Entire Bets India Can Help Solve the Infrastructure Problems of AI Agents
"If you drop a pin somewhere in India, there's a good chance you'll find somebody working on a particular technology, maintaining an open-source project or specialising in a particular area."
India’s importance in the next phase of AI-powered software development may not come simply from having one of the world’s largest pools of developers. Its bigger advantage could be the unusual overlap between enterprise engineering teams, open-source communities and developers who move between both worlds.
That is increasingly attracting companies building infrastructure for an era in which software is no longer written only by humans.
Entire, the startup founded by former GitHub CEO Thomas Dohmke, is betting that the transition from AI-assisted coding to agent-driven development will require a fundamental rethink of the tools developers have relied on for years.
If one developer can now deploy multiple AI agents—and those agents can spawn sub-agents—the volume of code being produced, reviewed and committed can quickly overwhelm workflows which were fundamentally designed with humans in mind.
Moreover, traditional Git workflows largely preserve the destination, the code and its changes, but not the journey that produced it. For enterprises, that missing context could become increasingly important.
As AI-generated code moves into production, organisations will need to know not just what changed, but why it changed, which model and tools were involved, who reviewed it and whether the resulting audit trail can be trusted.
That is the problem Entire is attempting to tackle, while also rethinking the infrastructure beneath Git for a world where machines generate and move code at a scale humans never did.
In an interview with The Left Shift, Karthik Rameshkumar, Field CTO at Entire, explains why the company believes Git-based workflows need to evolve, why machine-level concurrency could put pressure on existing infrastructure, and why India has become a particularly important market for the company.
Edited Excerpts
Karthik, let's start with the basics. What exactly is Entire, what developer pain points is it trying to solve, and where does it fit into the broader AI software development ecosystem?
Karthik: Anyone who tells me we're trying to build GitHub again, my core point to them is that GitHub does what it does really well. There's no point in building GitHub again.
GitHub was originally built when human developers were at the centre. The pull request, for example, was fundamentally designed around a human developer writing code, committing it and then asking another human developer to review it.
The problem statement we're dealing with today is quite different. When human developers were writing code, the pull request was an amazing and foundational unit of work. It gave everybody something to review and provided a way to attach automation around that work.
But now imagine a developer working with agents. You could have 15 different commits coming from 14 sub-agents and three worker agents, all eventually being committed into one pull request. A senior developer could suddenly open that pull request and find 6,000 lines of code changed.
So the volume of code has increased dramatically. Reviewer stress is real, and there is now a gap between the amount of code that agents can generate and the number of senior developers who can actually review that code effectively.
The bigger problem is that we are only capturing the end result. We capture the code that eventually gets committed, but we don't necessarily capture the journey that the agent took to get there.
Developers have their own ways of working with agents. They have prompts that work better for particular tasks, tools they prefer and decisions they make along the way. Agents are also making important trade-offs and decisions at every stage.
Today, very little of that is captured. At GitHub, for example, there isn't really a way to capture the entire journey an agent takes to arrive at an answer. You get the final code and, if you want, you can look at the resulting diffs.
That creates another problem for AI systems reviewing the code. The reviewing agent gets the final code and has to reconstruct a lot of the context itself by searching through the codebase. That consumes tokens, throughput and bandwidth unnecessarily. So Entire is trying to solve that problem while also looking at the infrastructure underneath it.
The Left Shift Bureau
From an enterprise perspective, why does that infrastructure need to change?
Karthik: Two kinds of concurrency have become important– human-level concurrency and machine-level concurrency. Traditionally, platforms were architected differently depending on whether the system was being accessed by a human developer or by machines making automated requests.
But now imagine every developer having five or six agents. Each agent can have its own sub-agents, and each sub-agent can be working on its own task, worktree or project.
You suddenly have a multiplication factor. One developer could effectively have several developers working underneath them. That's putting enormous pressure on existing infrastructure.
The third layer we're trying to address is therefore the Git network itself. We wanted to build an open, decentralised and independent Git network that can operate in a much lighter and more efficient way.
We've already stood up a cluster in Mumbai. It's an India cluster that developers in India can clone from. They can mirror their repositories there and start working.
It's really the sum of all these parts. An experience isn't an isolated piece. You have to look at every part of the software development lifecycle, every surface the developer touches and every point that causes friction.
Then you have to figure out how to turn that pain point into something that creates developer joy. That's the mission we're trying to build towards.
Enterprises already have GitHub, GitLab, Copilot, Claude Code, Cursor and countless other tools. What's stopping an enterprise from building something like Entire itself?
Karthik: One foundational difference is that we're building Entire with a very strong open-source foundation. The Entire CLI is open source under an MIT licence. We could have built a system where all the checkpoints were stored in one proprietary database somewhere, but that's not what we chose to do.
