Reimagining the Enterprise Computer in the Agentic AI Era

Enterprise software is heading into a generational shift. What once revolved around static workflows, disparate data silos, and surface-level automation is now moving toward systems that can reason, act, and learn in the flow of work. At the center of this shift is a new vision. It sees the enterprise computer not as a collection of apps but as a unified, intelligent memory.

Ahmed Bashir, Chief Technology Officer at DevRev, brings a rare perspective to this transformation in a conversation with Thomas George, Managing Editor & President, CyberMedia Group. After overseeing massive-scale systems at Apple for over a decade, he joined DevRev in 2021 to help architect a fundamentally different AI-native foundation for the enterprise.

How DevRev came together and why this problem space matters

I joined DevRev in 2021 after nearly 13 years at Apple. At Apple, I worked at a massive scale, overseeing systems such as iMessage, FaceTime, notifications, and iCloud storage. We operated at exabyte scale, delivering over two trillion notifications daily. By the time I left, I felt I had seen that chapter through and that we were entering another generational shift at the intersection of AI and enterprise software. When DevRev started, just as large language models were beginning to show real promise, the timing felt right.

When I joined, GPT 2 was still considered experimental. Even then, we were convinced that AI would need a fundamentally different enterprise foundation. From day one, we focused on three things: building the best multi-tenant vector database, creating a workflow engine that could handle everything from millisecond actions to multi-hour processes, and unifying enterprise data. Not just data, but metadata too, so context, permissions, and schema were never an afterthought.

What makes DevRev's “computer” fundamentally different

We think of the computer as the place where work happens. Every generation of computing has made the computer more personal, more accessible, and closer to the user. The next evolution isn't about screens or form factors. It’s about memory and understanding.

Our "computer" is built on the best possible enterprise memory: short-term and long-term. Memory means understanding who you are, the work you do, the products you build, and the customers you serve. But critically, we don't create artificial boundaries. Most enterprises separate sales from support, issues from opportunities, and tickets from enhancements. Real life doesn't work that way.

If you're trying to close a deal while critical customer issues remain unresolved, those contexts must coexist. DevRev's platform understands the intersections across sales, product, support, and customer success. That’s what enables meaningful action, not just answers.

This isn't about layering an agent on top of an LLM. It's about having a deeply unified data layer, a permissions-aware context engine, and a workflow system fully embedded in enterprise memory. Only then can agents truly act.

The hardest engineering challenges in building agentic systems at scale

Semantic and syntactic search at scale, while respecting permissions, is extremely hard. Most systems compromise on either accuracy or governance. Another major challenge was workflows. Many AI systems treat workflows as reactive: if something triggers, an action happens. Our workflows are observational. They continuously observe changes across opportunities, tickets, cases, meetings, events… everything. The system understands who the change matters to and why.

That requires tight coupling between memory and automation. Our workflow engine doesn't sit at arm's length from enterprise context. It’s deeply embedded in it. This allows for more intelligent, probabilistic actions rather than brittle, deterministic rules.

Data synchronization is another hard problem. Our Air Sync layer doesn't just ingest data; it also ingests metadata and schema losslessly. As schemas evolve, those changes are reflected automatically, which is essential for long-term enterprise adaptability.

Early adopters and strongest traction

The industry is gravitating towards problems with clear ROI. That usually means customer-facing functions first: customer support, customer success, and parts of sales. KPIs such as customer effort, time to resolution, NPS, and employee effort are measurable and immediate. Enterprises want to see value quickly, and these areas provide that clarity. We're also seeing strong traction in enterprise sales and productivity, particularly because DevRev offers a Google-like experience inside the enterprise, but one that’s not just read-only. It’s read write. You can act, not just search.

Specific customer outcomes and benchmarks

We've seen up to 80% ticket deflection in some environments. Companies like Bill.com, Descope, and Figma have become strong internal reference points. Razorpay is another great example. They're moving quickly across multiple customer-facing and productivity use cases.

What matters most to us is daily usage. When teams use the system every day, that’s the strongest signal of success.

Learning and adaptability in agentic system evolution

Agents cannot be one-size-fits-all. They need to be upskilled differently for individuals, teams, and enterprises. Our roadmap is heavily focused on creating the best upskilling experience so agents evolve with how people work.

We're also investing deeply in multiplayer collaboration. Conversations aren't just adjacent to work in DevRev. They are the work. Discussions become part of enterprise memory, enriching context and improving future retrieval, reasoning, and action.

This creates a powerful reinforcement loop: relationships, discourse, and work items all become interconnected. That’s how you achieve truly personalized, accurate, and fast decision making.

The current state of "agentic AI" and levels of autonomy

Honestly, the industry needs better benchmarks. Right now, everyone claims to be "industry leading," which is meaningless.

We're developing academically backed benchmarks that clearly define what Level 1, Level 2, and Level 3 autonomy mean: search, reasoning, action, delegation, and beyond. The goal is to make these benchmarks open, measurable, and comparable across vendors.

That’s how the ecosystem matures through a shared language and transparent measurement, not marketing claims.

How AWS partnership supports scalability and architecture

My relationship with AWS goes back to around 2010, so it is deep. DevRev has been built on AWS from day one, not just for credits, but for architectural depth.

We collaborated closely on OpenSearch for multi tenant semantic search. Our workflow engine is fully serverless, built on AWS Lambda. Proximity matters to us. India, Singapore, Frankfurt, the US, and soon the Middle East enable low latency experiences for our customers. We're also early with Amazon Bedrock. That partnership is still evolving, but there’s strong alignment around where agentic platforms are headed.

The future of agentic AI in the next few years

We think in quarters, not years, but directionally, agents will absorb work from two places: humans and microservices.

There are many things humans do inconsistently, unscalably, or reluctantly. Agents will take those over. At the same time, microservices are too rigid and complex for a world with hundreds of millions of builders. Agents are more flexible, more democratised, and easier to upskill.

Instead of over engineering deterministic systems, we'll deploy skills and agents that reason, prioritize, and adapt. That shift is already underway, and over the next two to three years, it will accelerate by orders of magnitude.