←
Arjun Malhotra3,285 posts
Arjun Malhotra avatar

Arjun Malhotra

@BadCapitalVC

Investing in companies that don't make money @GoodCapitalVC

◷ Joined February 2012▧ 490 Photos and videos
Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·10h

One of the interesting things about VC is that while relationships take years to build, the operational work around them compounds almost immediately. (more founders, conversations, follow-ups and more info to keep track of) The funny thing is, while the work has grown, our team hasn't. Our collection of in-house AI tools now comfortably outnumbers us!

Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·Nov 17

harder to replicate moats, compounds with time and usage: - regulatory stock. years of licenses and compliance you bank before a new entrant is even allowed to play. usual in spaces with a higher barrier to entry - workflow embeddedness. once a team runs a crucial workflow via your tool, the switching cost grows every week they stay - supply lock-in. the sellers, creators or capacity that signed with you first and have no good reason to move - owned distribution. a cost base where your next customer is cheaper to reach than your rival’s easier to replicate moats but compounds if you run it harder than everyone/have a first mover advantage: - content loops and founder voice - vernacular trust. show up in the tier-2 customer’s language before anyone else and the habit sets before the copy arrives - category education. be the first to name the problem and teach the market and the association would stick

Priyal Motwani @priyal_motwani6 Nov 16 Should we obsess over moats in consumer AI? The reality we have to accept - consumer AI companies will have fewer “natural moats” than pre-AI companies. Why? Cost of building has collapsed and features get replicated in weeks. Horizontals

Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·Nov 16

The principal theory here points at a clear gap, but I don’t think it lands on all AI equally. It’s far more true for vertical AI than for agents in general. For consumer use cases, agents already win by a huge margin as the stakes of the tasks are typically low. But something specialised, like health or legal AI, is a different kind of domain. A wrong answer can cost money or much worse. That’s where I see this theory fitting since vertical AI can carry the liability and give users an assurance a general model never can. Accountability is vertical by nature.

Andrew D'Souza @andrewdsouza Nov 13 Article Nobody Trusts Your Agent If 2026 was the year of agents, 2027 will be the year of principals. Nobody has ever trusted an agent, and nobody ever will. An agent is something that acts on someone’s behalf. A principal is the one

Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·Nov 10

If you were great at judging mobile growth hacks in the 2010s, you had a very good decade. But when every platform made sure ads were the best growth hack on it, that skill would've eventually stopped paying. Most investors hit this at some point. You get good at one cycle, everything you know gets shaped around it, and then the next cycle arrives and it’s a different animal. A way to escape this inevitability, acc to @byrnehobart - is to have people at your fund who have fewer preconceptions. And while i agree, for us a version of this is to keep an evolutionary thesis which is never built before proof of work on ground & then doubling down where conviction holds. I'm sure some of what we believe at Good today will look dated in a few years too, but I'd just like us to be among the first to notice.

Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·Nov 5

The difference between having a tool and being a tool is subtle these days and it’s easy to accidentally build yourself into being a tool. Don’t be a tool.

Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·Nov 5

been building chrome extensions to make software i use less annoying – @superhuman - removes the sidebar when i narrow the window so i can see just the emails @vimcal - event previews no longer popup when moving the cursor across my calendar; it needs an actual click

Arjun Malhotra avatar
Arjun Malhotra@BadCapitalVC·Nov 3

Wut

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…

Post attachment