Architecture Deconstructed: Re-architecting the Stack for the AI Era

In our first episode of Architecture Deconstructed, we sat down with two seasoned engineering leaders - Satish Raghunath, who leads Infrastructure Engineering at Salesforce, and Emil Ibrishimov, Head of Engineering, Product, and Design at recruiting-tech startup Gem. Hosted by Boris Bogatin (CEO, Catio) and Toufic Boubez (CTO, Catio), the conversation surfaced powerful themes around the shifting role of architecture in a post-AI world.

It was unscripted, unfiltered, and often refreshingly contrarian. The result? Practical lessons and lived experience that go beyond platitudes.

Below are three moments that cut to the heart of what it means to lead modern technical teams.

Watch the full episode here:

When AI Tools Reveal Unmet Needs: Evolution Through User Behavior

A fascinating pattern is emerging in how engineering teams adopt and adapt AI tools. Both engineers and technical leaders are turning to their familiar AI platforms to solve needs beyond the tools' original capabilities.

As Emil shares:

"People are using them also in ways that are outside of the direct way in which they were designed to be used... engineers... start their initial draft of a design doc in a coding co-pilot."

This organic evolution of tool usage extends to leadership as well. Emil describes his own shift:

"I used to... go to my Slack group with other friends in similar roles... now I first go to the AI, do initial research. If I need to, then I can go to the network."

This parallel trend - engineers using coding copilots for design documents while tech leaders turn to ChatGPT and Gemini for decision support - mirrors a familiar pattern in software evolution. Just as developers' early use of ChatGPT for coding eventually led to specialized coding copilots, we're seeing signs of new needs emerging through how users adapt existing tools.

This constant evolution suggests we're watching the early signals of where specialized AI tools might next emerge - particularly in the space of technical decision-making and architectural planning.

AI is Transforming Knowledge Access, But Context and Specialization Matter

AI is fundamentally reshaping how organizations access and use knowledge. Satish captures this shift vividly:

“I don't go to channels anymore. I just ask Slack AI. Even URLs for documents. I don't even search anymore. LLMs have been game changers for information discovery.”

But this new ease of access brings its own challenges particularly when it comes to deeply contextual or domain-specific work like architecture and technical planning.

Fragmentation and lack of specialized insight remain major roadblocks. As Emil explains:

“We have so much information in Slack… We use everything - Docs, Notion, Slack. That's great for experimentation, but it creates silos.”

Even with LLMs, Satish notes that general-purpose answers aren’t enough when the stakes are strategic:

“LLMs are great, but I need it to be thought through by way of architecture. Am I going to go to each LLM and go figure out what does this say about architecture?”

The conversation points to a clear set of emerging needs:

1. Unified Knowledge Layers that bring together scattered documentation, chats, and tools into a single, coherent interface.
2. Context-Aware Reasoning, tailored to domains like architecture, that goes beyond summarizing and starts synthesizing.
3. Outcome-Driven Insights that help teams not just find information, but make better decisions because of how that information is processed and framed.

The future isn’t just about better access, it’s about systems that understand why you're seeking knowledge, and can actively shape how it informs your strategic outcomes.

Beyond Uptime: Understanding and Managing System Dependencies in Mission-Critical Infrastructure

At Salesforce, one of the world's largest enterprise platforms, even a minor service disruption can significantly impact critical operations ranging from financial services to emergency response systems. For Salesforce, availability isn't just a metric - it's a mission-critical imperative. As Satish explains:

"Even a blip in terms of a downtime has an impact on day-to-day lives of people. People do finances, they do insurance, they do emergency services, they do a lot of really critical things on top of the platform."

This reality drives a fundamental shift in how architecture decisions are made. Availability and resilience can't be afterthoughts:

"It can't be a bolt on. We have to think right from the point when somebody starts a design document even before a line of code is written."

Satish calls this the "radius of impact", a concept that goes beyond traditional blast radius thinking. It raises critical questions: How much of your surface area is impacted by it? How many customers experience it? How do you measure it? How do you observe it? How do you recover from it?

The challenge becomes particularly complex at scale. As Satish notes:

"That's a very hard question to get right in production at scale... you have to ask, OK, what are my hard dependencies and what are my soft dependencies? If I'm building a system that needs to be a four nines system, I cannot obviously be dependent on a three and a half nine system."

What emerges is the critical need for real-time visibility into system dependencies. This is especially crucial when dealing with legacy systems, as Satish points out:

"We don't know what things it's impacting because everything else is modernized, but this thing isn't. And we have no idea actually how much it wipes out when it goes down."

The solution isn't just about better tooling—it's about fundamentally rethinking how we architect for visibility. Engineering leaders need real-time, comprehensive dependency mapping that evolves with their systems. This visibility isn't just about documentation—it's about having a living, breathing understanding of how systems interact and impact each other, enabling proactive rather than reactive architecture decisions. Only then can organizations truly understand and manage the radius of impact in their increasingly complex infrastructures.

Final Thoughts: Toward an Architecture-Aware AI Era

This conversation confirmed what many in the trenches already feel: architecture is becoming more dynamic, more contextual, and more dependent on real-time decision tooling than ever before.

AI can be a massive accelerant but only if paired with strong architectural foundations and a clear understanding of your team's workflows and leverage points.

That’s what we’re building toward at Catio. We believe architecture shouldn’t be a postmortem. It should be an active, evolving intelligence layer for your organization.

🎧 Watch the full episode of Architecture Deconstructed to hear more from Satish, Emil, and the Catio team.

‍