat to business continuity.

Based on my experience building hybrid solutions at VMware and now supporting enterprise ADC deployments at F5, here’s what I believe organizations need to do differently:

First, resist the urge to add more vendors when existing ones aren’t meeting your needs. The A10 data shows organizations managing relationships with multiple ADC providers, often out of frustration with their primary vendor. But vendor proliferation is complexity proliferation. Instead of adding another vendor to your stack, invest time in fixing the relationships you have or making deliberate vendor consolidation decisions.

Second, audit your API landscape ruthlessly. When 58% of organizations struggle with API sprawl, and working with vendor APIs is the most time-consuming automation task, you know there’s waste in the system. In my work with enterprises, I’ve seen teams managing dozens of different APIs just for application delivery and security. Every API represents operational overhead. Question whether each one is truly necessary.

Third, standardize before you automate. The F5 report shows that 95% of organizations are standardizing with observability tools like OpenTelemetry. This kind of standardization creates the foundation for automation. You can’t effectively automate chaos — you have to organize it first.

Finally, treat vendor selection as infrastructure architecture. When evaluating new tools or vendors, ask one question: ‘Will this reduce or increase our operational complexity?’ If the answer isn’t clearly ‘reduce,’ don’t implement it. The short-term gains aren’t worth the long-term operational debt.

The path forward I’m taking

In my current work leading F5’s BIG-IP management plane architecture, I’m trying to apply these lessons learned. The challenge is significant: BIG-IP serves as critical infrastructure for enterprises worldwide, and any changes we make have to account for the operational complexity our customers are already managing.

One thing I’ve focused on is reducing the operational burden for teams managing our platform across hybrid environments. Rather than adding more features that require more configuration, we’re working on intelligent defaults and automated policy management that reduce the number of decisions operators need to make manually.

The F5 research showing that automation has become the top use case for operational telemetry gives me hope. Organizations are moving beyond using data just for alerts — they want it to drive automated responses. This shift from reactive to proactive operations is exactly what’s needed to break the complexity cycle. I’m also paying close attention to the API sprawl problem. In my architecture work, I constantly ask: ‘Are we adding another API that teams will have to learn and maintain, or are we simplifying existing ones?’ The answer shapes our development priorities.

Looking ahead, I believe the enterprises that succeed with AI implementation will be those that solve their operational complexity first. The data shows organizations want AI to optimize application performance and even handle security responses automatically. But you can’t effectively automate systems you don’t fully control or understand.

From my vantage point working on infrastructure that supports AI workloads, I see the future dividing into two camps: organizations that simplify their infrastructure to enable AI capabilities, and those that remain trapped in complexity cycles. The gap between these two groups will only widen as AI becomes more central to business operations.

The irony of our current moment is that we have the tools to solve our operational problems, but our operational problems prevent us from using those tools. Breaking this cycle requires discipline, not just technology. The enterprises that recognize this will be the ones that successfully harness AI’s potential.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?