[POV] AAVA™: A New Model for How Software Is Conceived, Built, and Managed

Some organizations accept the chaos of software delivery as the nature of the work. It isn’t. It’s a process problem, and it’s specific enough to map.

Every engineering organization runs a version of the software delivery process. Requirements come through, design begins, code gets written, testing runs, something breaks, and it starts over. What looks linear in the methodology is back-and-forth in practice; designs get revised mid-build, gaps surface in testing that were missed upstream, and the cost of that pattern accumulates long before it appears on any project dashboard.

When designing Ascendion’s agentic AI platform, AAVA, the full engineering lifecycle was mapped end to end, from ideation through production operations. That analysis surfaced approximately a hundred to a hundred and fifty distinct friction points that were causing rework, delays, unnecessary costs, and forcing more labor than the work required. Nearly all of it traced back to the same root causes: inconsistency in how work gets completed, gaps in standards, and governance that weakens under pressure. These are structural problems that accumulate silently across the lifecycle until they show up as quality gaps, cost, and delay.

Friction Doesn’t Live in One Stage

The tendency in applying AI to engineering has been to target individual stages and attack them. Better code generation. Smarter testing. Faster deployment. Each improvement is valuable, but none address the inefficiency that compounds across the whole lifecycle.

Think of engineering as a supply chain—raw materials gathering through last-mile delivery in operations. Every stage in that chain depends on the quality of what came before it. A gap that passes unresolved from requirements through design and into code doesn’t stay contained. By the time it surfaces in testing, it has traveled through everything built on top of it. The further a gap travels, the more it takes down with it. That compounding action is what makes engineering expensive, and it runs across the full chain. Addressing one stage leaves the underlying inefficiency untouched.

AAVA is the orchestration engine for the entire chain, from the first moment a capability is conceived through production operations and ongoing management. When we built the platform, that scope was deliberate and rooted in a simple observation: friction doesn’t live in one stage. It accumulates across all of them.

When AAVA becomes that engine for our clients, the first thing they have to do is reimagine their engineering processes. Today, many organizations run processes that were designed for humans to move through each stage sequentially. They were not designed for agentic orchestration. Our goal with AAVA is to simplify and orchestrate that process. Consistency and uniformity are brought to every stage of the software delivery lifecycle in a way that human-managed handoffs have never been able to sustain reliably at scale.

AI Agents Designed for Precision

Our AI agent design follows a single responsibility principle. Every agent has one task, covering a small surface area, and can perform with genuine precision within that task. The more responsibilities assigned to a single agent, the more its accuracy across all of them degrades.

What makes these agents genuinely capable is how they understand context. We developed and embedded a knowledge hierarchy in AAVA agents that’s modeled on how the human brain stores and applies knowledge: tiered and context sensitive. AAVA agents carry four tiers of knowledge:

General knowledge can be applied anywhere. For example, before you cross a road, you wait for traffic to stop and look both ways. But crossing a road in India is different from crossing one in Manhattan or crossing one in a small town. That is the domain context, and it changes how general knowledge gets applied. Then there is the application layer. The shoes you’re wearing matter. In heels, you watch for drainage grates in the pavement, whereas, if you’re wearing sneakers, you may not even acknowledge the grates. Each layer changes what the right action is, even though the underlying task is the same.

An agent built for a retail banking organization reflects retail banking specifics as opposed to commercial banking, without needing another explanation with every prompt. It also knows the specific tools within the client’s ecosystem that it needs to perform its task. When a new joiner gets onboarded on an account, they undergo document reviews, training, and gradual exposure all before genuine domain expertise develops. That takes years. An agent absorbs it faster, and once it has it, consistently holds it in a way that individual contributors working across different engagements simply cannot.

Agents First. Humans in Control.

The default assumption about humans and agents in engineering is that humans produce and agents review. AAVA was built with the opposite idea in mind. Agents go first. They produce the artifact, whether that is a user story, a block of code, a test case, or an architecture document. The human comes in as the reviewer and decision-maker. They own the judgment.

