Darwinbox Cortex: The Innovator’s Dilemma Comes for HCM

This may be the most notable HR technology story I have come across all year.

Darwinbox was founded in 2015, making it roughly a decade younger than the next-youngest company in Gartner’s Magic Quadrant for HCM.

The company began with several architectural advantages. Its founders could study the systems that came before them, incorporate newer technologies and build a global enterprise HCM platform on a clean foundation.

Then AI arrived.

Despite having one of the newest architectures in enterprise HCM, Darwinbox saw where the industry's approach of layering AI onto existing platforms would eventually plateau. Rather than wait to hit that ceiling, the company chose to rebuild ahead of it.

Extending that AI-native foundation for the next era required starting over

That decision has major implications for the rest of the market. If the youngest enterprise HCM vendor reached the limits of its architecture, what does that suggest about companies applying AI to systems designed much earlier?

I recently spoke with Darwinbox co-founder Jayant Paleti about the company’s original platform bet, the difficult decision to rebuild and what the resulting product, Darwinbox Cortex, tells us about the next era of HCM.

Building the platform before the modules

Darwinbox entered a 40-year-old software category with ambitions to serve large, complex and global enterprises.

Jayant and his co-founders saw an opportunity to learn from established HCM providers while incorporating architectural ideas from companies such as Facebook and ServiceNow. Darwinbox became an early adopter of technologies including graph databases and Kubernetes.

The company also made an important product decision.

Rather than building each HCM module independently, Darwinbox began by creating a central platform containing the objects and infrastructure that every module would need. That included workflows, permissions, reporting and other shared components.

“So it took us two years to build the platform,” Jayant said.

From the outside, that progress looked slow. Darwinbox had spent two years building only a small number of visible modules.

Inside the company, the foundational work was beginning to pay off.

Jayant estimated that approximately 60% of what each new module required could be inherited from the central platform. The team could focus on the elements unique to recruiting, rewards, engagement or another specific function.

Once the platform existed, Darwinbox began releasing new modules approximately every six months.

This is a useful way to understand the platform concept in HCM. Modules become specific expressions of a common platform. When the workflow, permissioning and reporting layers are already established, the visible application can come together far faster.

That architectural decision helped Darwinbox build a broad HCM suite within its first decade.

It also gave the company enough flexibility to recognize when another foundational shift was required.

When the ground kept shifting

Darwinbox began releasing AI-powered functionality as generative AI entered the enterprise software market.

The company was shipping quickly, especially compared with larger and older competitors. Still, the work left the team dissatisfied.

“It felt like the ground was constantly shifting under our feet with the AI wave,” Jayant said.

Every few months brought a new generation of models, capabilities and assumptions about how enterprise software should operate.

Darwinbox kept adding AI functionality to the platform it had spent the previous decade building. Over time, the team recognized the ceiling of that approach.

Jayant said “all that we were doing was AI assisted at best, AI enabled at best, but was not really AI native.”

That realization led Darwinbox back to the same question its founders had asked in 2015.

How would they build an enterprise HCM system today if they could begin with a blank page?

The company created a small, talent-dense internal team and physically separated it from the broader organization. Its mandate was to imagine an HCM platform built specifically for the AI era, unconstrained by the existing Darwinbox product.

Jayant described the process as “incredibly liberating.”

The result became Darwinbox Cortex.

A system that begins the work

The first major architectural change inside Cortex is an events and signals layer.

Traditional systems of record are generally passive. A user opens a page, initiates a process or submits information before the system begins doing anything.

Cortex is designed to continually recognize changes in HR context.

An employee joins the company. Someone becomes a manager for the first time. A payroll discrepancy occurs. A performance cycle reaches its midpoint.

Each change becomes an event the system can recognize. Cortex can then launch the appropriate playbook without waiting for someone to initiate it manually.

Jayant encouraged HR leaders to think about the employee lifecycle as tens of thousands of events rather than a fixed collection of modules.

That shift could make the HR system far more proactive. It also changes the role of the administrator, who can spend less time identifying when a process should begin and more time defining how the organization should respond.

Christopher Nolan season comes to HCM

The second foundational layer is the context graph.

Jayant defines this as the collection of domain knowledge inside the product, organized around each customer, use case and persona in a form that AI can understand.

The goal is to let HR teams build within the system without extensive product certifications or outside technical resources.

To explain the idea, Jayant turned to one of the more unexpected analogies I have heard in an HCM interview.

“This is Christopher Nolan season, right, with Odyssey coming through,” he said.

Jayant had recently rewatched Inception and was struck by the film’s world-building sequence, where a character can imagine a structure and bring it into existence almost immediately.

He sees the context graph creating a similar experience for HR teams.

Consider a company that has spent months developing its performance philosophy. Historically, translating that philosophy into the HCM platform could require a product expert, an implementation partner and extensive configuration work.

With the context graph, the company could upload its philosophy directly. The system could interpret it, identify missing information, ask clarifying questions and create the necessary configuration.

“You can just approve and it can go live,” Jayant said.

The same approach could reduce the time and expertise required to implement and maintain the broader system.

Turning the autonomy dial

Cortex also includes an agentic execution layer that allows AI to complete work inside the system.

The most useful concept Jayant introduced was the autonomy dial.

“You can turn the autonomy dial from fully manual to fully automated,” he said.

A workflow might contain ten steps. The organization could allow the system to complete some automatically, use AI to assist with others and require a human decision at the most sensitive points.

In Jayant’s words, “autonomy just becomes a switch.”

His examples helped make the idea tangible.

When an employee calls in sick and needs time off approved, an organization may feel comfortable allowing the system to handle the process automatically.

Executive compensation requires a different setting.

“The compensation for a VP, you don't want the system to decide. The human always has to be in the loop, right?”

This framework is more practical than treating autonomous AI as a single company-wide decision. Organizations can adjust autonomy based on the workflow, risk and consequences involved.

They can also increase it over time as confidence in the system grows.

The larger market implication

During our conversation, I told Jayant that Clayton Christensen would have been proud.

Darwinbox faced the innovator’s dilemma from an unusual position. It was already the youngest company competing at the highest level of enterprise HCM. Its product had been built using architectural principles that were considerably newer than many of its competitors.

Still, the company decided that applying AI to its existing foundation would eventually become a constraint.

Darwinbox chose to disrupt its own product.

The decision was expensive and difficult. It also required the company to run two realities at once: supporting the platform that customers use today while rebuilding for a future that remains difficult to predict.

Jayant said the decision became easier because of his time horizon.

“This is a company I want to run for the next 20 years,” he said.

A company planning for the next quarter may hesitate to absorb the cost of a foundational rebuild. A founder planning to operate the company for another two decades has a different calculation.

Even with that conviction, Jayant acknowledged the toll of the transition.

“The switch was hard,” he said. “Rebirth is painful.”

The experience also creates a new set of questions for HCM buyers.

How much of a vendor’s AI product was built specifically for this era? Can the system recognize events and initiate work? How deeply does it understand the customer’s context? Can autonomy be adjusted at the workflow level? Does the permission structure remain intact when agents begin taking action?

The term AI-native will appear on nearly every product roadmap. The underlying architecture will determine how much that label actually represents.

HCM survives, while implementations shrink

Jayant remains confident that HCM will continue to exist as a core enterprise software category.

“HCM as a category will continue to remain,” he said.

HR platforms contain deterministic processes where accuracy is essential. They also operate within legal, regulatory and compliance environments that standalone language models are poorly equipped to manage on their own.

AI will still compress parts of the category.

Jayant expects implementation timelines to fall dramatically. Product development will accelerate. Some workflows and use cases may disappear as AI absorbs the work.

“What took two years can happen in three months now,” he said.

As the mechanics of building software become easier, the quality of the company’s product judgment becomes more visible.

“If you're clear on what to build, building is easy. Building is almost instantaneous, right?”

The differentiator will increasingly come from the depth of knowledge inside the application and its understanding of each organization.

“Your difference is going to be context,” Jayant said.

That may be the clearest summary of Darwinbox’s bet.

The company spent its first decade building one of the newest platforms in enterprise HCM. It then concluded that the next era demanded a different foundation.

Darwinbox accepted the disruption, created an internal startup and rebuilt.

Rebirth may be painful.

Darwinbox chose it anyway.