When Mukund Jha co-founded Dunzo — one of India’s most recognised hyperlocal delivery startups — few would have predicted his next act would involve shipping an AI platform that renders coding optional. But that is exactly what happened. In 2025, Mukund and his twin brother Madhav Jha launched Emergent, a platform that lets anyone turn a plain-language idea into a production-ready application without writing a single line of code. Eight months later, Emergent had crossed $100 million in annual recurring revenue, with more than 8 million builders across 190+ countries deploying software on the platform.
That momentum brought a harder question: what comes after building? The answer, Jha decided, was running. Last month, Emergent launched Wingman — an autonomous AI agent that lives not in a dashboard or a standalone app, but inside WhatsApp, Telegram, and iMessage. It schedules meetings, runs sales outreach, drafts content, preps briefings before calls, and manages your inbox — continuously, in the background, through the chat interfaces that founders already use. LAFFAZ covered the Wingman launch when it dropped; what follows is the conversation behind it.
The company is backed by Khosla Ventures, SoftBank, Lightspeed, Y Combinator, Prosus, Together, and Google’s AI Futures Fund, with $100 million raised in total.
LAFFAZ sat down with Mukund Jha to understand the internal logic behind Wingman, how Emergent thinks about trust in autonomous systems, and what the rise of AI agents really means for human coders.
LAFFAZ: Wingman feels like a major expansion from Emergent’s vibe-coding roots. What was the internal insight that convinced you the next product should be an autonomous agent rather than another layer of software-building tools?
Mukund: Emergent was built on a simple premise: that the best technology should be accessible to everyone, not just those with technical backgrounds. Wingman is the next step in that vision, applying the same agentic engineering foundation to autonomous agents. We know that professionals can lose hours of valuable time to day-to-day admin, such as outreach, task delegation, meeting prep, follow-ups, and general management.
We designed something to give users those hours back so they can focus on driving results and serving clients. By relieving the mental load that scheduling and digital discourse require, people can reclaim that time and direct their attention towards growing for the future.
LAFFAZ: Users want autonomy but also peace of mind. Can you walk us through how Wingman’s “trust boundary” works in practice? For example, how does it decide what tasks are low-risk enough to handle independently, versus those needing user approval—and what’s the process if it makes a misjudgment?
Mukund: Our trust boundary is essentially how Wingman distinguishes between low-stakes tasks that can be automated and consequential actions that still require input from the user. So for routine tasks, such as moving a meeting time or generating a research summary, the logic is binary, predictable, and confidence is high. However, sending messages to large groups or modifying important data are considered to be high-risk requests where the agent must interpret intent, meaning that it will prepare the actions, but will need the green light from the user in order to proceed. These trust boundaries can also be established by the user upon sign-up, ensuring that they are always in control and can specify any task they would like to directly oversee.
LAFFAZ: Wingman builds memory across sessions and adapts to your tone and communication style. How do you design this personalization to feel genuinely helpful and reliable, without coming across as intrusive or overly dependent on user trust or preference?
Mukund: Wingman adapts to a user’s ‘normal’, which means that they never have to start from zero. It retains context, saves routines, and recalls preferences across sessions, meaning that every new session is a continuation of the last. This means that the user doesn’t have to be overly specific in their prompts as the agent quietly observes their behaviors and evolves to mirror them. However, while Wingman notices these patterns, it collaborates with and seeks verification from the user before acting autonomously. Instead of being overly reliant on user preferences, the agent is trained through application and analysing feedback, which makes Wingman feel more like a trusted operator rather than another tool to manage.
LAFFAZ: A lot of AI agents promise broad utility, but many struggle with ambiguity and edge cases. Where do you see Wingman’s current limits today, and what kinds of tasks still require a human in the loop?
Mukund: Although being able to act autonomously for more routine tasks, Wingman always requires human insight in order to carry out more ‘high-risk tasks’. How this works is that the agent will calculate the risk associated with a task by detecting intent and priority, and then validate the action with the user before executing. So humans will still need to provide their input for any outbound actions, such as sending an email or meeting invite, but this also means that they are in complete control of their professional admin. Because Wingman lives in the apps that humans use every day, such as WhatsApp or Gmail, it can observe your behaviours and establish patterns but it will always clarify with the user first. Wingman also reconfigures these permissions for every user, so any disruptions will redirect the agent back to you for final confirmation.
LAFFAZ: From a market perspective, Emergent’s vibe-coding hit $100M ARR in 8 months. But Wingman pivots to task bots—what if users stick to free agents from OpenAI or Anthropic? How do you win?
Mukund: What separates Wingman from its competitors is that it is built for how people already work. By embedding itself into WhatsApp, Telegram, and iMessage, it eliminates the biggest pain point in AI adoption and meets people in their existing workflows. There’s no app to download or onboarding to sit through; you can open a chat and start delegating. Most AI agents today live inside developer environments or standalone platforms that require technical setup or conditioning to operate in a way that is most efficient and effective for the user.
The most successful AI agent will be the one that founders and businesses actually use, which is exactly what this platform was built for.
LAFFAZ: Vibe-coding agents like Wingman let anyone prompt apps without writing code. They’re clearly helping non-tech founders, and Indian startups are already using AI for 40-80% of coding. But how do you see these tools impacting human coders?
Mukund: The human impact of AI is not something to dismiss. It’s a real concern and one that I take very seriously. However, I believe that AI tools can be an extension of talented coders rather than replace them. AI can help streamline operations and make coders dramatically more productive, meaning they are able to take on more ambitious projects and redirect their focus onto measurable outcomes instead of arduous processes.
AI is redistributing opportunities to people who never had access to them, as opposed to removing the need for human input.
Companies with growth ambitions tend to expand what they build rather than shrink their teams, so they must design systems that can evolve along with AI innovations and complement human potential.
What comes through in Jha’s answers is a consistency of principle rather than just product positioning. The same thesis that drove Emergent — powerful technology should reach everyone, not just developers — runs directly through Wingman’s design logic: no new app, no setup, no learning curve. Just a chat and a delegation.
Whether Wingman can hold that promise at scale, against well-capitalised competitors from OpenAI, Anthropic, and Google, remains the open question. But Jha’s track record — from Dunzo’s hyperlocal infrastructure to Emergent’s eight-month sprint to nine figures in ARR — suggests that this is a founder who moves faster than the debate. The next phase of AI won’t be won by the most capable model. If Jha is right, it will be won by the one already sitting in your inbox.