A travel agent can book your flight. But your preferences (aisle seat, exit row, specific departure time) are yours to specify. The travel agent operates within the parameters you set. In the same way, an AI agent works for the engineer, and that relationship has to be designed in from the start.

Agents enforce engineering rigor and discipline, as well as the structural work of executing tasks consistently, holding to guidelines, and producing output that doesn’t drift. Humans apply judgment, situational expertise, and governance. Every consequential decision and checkpoint stays with people.

Designed for Adoption from Day One

Any technology organization does a finite set of things. The engineering processes that run through those organizations, such as the way software gets conceived, designed, built, and managed, are knowable and mappable. From day one, AAVA diffuses AI across the software delivery lifecycle, rather than adding it in as an afterthought. It comes with out-of-the-box agentic engineering processes that teams can use immediately, without needing to understand the design behind them. The moment it is deployed, the team is already working with something useful.

From that foundation, Ascendion works with teams to design with AI, teaching them to build and adapt their own agentic processes to their specific context and needs, without requiring deep AI knowledge to do so. People work confidently when they feel in control, and confidence comes from using something that already works. That supported progression, from user to designer, is what makes adoption hold over time.

Engineering for Outcomes

In many enterprises, engineering still moves in stage-based timelines. Each stage has its own metric, its own definition of “complete”. Optimizing individual stages doesn’t necessarily produce a better outcome at the end. A team can hit every stage metric and still deliver late, over budget, and be misaligned with what the business actually needed.

The more useful measure is the business outcome. When does this capability need to be in the hands of customers? What does the business need to be true in two months, or six? Engineering’s job is to work backward from that, to orchestrate every stage of the supply chain in service of that end result rather than in service of its own internal metrics.

This is what changes when the full lifecycle is orchestrated rather than managed stage by stage. The question shifts from how long does each phase take to what does the business need and when. Technology change becomes something an organization can execute, at the pace its market demands. The organizations that struggled most during the digital transformation era weren’t short on strategy. They knew what needed to change. What they couldn’t do was move technology fast enough and affordably enough to keep pace with it. That is the gap a well-orchestrated engineering lifecycle closes.

Engineering to Impact a Billion Lives

The people on the other side of software delivery feel the impact of how well engineering works. Fifty thousand new Medicare members gaining access to health coverage every quarter. Over fifty-four million athletes whose records power a more personal experience. More than fifteen thousand auto dealers with access to a bug-free digital portal. That is why engineering cycles that run too long, cost too much, and still fall short of what the business needs are worth agonizing over.

AI is the best available means of addressing that today. Tomorrow it may be quantum computing, or something that doesn’t have a name yet. The tools will change. The drive to solve it won’t.

About the Authors

Arun Varadarajan is Chief Commercial Officer at Ascendion, where he leads global growth and drives the commercial strategy behind AAVA. Previously VP and Global Head of Data at Cognizant, he built a $1.1B high-growth data practice and accelerated digital transformations for Fortune 500 clients. With deep technical acumen and a career spanning startups, global tech giants, and large-scale transformations, Arun brings a rare combination of engineering depth and business instinct to every engagement. His focus has always been the same: technology that produces outcomes, at the pace and scale organizations need to be winners in their markets.

Karthikeyan Girijanandan (KG) is Global Head of Technology & Architecture at Ascendion, where he leads the technology roadmap for AAVA, across North America, APAC, and Europe. With nearly two decades of experience spanning enterprise architecture, digital commerce, and cloud-native transformation, he has architected solutions for global organizations across retail, financial services, and healthcare. KG’s work sits at the intersection of AI strategy and engineering execution, translating emerging technology into systems that deliver measurable business outcomes. He is a recognized thought leader in GenAI adoption, human-AI collaboration, and responsible AI implementation at enterprise scale.