Naren Sathiya

founder @TelivySecure (yc s21, acquired)
ai × cyber risk × systems
research, experiments & writing →

[Naren Sathiya @bui1dcathedrals]

AI can already generate beautiful interactive artifacts. The problem is sharing them.

This weekend we built Relay.

Ask Claude or Cursor to publish an HTML artifact, share it with someone, and track who viewed it.

Demo ↓

Try Relay: https://t.co/FNu3gnxfAO

[Naren Sathiya @bui1dcathedrals]

Building this at Tether Labs. We’ve been researching agent trust, evaluation, and supervision, informed by deploying AI into SMB cybersecurity at Telivy/Cytracom. Now exploring how intelligent systems learn who to trust and how evaluators themselves should evolve. Would love to chat.

[Naren Sathiya @bui1dcathedrals]

https://t.co/7GvLspzfUB

[Naren Sathiya @bui1dcathedrals]

I've been thinking a lot about what happens when AI agents start making more and more decisions on our behalf. Not just what they can do, but how they decide.

- Who do they trust?
- When do they switch to another service?
- How much do latency, failures, or reputation matter?

I wrote down some thoughts and why I think this is an interesting area to explore. I'm also starting work on an open simulator to study these questions. Link below ->

[Naren Sathiya @bui1dcathedrals]

@benln Hey hey

[Nikesh Arora @nikesharora]

Time to go FDE! So the new craze, let's build an FDE business because customers need it. What is an FDE -
1. well versed in AI? - yes
2. Well versed in applying their knowledge to make implementation successful for what I bought from you - yes, Palantir, Frontier Models, now Hyperscalers.
The real question is:

Is the FDE actually building unique product capability in the field that will then be integrated into the product eventually back at the ranch? Or is the FDE building unique capability for my enterprise?

The answer to the question helps determine whether the FDE provider accelerates product development and value capture.

For the customer - higher probability for driving outcomes for sure, but if you have ever worried about vendor reliance - this is the moat of a lifetime.

Use FDEs for sure, but understand the architecture, make sure you understand where the intelligence is being retained in your enterprise:
1. Model weights? Do you have a fine tuned model?
2. Generic model - but organizational memory, context layer and harness?
3. Unique application context and data being leveraged?

No prizes for guessing where the different FDE providers will build that value for you, hence creating the moat for themselves.

Go FDE but tread thoughtfully.

[Naren Sathiya @bui1dcathedrals]

@levie Love seeing more tooling around HTML artifacts.

I've been building Relay around the publishing/sharing workflow for AI-generated HTML. It feels like this is becoming a much more common pattern.

[Naren Sathiya @bui1dcathedrals]

@renkelvin21 @trq212 This is awesome. I've been using HTML a lot more for AI outputs lately too.

I ended up building Relay last week because I kept running into the next problem... sharing those HTML artifacts outside the editor.

I think these two fit together really well.

[Naren Sathiya @bui1dcathedrals]

I've been on the same path. HTML has become my default output too.

The only friction I've run into is sharing it. Once it's outside Claude/Cursor, it gets surprisingly clunky.

I ended up building Relay over the weekend to make that part easier. Curious if you've found a workflow that works well.

[Naren Sathiya @bui1dcathedrals]

@iAmHenryMascot @trq212 Completely agree. I've been generating way more HTML artifacts lately.

The next friction point for me was sharing them, especially across different AI tools. I ended up building Relay over the weekend to solve that.

[Naren Sathiya @bui1dcathedrals]

@trq212 access it here - https://t.co/VGt2uRtPve

[Naren Sathiya @bui1dcathedrals]

I've really been enjoying generating HTML artifacts lately, but sharing them is still surprisingly clunky. Claude's built-in sharing is great, but it doesn't really work once you're outside that ecosystem.

I spent last weekend building Relay to make that easier. Publish HTML artifacts from Claude, Cursor, or any coding agent as secure share links with permissions and analytics.

Curious if others have felt this pain too.

[Chandana 🌻✨ @RoseOnX9]

Innovation starts with noticing problems like this.

This is how you create lifelong customers.

[Naren Sathiya @bui1dcathedrals]

One idea from A Timeless Way of Building that’s reshaping how I think about product strategy:

Every design is a response to competing forces.

Context → Forces → Configuration

For Cursor, the forces might be:

Developers want to move faster.
Developers don’t want to lose control.

The configuration becomes inline completions, chat, diffs, and approval before applying changes.

The features aren’t the product. They’re the resolution of the forces.

[Naren Sathiya @bui1dcathedrals]

One product lesson I’ve been thinking about:

It’s often easier to sell outcomes than improvements to expertise.

Building accounting software for accountants is hard. You’re selling to experts who know every edge case, every workflow, every shortcut. Your product has to earn its place.

Selling accounting services to businesses is different.

The customer doesn’t care how journal entries work.

They care that they’re compliant, taxes are filed, reports are accurate, and audits go smoothly.

Same with cybersecurity.

Selling to security engineers means competing on detection quality, workflow efficiency, and technical depth.

Selling security to a business means selling peace of mind: lower risk, insurance eligibility, compliance, and business continuity.

One market buys better tools. The other buys better outcomes.

I think AI is pushing more software toward the second category.

[Naren Sathiya @bui1dcathedrals]

One product lesson that’s taken me a while to learn:

Expanding into a new product is much easier when you’re extending the same use case, not just selling to the same customer.

At Telivy, we started with cyber risk assessments and expanded into continuous monitoring.

Same customer.

Different job.

Risk assessments were about proving posture. Monitoring became an operational SecOps workflow owned by different people with different incentives.

Today we’re building multiple products for cyber insurance instead:
• application analysis
• policy analysis
• continuous compliance

They’re different products, but they’re all helping users accomplish the same underlying job.

The easiest product expansion isn’t customer expansion. It’s job expansion.

[Naren Sathiya @bui1dcathedrals]

@trq212 Built something similar for sharing AI-generated HTML outside Claude w/ MCP support.

Upload HTML or zip bundles → get a shareable link with permissions and analytics.

https://t.co/FNu3gnxfAO

[Naren Sathiya @bui1dcathedrals]

Interesting lesson from hacking on Relay (https://t.co/VjSHGDdoIQ):

Adding MCP support early changes product thinking.

For users, you design interfaces. For agents, you design capabilities.

The questions become:

What context is needed?
What actions are allowed?
What outcomes should be possible?
How do we supervise them?

Feels like we’re moving from designing experiences to designing environments.

Curious what others have learned building agent-native products.

[Paul Graham @paulg]

Prediction: In the AI age, taste will become even more important. When anyone can make anything, the big differentiator is what you choose to make.

https://t.co/3GQUlfH58t

[Naren Sathiya @bui1dcathedrals]

One of the most supportive folks you can have on your cap table!

[Restive Ventures @restivevc]

We're funding new companies building AI applications and infrastructure for financial services.

We write founders their first check, and we want to back at least 3 more cracked teams over the next month.

If you’re building, or thinking about building, we’re opening up public applications here. Apply now.

Not an accelerator. No cohort, no demo day, no program terms. We invest on market terms.

Apply by midnight PST July 20: https://t.co/F4biQZn48r