The data from checkpoints is versioned in your own repository, wherever you choose to keep it. You can then decide how you want to visualise that data.
You can also mirror it with Entire and get a more visual representation as a graph or timeline, with filters and other capabilities. But the choice remains with the developer or enterprise.
We intend to remain an open platform at the foundational level. Entire Graph is also open source under an MIT licence, and we're working towards open-sourcing the underlying engine that powers our Git network, which we call Entire DB.
The idea is that we build these foundational primitives in the open, let the community tell us what works and what doesn't, and then build the enterprise capabilities on top.
That's how we see differentiation happening. If the community wants to take those primitives and use them without ever touching an Entire surface, we're perfectly comfortable with that. At the end of the day, developer joy comes first.
Pritam Bordoloi
From a technical perspective, what are some of the biggest problems you need to solve to make this infrastructure ready for AI agents?
Karthik: One of the biggest things we've realised is that the infrastructure underneath the major software forges has become highly centralised over the past two decades. That made sense when developers were working at human scale.
Take a developer sitting in Coimbatore, a city in India where I live, for example. Their Git push might travel all the way to Virginia before being committed, even if their employer is based in Electronic City, Bengaluru.
When you're a human developer, that's generally not a major problem. You clone once, work locally and perhaps push once or twice a day. Agents, however, operate very differently.
You don't want an agent to push code to a remote repository once or twice a day. Small, incremental units of work being committed to the remote quickly are becoming part of how agentic development works.
Now multiply that by every agent and every sub-agent working on a developer's tasks. The latency that didn't matter at human scale suddenly becomes much more important at machine scale.
A human can wait a couple of seconds. They can take a sip of water while something completes. An agent doesn't work that way. So when you design infrastructure for agents, you have to rethink those assumptions.
Entire also captures the agent's journey. We've recently seen reports of AI agents going rogue or making unexpected infrastructure changes. Could that become a major selling point for Entire, particularly with enterprises?
Karthik: Absolutely. If you've worked with enterprise software for a long time, you know that audit trails are fundamental. Compliance frameworks such as ISO and SOC 2 require organisations to maintain records of changes, checks, approvals and access.
You need to know who checked whose code, when it was submitted, who viewed information and what happened at different stages.
With AI-generated code, the same principle becomes even more important. From the business leaders we've spoken to, being able to attest to AI-generated code — what was written, how it was written and who signed off on it — is becoming a significant requirement.
Today, even with the open-source CLI, you can ask questions such as: Who wrote this function? When was it written? Which model was used? Which commit and session was it associated with? What was the prompt? You can also see which tools were used and which files were accessed.
The next step is creating a layer that allows enterprises to trust that information. It's not enough to say that a record exists. Enterprises need confidence that the record is legitimate and hasn't been manipulated.
If we can give enterprises that trusted trail, they can understand what an agent did, why it did it and what ultimately happened. That could make enterprises much more comfortable adopting AI at scale.
There is still resistance in some organisations because people don't know what happened when something goes wrong. If you can give them a complete trail and allow them to backtrack and understand the decisions, that changes the equation.
Since much of the platform is open source, what does the business model look like? What remains free and what becomes an enterprise feature?
Karthik: We're working through multiple areas of that and will have more announcements in the coming months. But our approach to the primitives will remain broadly the same. We want to build those foundational components in the open and allow the community to tell us what works and what doesn't.
Take checkpoints again. Today you can already ask who wrote something, when it was written, what prompt was used, what tools were called and which files were accessed.
A strong enterprise feature could be independently signing all of that information so that an organisation can trust the record. That's an example of where enterprise capabilities can add value.
We're also taking a design-partner approach. Rather than building enterprise software in isolation and then going to customers with a product and asking them which of a thousand customisation knobs they want to turn, we're building with customers.
We launched a design partnership motion alongside our India launch. We're working with multiple design partners in India, including large enterprises that are testing the products and telling us what needs to change.
We're also working with large open-source projects in India. The developer community here is incredibly vocal. They'll tell you immediately when something works and when it doesn't. That's exactly the kind of community we want to build with.
Pritam Bordoloi
Why is India so important for Entire? Is it primarily because of the developer ecosystem, or do you also see a large market opportunity here?
Karthik: India's developer ecosystem is unique. It's incredibly fast-growing, but that's not the only reason. What's interesting is the diversity of talent and capability across the country.
At almost every skill level and in almost every specialisation, you can find people doing interesting things. If you drop a pin somewhere in India, there's a good chance you'll find somebody working on a particular technology, maintaining an open-source project or specialising in a particular area.
After the US, I would put India among the geographies where that concentration of talent is particularly strong. That's one reason Entire deliberately chose to have a field CTO presence out of India.
