Aditya Kajla

Co-founder @radialhq 🛠️ prev co-founder @warrant_dev (YC S21, acq'd), eng @ WorkOS, Lyft, Medallia & Yahoo. Also a big Lakers fan 🏀

[Aditya Kajla @akajla09]

@KaranKajla and I are back to building and excited to announce @radialhq, a customer experience platform built from scratch for our new AI-native world.

Customer service still sucks. Support teams try to keep up, but the traditional model of hiring and training more agents doesn't scale. AI was supposed to solve this problem, but first-generation tools stop at FAQ chatbots and basic tier-1 automation.

Unlike existing AI support platforms that focus on deflection, Radial aims for full, autonomous resolution. Our AI agents integrate with existing systems and data to investigate issues and take action, all while operating within company policies.

Much of what we're crafting into Radial comes from our experience building and supporting Warrant. As founders, we spent countless hours supporting developers integrating with Warrant APIs. Radial is the support platform we wish existed back then.

We're now in beta and working with early adopters. Reach out!

[Aditya Kajla @akajla09]

@thdxr More RPC?

[Aditya Kajla @akajla09]

@nedoleary the market is massive!

[Aditya Kajla @akajla09]

So is Disclosure Day a Wemby biopic?

[Aditya Kajla @akajla09]

Enterprise sales is super interesting.

Each Gartner and Forrester category typically has one, or a handful of analysts that decide the "leaders" in a category.

Once enterprise companies reach a certain size, a lot of money and resources go into "winning" over these analysts every year because so many Fortune 500 purchasing decisions happen based on what's in these reports, and not necessarily what end users or employees think.

[Dr. Julie Gurner @drgurner]

Action reduces fear.
Hesitations amplify it.

[Karan @KaranKajla]

When we designed the @radialhq API, we made a deliberate choice: favor RPC-style operations over REST.

REST was built with resources in mind. It maps cleanly to database tables and CRUD operations, which is great when a developer is browsing documentation or wiring up a frontend. But when an AI agent is the client, that design gets expensive fast. The agent has to translate what it wants to do into which resources to create, update, or delete. Extra reasoning, extra tokens, more room for error.

LLMs already think in actions: tool use and function calling ("do this thing with these parameters"). An RPC-style API lets the model speak that language directly. POST /orders.issueRefund instead of figuring out the right REST choreography across three endpoints.

MCP feels like a symptom of the same problem. A lot of what people are doing with MCP servers is wrapping RESTful APIs in a tool-oriented interface so LLMs can actually use them. That's useful, but it's also an extra layer. If your API is already action-oriented, you're closer to what the model needs in the first place.

More and more, AI won't just read your docs. It will call your APIs directly. That's going to push developers to rethink how they design systems, not just how they document them.

Curious to hear what others think about this.

[Aditya Kajla @akajla09]

if you're building AI agents, you need good guardrails.

one of the first things users try to do with @radialhq is "break the bot" by going off-topic.

if you don't have guardrails to prevent/recover from this, you can't build trust.

no trust = no customers.

[Aditya Kajla @akajla09]

@theCTO Working on this!

[Aditya Kajla @akajla09]

@samwillis Interesting 🤔 this is how we’re building our agents @radialhq

[Aditya Kajla @akajla09]

@techgirl1908 Not yet if one of the biggest agents (Claude) hasn't fully embraced it

[Aditya Kajla @akajla09]

Lots more exit liquidity

[Sawyer Merritt @SawyerMerritt]

Fidelity has announced that it is making the SpaceX IPO available to any customer with a retail brokerage account with $2,000 or more in the account (down from up to $500k before).

"SpaceX has decided to reserve a much higher percentage of the offering (up to 30%), which means there should be more shares available to retail clients, which is why we have decided to reduce IPO eligibility for this offering."

[Aditya Kajla @akajla09]

@petergyang Add @AGENTS.md in an otherwise empty CLAUDE.md file. That’s all you need

[Aditya Kajla @akajla09]

Finance bro summer in full swing. Morgan Stanley and Goldman bros watching Brunson take over game 1 with Anthropic and SpaceX IPOs on deck 🚀

[Aditya Kajla @akajla09]

@0xZoZoZo Dental offices are actually pretty tech forward. AI is good at admin/comms and even collections/talking to insurance. Offices love any efficiency gains.

[Aditya Kajla @akajla09]

one of the biggest benefits of being on the east coast is that @cursor_ai bugbot is much faster in the morning before SF wakes up

[Aditya Kajla @akajla09]

@lizwessel Sounds like these companies need new IT departments

[Aditya Kajla @akajla09]

@garrytan @jedwhite But labs are increasing per token cost

[Aditya Kajla @akajla09]

If you’re not building general purpose coding agents, I don’t get the need for sandboxes.

Most domain-specific AI agents should do fine with purpose built tools, APIs, or a stripped down web browser.

[Jonathon Belotti @jonobelotti_IO]

There's about 80 products in the agent sandboxing space right now. By YC Summer '26 we could hit 100

[Aditya Kajla @akajla09]

One thing I learned early on in biz is that you can just choose not to answer a question if you don’t want/need to.

As a founder, you will (and should) have conversations daily with other founders, customers, prospects, investors etc. This includes video chats, in-person, email, slack, whatever.

Some people will be much more savvy than you and focus on extracting information for whatever reason (competition, investment).

Most people (even inexperienced) can pick up on when this happens to them. When you do, simply ignore/deflect and move on. Never lie, but you don’t owe anyone an answer.