Gene Linetsky

I sat down with Frontegg CEO Sagi on scaling CIAM for Insight Health. A few highlights from our convo, • Day-one enterprise readiness (HIPAA, multi-org, customizable access policies) • Why “build your own auth” slows growth (SSO/RBAC/SCIM, audit logs) • Delegated admin so partners self-manage without pinging engineering • CIAM as a GTM enabler for AI agents and clinician workflows Replay + transcript: https://lnkd.in/grRed-C5

47

1 Comment

This week, we shipped a predictive route detector and stabilized Tycho execution flows, inviting a few partners for short onboarding calls. This week, our team shipped a focused, production-oriented step toward private, safe MEV routing. We integrated an FMD-based profitable-route detector into the execution stack, refined Tycho route encoding and Permit2 flows, and stabilized our canonical two-hop example, enabling contributors to reproduce realistic simulations locally. These changes aren’t just code. They cut down wasted simulations, surface higher-quality candidate routes, and make private execution paths more deterministic and auditable. Quick technical highlights: - FMD detector: physics-inspired prediction (TPS/committor) ranks route candidates before expensive REVM validation. - Safe execution: Tycho-encoded solutions + Permit2 user-transfer flow — never broadcast to the public mempool. - Developer flow: fixed example stability and clearer env checks, so onboard cycles are shorter. Why it matters: Extracting value from fragmented DEX liquidity is both an engineering and research problem. Predictive filtering reduces noise, and deterministic encoders reduce execution risk; that’s what moves this from lab experiments to usable tooling for desks, wallets, and integrators. We’re inviting a small set of partners for a brief 20 to 30-minute onboarding call to walk through our approach, discuss integration pathways, and gather feedback on our docs and APIs. If you’re building MEV-safe routing, wallet integrations, or execution infrastructure and would like a private walkthrough, message me or email joshua.davidson@orkidlabs.com, and we’ll schedule a quick call. If you're interested, please preview our new whitepaper below. DM for a copy if you surf chaos like Paul Atriedes rides the sandworm. Negentropy I'll send you the version properly processed with LaTeX. I promise. 🙏 — JC, ORKID Labs p.s. Beta Version of ORKID's Uniswap V4 hook deployed: https://lnkd.in/gTn46y6Q #MEV #MEVProtection #DeFi #Web3Infrastructure #Rust

4

The cofounder and CTO of Perplexity, @denisyarats, just shared internally that Perplexity is moving away from MCP-based integrations in favor of direct APIs and CLIs. That might sound like a minor implementation detail, but it actually says a lot about where agentic AI is heading. MCP (Model Context Protocol) is a powerful standard for letting LLMs discover and call tools in a uniform way, and it’s been embraced by many players in the ecosystem. It helps agents dynamically find tools, maintain context, and orchestrate complex workflows without bespoke glue code for every integration. But there are tradeoffs. Many teams are realizing that: - Mature REST APIs already provide stable, well-understood contracts for high-stakes workflows and data access. - CLIs give agents a compact, text-native, stream-oriented interface that’s easier to parse, cheaper in tokens, and inherently composable via pipes. - Existing auth, logging, and DevOps tooling are already built around APIs and CLIs, not new protocols. In that context, Perplexity’s move is interesting: instead of betting everything on a “universal” protocol layer, they’re doubling down on the primitives that have scaled the web and cloud for decades—HTTP APIs and command-line interfaces. My takeaway: the winning pattern for AI agents might not be a single grand protocol, but a thin intelligent layer on top of the interfaces engineers already trust—combined with careful choices about where standards like MCP make sense, and where direct API/CLI integration wins on speed, cost, and observability. Curious how other teams are thinking about this balance between MCP, APIs, and CLIs right now.

11

3 Comments

Cyber insurance capacity is not constrained by premium volume. It is constrained by correlation. Across portfolios, insureds depend on the same cloud providers, identity platforms, and core software stacks. When one of those shared dependencies fails or is compromised, losses are not independent. They accumulate. That is the structural challenge carriers and reinsurers manage. Underwriting has responded by becoming more evidence-driven. Controls such as MFA, segregation of duties, and privileged access governance are increasingly corroborated rather than accepted at face value. The next step is continuous verification. Controls degrade between renewal cycles. When that happens, underwriting assumptions drift and severity modeling becomes less reliable. This week, Tristan Morris examines how correlation defines the practical boundary of insurability — and why verifiable control design contributes to underwriting discipline: 👉 The Math of Resilience: How Verifiable Controls Support a More Stable Cyber Insurance Market https://lnkd.in/gWgUdYKG Follow SplitSecure for perspectives on cybersecurity, privileged access, third-party risk, and operational resilience under #DORA, #MiCA, and #NYDFS — along with weekly breach briefs examining how eliminating stored secrets and unilateral administrative authority reduces irreversible loss exposure.

26

1 Comment

Andrej Karpathy on AI-assisted coding: "I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in December. [...] It hurts the ego a bit" Full post with details and predictions on X: https://lnkd.in/dxP24SDf Andrej was recently on record (on Dwarkesh podcast) saying AI coding agents were not good enough for non-boilerplate coding tasks, which makes the new post so surprising (and yet not, to people watching closely what's been happening with the coding agents). Anecdotally, I've been hearing the same from *many* senior software developers who know their stuff (ie. aren't just blinded by hype). The development workflows are being structurally redefined very quickly. In my recent talk on AI-assisted software engineering I tongue-in-cheek joked that the information has a "best before" date of a couple of months. Seems right on spot ... I'll need to update the talk (again).

22

2 Comments