I was also the company's first revenue hire, and the thinking was that India could bring not just commercial value but also community and developer feedback.
There's another thing I find fascinating about the Indian developer ecosystem. You can have someone attend a community event on Saturday, contribute to open source, mentor other developers and review pull requests — and then see the same person at an enterprise event representing their company.
They have this split persona. They're enterprise developers during the week, but they are also members of the broader developer community.
That vibrancy is very valuable to us. India is important to us from both sides– firstly, the developer community that can push the product forward and secondly, the enterprise ecosystem that is willing to try new technology early.
Can you give examples of how developers or enterprises are using Entire today and how their workflows have changed?
Karthik: We already have a lot of feedback from the open-source community. Some well-known open-source developers have left feedback through pull requests and issues, and some projects with thousands of stars are already using Entire.
One thing we're hearing consistently is that people like the fact that we're preserving context and intent that previously disappeared. In large open-source projects, you might have a contribution where you can see the code but not necessarily understand the full intent behind how it was produced.
With checkpoints, that context can be preserved wherever the project chooses to keep it. We're also seeing the community directly influence the product. For example, we didn't previously have the ability to track the work of sub-agents operating inside an agent. The community helped us build that capability and gave us feedback on how it should work. That's what open source should look like.
What does that translate into from a productivity perspective?
Karthik: One example is multi-agent reviews. Developers can start enabling reviews where multiple agents look at the work, rather than relying only on the coding agent that produced it.
Those agents can also review the work against the original intent and prompt. That gives developers a much richer review experience than simply looking at the resulting code. As the platform becomes more widely used, we'll see more of those stories emerge.
You speak to developers and business leaders across markets. How much agent adoption are you seeing among Indian enterprises? Is it really happening at scale, or is most of it still in the pilot and proof-of-concept stage?
Karthik: It's extremely diverse. There are enterprises that are very far ahead on the adoption curve, while others are still anxious about touching certain surfaces.
A lot of that comes down to compliance, regulation and governance. Financial services is a good example. Organisations in that sector have multiple compliance requirements that they have to satisfy before deploying a new tool.
Government technology has similar challenges. But despite that, I think the conversation among enterprise leaders has changed.
A few years ago, the conversation was, "Show me the value." I don't think that's the primary conversation anymore. Now it's more like, "How much value can you show me compared with someone else I'm already working with?"
There's a huge diversity of agent tools and approaches available today, so organisations can make choices much more quickly.
If an organisation doesn't have significant regulatory boundaries and can innovate while keeping its customers safe, we're seeing some companies push the envelope very quickly.
India also has a unique economic incentive. A lot of Indian technology organisations have historically competed through cost arbitrage. But that advantage is changing because agents can perform some of the work that previously required large teams.
So leaders are asking a different question: How do we use AI to multiply what our existing teams can accomplish? I think that makes India a perfect storm for AI adoption.
Pritam Bordoloi
What are Entire's plans for India over the next two or three years? How large is the team today, and do you plan to expand engineering, partnerships and go-to-market here?
Karthik: Entire is still very early. We have around 40 people globally today. We intend to find the best developers wherever they are. It doesn't matter whether they're in Karachi, Istanbul, India or somewhere else.
One example I love is a 14-year-old developer in Ohio who became a maintainer of Entire Graph. He contributed to the project through open source, kept building and eventually became an engineering intern with us.
That's the kind of talent we're looking for and I have a strong conviction that a lot of that talent exists in India.
We're already looking at multiple individuals and working on plans to expand the team here. We have a strong intention to build a team in India, both from a go-to-market perspective and from a community perspective.
It's not just about India as a market. India is a place where a lot of the skills we need exist. So if you're a developer reading this and you want to be part of Entire, go to our open-source repositories. Contribute. Show us what you can build with agents. Show us your skills, and you could potentially be part of Team Entire.
Finally, what about partnerships? Do you see Entire eventually working with large IT services or consulting companies such as Deloitte or Accenture to take the platform to a broader set of enterprise customers?
Karthik: We're already working with partners. There are multiple partners we're working with, although I can't disclose their names yet. Some of them are incredibly large organisations, and we're already discussing multiple areas of collaboration. There will be announcements over the coming weeks and months.
We're treating these partners as design partners as well. They bring a wealth of enterprise experience that we might not necessarily have ourselves. Their architects have worked across financial services, aviation, energy and other industries, and each industry has different compliance and governance requirements.
So we're learning from those conversations. What governance requirements do enterprises have? What needs to be built into the platform? What does an enterprise-grade agentic development workflow actually need? Those conversations are happening right now, and we're actively building around that feedback.
Pritam Bordoloi
Pritam Bordoloi
