alphalist.CTO Podcast - For CTOs and Technical Leaders

Pat Casey was the first person besides founder Fred Luddy to write code at ServiceNow back in 2005, when it was called Glide and lived above a friend's restaurant. Twenty years later, he's CTO of a company where 85% of the Fortune 500 are customers, and until recently ran all of engineering: 10,000 people, 7,000 of them writing code. Almost nobody survives the journey from first engineer to public-company CTO. Pat did.

Tobi and Pat dig into how ServiceNow actually works under the hood: a metadata processing engine running 90,000 single-tenant databases and over 25 billion queries an hour, why they bought a 15-person German database company and turned it into RaptorDB, and why tearing apart a 20-year-old monolith is harder than every senior engineer thinks.

Then the conversation turns to AI. Pat bought 7,000 Windsurf licenses and measured a real, but unglamorous, 15% productivity bump, with a small subset of engineers going 5–6x while most barely changed. His thesis: AI coding is like playing five chessboards at once, and it's reshuffling the deck on who the top engineers will be. On agents, ServiceNow's answer is disarmingly simple: create a user called "AI Pat," assign it cases, and make it follow the exact same rules as humans because you should not trust an LLM more than you trust a human being.

Topics covered:

- From Atari 400 and floppy-disk jockey at Aldus to first engineer at ServiceNow
- Scaling engineering from a stuffed fish on a monitor to 10,000 people — and the productivity trough at ~100 engineers
- Single-tenant architecture: 90,000 databases, 25B+ queries/hour, and the monolith-to-Kubernetes migration
- Why ServiceNow bought Swarm64 and built RaptorDB on a Postgres fork
- 7,000 Windsurf licenses, Claude Code, and the real numbers on AI coding productivity
- "AI Pat": the anthropomorphic model for enterprise agents outcomes, not toolkits
- Whether AI kills seat-based SaaS, and why incumbents may have the inside track
- Pat's advice to CTOs: this is not a time for excessive caution

Before he founded Render, Anurag Goel was the fifth engineer at Stripe, where he watched roughly a fifth of the engineering team disappear into managing AWS, writing brittle, repetitive, error-prone infrastructure scripts that had nothing to do with the actual product. That experience became the seed for Render: a platform that automates away the undifferentiated DevOps work and lets application teams ship without standing up their own cloud team. Today, millions of developers build on it, and Render has raised over $260M from Bessemer and General Catalyst.

In this episode, Tobi and Anurag get into what's actually changing as AI moves from hype to production. Anurag makes the case that agents are simply a new kind of application, long-running, stateful, tool-heavy, and a new kind of end user you have to design for. He explains why Render deliberately refuses the "AI cloud" label, what he's building with Workflows and sandboxes, and why the hardest part of shipping agents isn't building them but seeing inside them.

The conversation also goes wide: how to hire executives when interviews lie, why short-lived keys and blast-radius thinking matter more than container escapes, how distribution is shifting from SEO to getting ChatGPT and Claude to recommend you, and why, despite all the "SaaS is dead" noise, specialization isn't going anywhere.

Topics covered:

Why ~20% of Stripe's engineers were stuck managing AWS and how that became Render
"We're not the AI cloud, we're the application cloud," and why the distinction matters
Agents, as a new type of application (and a new end user), you have to build for
Render Workflows and sandboxes: the consolidated AI runtime
Hiring executives when interviews are an imperfect signal
Security as blast-radius management: short-lived keys over "admin forever"
The shift from SEO to GEO, getting chatbots to recommend your product
Why SaaS isn't dying, and specialization still wins

Max Deichmann is the co-founder of Langfuse, the open-source LLM engineering platform that became the observability layer of choice for teams building production AI agents, before being acquired by ClickHouse. He started as a business student who taught himself to code via CS50 on a beach in Singapore, pivoted through Y Combinator, fired his own customers mid-batch, and built Langfuse out of a Sunday night conversation about what they'd actually want to build if nothing was in the way.

In this episode, Tobi and Max dig into what it really means to build and operate AI agents in production, not the LinkedIn version, but the 3 am alert, copy-pasted into Codex version. They cover the full loop: from pre-production experimentation and prompt iteration, to tracing, online evaluation, and the emerging architecture of agentic incident response. Max is unusually honest about where Langfuse itself still falls short, and what the next 12 months of the engineer's job actually look like.

What CTOs will take away: a clear mental model for LLM observability vs. traditional observability, a practical blueprint for agentic on-call workflows, and a grounded view of where agents are genuinely working in production today, and where the hype still outpaces reality.

Topics covered:

Why traditional observability tools fail for non-deterministic AI applications

The Langfuse loop: pre-production testing, tracing, online evaluation, and iteration

How the ClickHouse acquisition happened, and the half-page doc that decided it

Open source as a go-to-market strategy: adoption without a sales team

Agentic on-call: how Max's team handles 3 am incidents with Codex today

The "decision inbox" model, what the engineer's job looks like when agents do the work

Where agents are genuinely succeeding in production (and where LinkedIn is lying to you)

Alex Nahas is 28 years old and has already initiated a W3C web standard. Working as a backend engineer at Amazon, he ran into a problem most enterprises face: MCP requires OAuth, but most enterprise infrastructure runs on SAML. His solution was elegant: run the MCP server in client-side JavaScript, letting AI agents use the browser's existing authentication context rather than rebuilding auth from scratch.

What started as an internal tool became an open source project, then a viral Hacker News post published while under anesthesia, and ultimately an invitation from Google and Microsoft to help shape WebMCP as an official web standard.

In this episode, Alex and Tobi explore what WebMCP actually is, why the browser is the most underestimated sandbox in AI development, and what the agentic web might look like two years from now.

Topics covered:

What MCP actually is and why it's just an RPC framework at its core
Why OAuth is a dealbreaker for most enterprise infrastructure
How WebMCP lets AI agents operate within existing browser authentication
The Hacker News post that started it all, and why Alex doesn't remember posting it
How Chrome is natively building WebMCP support
The chicken-and-egg problem of standard adoption
Real-time bidding for agents and what it means for digital advertising
Why agents don't need their own identity
Where the agentic web is headed in the next two years

Most people assume the web runs on Google. The reality is more concentrated: only three companies on earth operate truly independent search indices — Google, Bing, and Brave. Jean-Paul Schmetz helped build one of them.

In this episode, Jean-Paul traces the arc from writing appointment software in a Belgian Radio Shack in 1981, through founding and selling Clix — a European search engine backed by Burda — to his current role as Chief of Ads at Brave, where he now sells search infrastructure to the AI companies that need it most.

For CTOs, this is a rare look inside an infrastructure layer most take for granted: how search indices are actually built, why it takes decades and hundreds of millions to do it properly, and why the entire AI grounding market quietly runs on infrastructure a small group of engineers spent their careers building.
Topics covered:

- Why only Google, Bing, and Brave have truly independent global search indices
- How AI companies use search grounding — and what happens when Google and Bing cut them off
- The SERP API gray market and why it probably has a two-year shelf life
- What it actually costs to crawl and index the web at scale
- The advertising model that will eventually come to AI — and why it's inevitable
- Jean-Paul's Stanford years: machine learning with Andrew Ng, and what was obvious in 2013 that took until 2022 to matter
- Build vs. buy for search infrastructure in 2025