Good Capital

"If you make a big change but you do it halfway, you lose what you're anyway losing. Without the gain." In the third episode of #LegendaryPivots, Meesho's Vidit Aatrey explains why the hard part of a pivot is not the call; it is everything that comes after it. Meesho was the leader in the reselling category, at roughly $1 billion in GMV, with growth capital raised specifically to scale that model. The evidence for change came from the data: consumers signing up as resellers and buying only for themselves, a share that rose every month through the pandemic. Vidit says the decision itself took no time. Everything else took two years. People had joined for the reseller mission. The concern inside the company was that a consumer business would compete with the sellers it had spent years bringing online, at the peak of the pandemic, when Meesho was the only income many of them had. It took two months of town halls before the switch and roughly two years of public doubt after it. Before it was clear the move had worked. Watch the full episode: https://lnkd.in/gdJ3v94E

Orange Health Labs reported Rs 138.6 crore revenue in FY26, up 65% from Rs 84 crore in FY25. The diagnostics startup’s loss before tax increased 11% to Rs 97 crore, while a deferred tax expense pushed its overall loss to Rs 146 crore. Total expenses rose 36% to Rs 240.3 crore, led by employee costs of Rs 72.2 crore and diagnostic material costs of Rs 42.9 crore. Orange Health’s EBITDA loss stood at Rs 92.4 crore, although the company claims to have achieved 20% EBITDA profitability in Bengaluru. The company added 60+ company-owned collection centres during FY26 and is now raising $30 million to support its expansion across Tier I cities.

Entri has built its scale by taking career-focused learning beyond English-first classrooms, reaching millions of learners across Indian languages. Its collaboration with IPTIF brings that distribution and learning engine together with IIT Palakkad’s deep-tech expertise. We are excited for the partnership as the result could be emerging-technology courses that are academically grounded, career-oriented, and accessible to learners well beyond traditional technical institutions. This is a meaningful expansion of what Entri does best: opening up access to skills that can lead to real economic opportunity. Congratulations to Mohammed Hisam and the entire Entri team! 🚀

A decade ago, the ambition for most Indian startups was to build above the mess - put a clean app in front, rent the operations underneath, stay light. But looking back at what actually compounded in our portfolio, the opposite is true. The companies that lasted went down into the complexity and chose to own it: → Orange Health Labs runs its own accredited labs and rider fleet, so a sample stays inside one company from your arm to the report. A late or wrong report has exactly one owner, and that accountability is the product. → SolarSquare owns installation and after-sales service in a market where install quality is inconsistent and service is mostly fictional. That control is what turns homeowners into referrals - 90% say they'd recommend it. → LEAD Group owns the full classroom stack, from curriculum to teacher training to assessments, becoming the school's operating system. Removing it would mean rebuilding the school's entire day, which is why partnerships endure for years. → Meesho grew on social trust early, then made low prices sustainable at scale by owning fulfilment via Valmo. None of this is asset-light, and when AI crushes the cost of building software, code alone no longer defends anything. What AI can't commodify is the physical operating layer underneath - the labs, the installer networks, the classrooms. And it's now what makes those heavy businesses scalable in the first place. Full piece on supply-first models in an AI world 👇

We're trying to run as much of the firm on AI as we can internally. So I thought I'd share where my own setup has gotten to: My day was spread across email, calendar, WhatsApp, Granola, to dos, trip plans and most of the friction was just me moving between all of them, everyday. So we built a dashboard that pulls from all of those sources into a single place. It works like a card feed, and I swipe right if an action is done, left if it's deferred, and I can add notes if something needs to be picked up by the team. It refreshes 3 times a day: morning, midday, and end of day, as real time is expensive and I don't really need it. The triage rules are simple: LP comms are higher priority, cold inbound goes to the team, warm intros come to me, newsletters get routed to my readwise. Interestingly, the model turned out to be the least important piece when we were building this. There are 4 custom layers underneath: memory, context, skills, and tools. The model sits on top and is basically swappable. Everything that took us months to get right lives in those layers, so switching from one model to another is close to a one-line change. There's plenty still to figure out. Right now it's running at something like 2 in 10 false positives, so there's always a human in the loop, and nothing goes out until I've approved the draft myself. Even with that, it's insane how much time it gives back. I'm not switching between 6 apps a day anymore and that alone was worth the build. Here's a dummy reference:

🍊 For a company like Orange Health Labs, regional expansion is much more than adding a city to an app. The promise has to be built on the ground - through local teams, fast sample movement, reliable operations, and neighbourhood-level trust. Great to see Orange Health expanding its footprint across Mumbai & Pune while meeting customers right where they are. 🚀

Some movements deserve an anthem. 🇮🇳 India’s energy independence is being built one rooftop, one household, and one conscious choice at a time. Proud to see SolarSquare give voice to this movement and to the people powering it. ☀️

Some obvious & non-obvious reasons I think AI agents may not have really been widely adopted yet, even though the tech is ready: 1/ It's not prompt in, answer out. An agent is a process you set up and steer while it runs, and the chatbot muscle memory most people have doesn't transfer. 2/ Prompting an agent is closer to writing a spec than asking a question. You have to scope the task extensively and define what "done" looks like. 3/ As Paras puts it, this needs a lot of delegation, which is a hard soft skill to build. It's very close to managing an employee, and most people have never done this. 4/ A lot of the actual power still lives inside Codex or Claude Code, which is terminal-shaped and a little technical. You have to be comfortable doing the messy setup, so it self-selects for a narrow crowd. 5/ One agent is also just a tool. The unlock is running several at once and getting them to talk to each other like a team, and that handoff between agents is still mostly DIY. 6/ Same problem across people. Your agent's context has to reach your colleagues, or everyone ends up working in silos, and right now that process is way too manual. 7/ Trust is a ratchet. A chatbot that's wrong wastes 10 seconds, but a wrong agent sends the email or edits the important file. The downside is asymmetric, so most people keep it on a short leash. 8/ Lastly, there isn't a job-to-be-done the public actually feels yet. Autonomous agents will always be a solution looking for a problem.