AI agents bring new demands to the infrastructure connecting them to models, tools, and data. Each agent can have its own permissions and state, and spend much of its time asleep before waking up to make another request. Supporting that behavior at scale changes how the network routes traffic, loads policies, and manages resources.

In this episode of Alexa’s Input (AI), I sit down with Yan Avlasov, Staff Software Engineer at Google and a senior maintainer of Envoy, to talk about building the networking stack for agents.

We get into why Google is building on Envoy, what changes between serving inference and supporting agents, and why enforcing policies for individual agents requires a more dynamic control plane. Yan also shares what he’s seeing from AI-powered security scanning, including how models combine subtle bugs into serious vulnerabilities and why fixing them can mean changing behavior that production systems have relied on for years.

From the episode:

- Model selection, cost controls, and fallback across providers
- Loading policies dynamically as agent traffic changes
- Managing personalized state and agents that suspend and resume
- Where Model Context Protocol (MCP) fits alongside the protocols agents already use
- Semantic routing and choosing models based on what a request needs
- Security findings involving policy enforcement, path normalization, and JSON parsing
- The engineering work and cost of making security scanning continuous

Chapters
00:00 Introduction
01:12 Welcome, Yan
02:30 How AI is changing Google
07:56 Yan’s networking background
10:00 New networking layers for agents
13:11 AI gateways, cost controls, and fallback
16:38 Why Google builds on Envoy
22:21 Per-agent policies and a dynamic control plane
26:56 Inference versus stateful agents
31:58 Project Substrate: running agents at scale
33:31 MCP and the other protocols agents use
37:32 Semantic routing and model costs
40:40 AI security scanning and the work of fixing bugs
43:55 Policy timing, path normalization, and JSON
47:11 Making security scanning continuous
50:40 What’s next for agentic networking
53:32 Closing

Read the episode article on Substack

Watch on YouTube

In this episode of Alexa's Input (AI), I sit down with Alex Zenla, founder and CTO of Edera.

Alex grew up in a small town in Alabama, found a computer young, and started building. Her story is unlike many in tech. She taught herself to program and got a job in tech at 14 years old. Since then, she's been actively building and involved in open source. She's currently the founder and CTO of Edera, a company whose product integrates security into the lowest layers of the platform without sacrificing performance or velocity.

In this episode, we get into where that path started, what it costs to be different in founder and venture rooms, and what breaks when infrastructure still ships with security off by default.

From the episode:

• Growing up in small-town Alabama without a path into tech
• Southern niceness as theory versus practice
• