DevRev claims it approaches the integration of data for AI purposes in a new and better way compared to everyone else in the market. To understand what the company wants to achieve, we recently sat down with Dheeraj Pandey, one of the founders of DevRev and its CEO, at DevRev’s Leadership Circle event in Amsterdam. “We’re not coming up with a new problem. We’re just saying AI is big challenge, and the headache is integration, and we’ll left shift that so you don’t need system integrators to do that integration.”
Pandey is a very familiar name to anyone who has been following enterprise IT for a while. He was one of the founders and the former CEO of Nutanix. Given the success of Nutanix, and the contribution he must have had to that success, a new company that Pandey founds (together with Manoj Agarwal, also ex-Nutanix) deserves our attention.
Founded in 2020, the past few years have been mostly about finding the right message for what DevRev wants to bring to the table. As it is trying to do something that has not been done before, that’s not always easy. It is equally challenging for us to make sense of it too. However, things are becoming clearer and clearer, so it is time now to dive into what DevRev built and has to offer.
DevRev takes a different approach
From a high-level perspective, DevRev fundamentally rethinks the enterprise software stack for an AI-first world. It doesn’t treat AI as an incremental enhancement or add-on to existing SaaS products, but builds a new foundation centered on what Pandey calls “enterprise memory”. This is a unified knowledge graph that integrates data from disparate sources while maintaining bidirectional sync with legacy systems.
The enterprise memory foundation that DevRev has developed is SaaS, but is not SaaS at the same time too. That is to say, it operates at the same level as SaaS, but differs from it in fundamental ways too. We will get into that a bit more later.
Data integration: the old and the less old
So, what is the problem that DevRev is trying to solve here? Pandey starts our conversation by identifying what he sees as the core challenge that has persisted for five years: integrating disparate SaaS sources into a unified knowledge graph. “AI’s real challenge and opportunity is integration,” he says. This is something system integrators can do, or you can do this “just in time”. The latter is what a lot of conversational apps are doing. “They’re using MCP and talk to legacy SaaS. But our belief is that you have to redigitize data”, Pandey argues.
According to Pandey, then, MCP isn’t the answer to the integration challenges posed by AI. He compares what MCP helps us do to having a music app on your phone that wirelessly controls a robotic arm changing CDs in a physical CD player. It works to a certain degree, but it’s far from ideal. If you really want to move forward, you digitize the CDs to iTunes or any other piece of software that is actually on the device you use to control it. That is exactly the task DevRev wants to undertake, but for enterprise data. That layer needs to be fundamentally restructured for AI consumption rather than simply accessed through existing APIs (which MCPs are too, to large degree).
APIs were designed for human agents working through graphical UIs in Salesforce or ServiceNow, not for autonomous AI agents. Critical capabilities that agents need, such as rolling back an errant session or performing point-in-time recovery, simply don’t exist in legacy API architectures. “There’s no such API for that in Salesforce,” Pandey claims.
Enterprise memory is the foundation
In order to achieve this redigitizing of data, DevRev focuses on creating what Pandey calls “enterprise memory” or “computer memory”, already mentioned above. This unified layer brings together data from multiple SaaS systems into a knowledge graph. Important to note here is that this doesn’t mean migrating data away from source systems. “We leave the data where it is. That’s the primary copy. We make a second record,” Pandey clarifies.
More important still is what Pandey says next: “We do two-way sync so we can actually sync it back to legacy environments.” This is important because in his view the transformation from SaaS with AI bolted on, as it is now, to AI will not be done very quickly. “This will be a five year journey”, he states. With two-way sync, the data remains the same on both sides. That is extremely important when parts of the business run on on side and parts on the other, but both need to access the same data.
In other words, organizations can’t simply rip out their Salesforce, ServiceNow, and Zendesk implementations. Instead, DevRev wants to enable a gradual transition where new AI-driven experiences coexist with traditional SaaS interfaces, with changes flowing bidirectionally between the two worlds.
It is this “memory” that DevRev offers organizations that makes what the company does relevant from an architectural perspective. Pandey isn’t all that interested in the reasoning part of AI anyway: “Reasoning will commoditize as everybody just keeps throwing more hardware at the problem. The thing that’s unique to you will be memory.”
DevRev Computer
Now that we know that DevRev offers enterprises a memory for their data, the question arises what that actually brings them. Without anything else, not so much. This is where DevRev Computer comes into the equation. Computer is what Pandey calls the “conversational experience” DevRev offers customers. The idea is that conversational apps are the new convergence of all business software in the age of AI. Computer offers a conversational interface for interacting with business software.
Rather than logging into separate applications for sales, service, support, and development, Computer lets users engage through natural language with an AI that can access and act across all these domains. Combined with the enterprise memory we talked about before, this convergence on the user side creates a new stack, Pandey argues. The integration of data and the memory that comes with it is the back-end, Computer the front-end.
In order for Computer to give AI agents the right answers, DevRev added something else, namely text-to-SQL. Pandey readily admits that he didn’t believe in text-to-SQL until about a year ago. The technology really took off since then. It gives DevRev (and Computer) the ability to turn unstructured data into structured data on the fly. That in turn should lead to much better answers. That’s what Pandey wants Computer to be, an answers engine rather than a search engine. “You want to use LLMs to generate SQL on the fly, but really go to databases as opposed to go to unstructured documents because the amount of hallucination will be too great.”
Also read: DevRev Computer: AI assistant becomes AI colleague
Conversational interface is a process
Even though Pandey is sometimes quite philosophical in how he speaks about what DevRev is trying to achieve, he is also very realistic. The conversational interface for Computer will not be adopted by everyone overnight. “The graphical UI is not going to disappear overnight. 20 to 30 percent of specialists still want to point, scroll and click”, he says.
For other people inside organizations, though, the conversational approach that Computer wants to bring, with its focus on answers rather than search, could potentially bring a lot of benefits. “Even executives that dreaded SaaS can now talk to their data, without being at the mercy of middlemen who are data analysts”, Pandey says. According to him, that only became possible once DevRev tackled the text-to-SQL problem mentioned above.
Apps are still there, but very targeted
In order to offer something to both sides of the UI vs. text-to-SQL debate, DevRev cannot get away from SaaS apps completely (yet). That’s why it built a couple of apps that interact with Computer: Build, Support, and Observe. These help specialists at organizations support their customers better, build products faster and improve user experience for customers. These apps should take care of the 20-30 percent of specialists who still want to point, scroll and click, Pandey points out.
Given the allergic reaction Pandey seems to have when it comes to SaaS, it might seem surprising that DevRev offers these types of apps. Should they be considered as SaaS apps? According to him, that isn’t necessarily the case. The pricing, for one, is different. There’s no subscription, DevRev prices is by consumption and outcomes. Besides that, it’s mostly just another way of getting to the right answers through Computer. Pandey compares it to the usage of mobile versus computers with larger screens. Most of what we do, we do through mobile apps, but about 20 percent we need a big screen. In other words, the apps are simply a necessity, but DevRev wants to make them as un-SaaSy as possible.
Built-in resilience
One of the issues with the world of agents and autonomy when it comes to data is that agents are not infallible. “Agents will get autonomous and they will make mistakes, way more than humans did as operators”, Pandey states. That’s not because agents are inherently worse performers than humans (which would make for a bad business case, of course). It’s just a matter of scale. There will be way more agents than humans, so in absolute terms the numbers of mistakes will be greater too.
For DevRev, this is an important piece of the puzzle they are putting together. In the words of Pandey: “There’s one thing that we are very excited about and that is data safety.” He compares the point we’re at when it comes to securing data in a world of (autonomous) agents to the 1990s and 2000s. That was when things like rollbacks, point in time recovery, back-up and restore came up to make sure data was protected.
He reiterates here that legacy APIs have “no notion of time travel or rollback”, whereas the technology DevRev developed does. Of course there’s the concept of rollback in something like Salesforce, but not for a specific session of an agent that went rogue, he nuances his position after we ask him to elaborate. “Safety is an important puzzle, because if agents become autonomous, memory needs to be recoverable.”
How to get started
Conceptually, the story that Pandey tells us is quite compelling. Having a layer (at the SaaS level) that harmonizes all the data and makes it fit for use by AI agents is something many organizations will be keen on. But how do they get started with this? Do they have to do a lot of work to prepare for deploying DevRev/Computer? Or does DevRev have a “magic button”?
Based on what we described above, the adoption shouldn’t be too hard. That is, if roughly 70 to 80 percent should be able to work based on the conversational interface alone, it should be relatively easy to put the application into the hands of employees. They can then start on what Pandey calls their “crawl, walk, run” trajectory. These three terms loosely align with another collection of terms he uses regularly, which is “search, answers, actions”. So even though DevRev at the moment is all about answers, people still need to start with search, even if only for a short while. Actions is the next phase, also for DevRev.
All of this sounds nice, but is still not very concrete, other than “give your people access to the application”. However, they can start using it very quickly too, we hear from Pandey. In other words, there is a magic button, and it’s called AirSync. “We connect it to all your legacy systems, we bring it here, we index it, re index it, organize it the way we need to. It’s all one click in that sense. And by the way, if you’re going to make changes here, we’ll reflect it back there as well”, Pandey says. This means there are two copies of the data. If you change something on either side, the change will also be done on the other side.
Quick wins
Customer support represents the most obvious initial use case for what DevRev has to offer. The reason for this is simple. In customer experience, the number of services an organization offers is relatively limited, so it’s not a very complex environment. On the other hand, the number of users is very large. This means that AI agents can reduce the number of actions humans have to take significantly, without the AI agents being befuddled by an environment that’s too complex for them. The conversational experience DevRev brings to the table can sit alongside existing Zendesk or Salesforce implementations, Pandey says. That way, it can prove its value before organizations commit to DevRev’s conversational approach fully.
On the topic of investments in AI, Pandey argues we should evaluate them differently than we are used to for other parts of the enterprise stack. Rather than considering tool costs in isolation, he argues organizations need to look at payroll plus tool costs together. “People are looking at payroll cost plus tool cost together for the first time”, he says. That means it’s not about getting extra budget for tools when somebody wants to make the pitch to go for DevRev. “It’s also about making current people more productive, which means they don’t have to hire more people, or they can reduce the number of people they have”, Pandey argues.
This reframing suggests that the four and a half trillion dollar market for AI extends well beyond the traditional $300 billion SaaS market. It encompasses labor costs that can be reduced or reallocated as agents take on more work. The pricing model shifts from per-seat subscriptions to consumption-based or outcome-based pricing that better aligns with the value delivered.
Integration and convergence without lock-in
Integration assumes convergence into a central place, in this case DevRev Computer, more specifically its memory. What does that mean in terms of lock-in? That’s not something organizations usually want, even though a little lock-in is generally acceptable.
When we ask Pandey about this, he once again stresses the bidirectional sync capability. Organizations can take their data out of DevRev at any time. More importantly, the real-time two-way sync means they never fully leave their existing SaaS systems behind in the first place. Mind you, the goal is to replace those systems, so that is a temporary state, we assume. However, even if organizations replace their existing SaaS with DevRev, they can always take their data out.
SaaSpocalypse may originate in SaaS layer
The SaaSpocalypse as it has been reported on by most people is never going to happen, in our opinion. That is, vibe coded software will never replace SaaS solutions. Sure, it may be possible to build something that works (for a while) and more or less does what it needs to do. And for some solutions, that may be perfectly acceptable. For big SaaS solutions we don’t see that happening. Those stacks are much more difficult to build, and, just as important, to maintain.
In a way, however, what DevRev is doing could also be seen as contributing to the end of SaaS as we know it. That is, if the company is successful. Converging all enterprise data into one conversational experience (Computer) that sits at the same level as SaaS solutions, has an enterprise memory and enables AI agents to access the data they need in a secure and well-governed way, certainly sounds good. The fact that it can be deployed relatively easily in existing environments with lots of legacy, makes it potentially even more interesting. DevRev also makes promises about not locking in customers, so they don’t run the risk of going from one locked-in environment to another.
DevRev operates in a part of the industry that changes quickly and fundamentally, though. Incumbent players like Salesforce, ServiceNow, Zendesk and many other SaaS solutions are also in constant flux, mainly because of the emergence of AI and AI agents. The fact that both Salesforce and ServiceNow have made moves recently to become headless indicates that they also see that their “point, scroll and click” front-ends may not be a vital part of their future. It’s going to be interesting to see how DevRev will find its way in this ever-changing landscape.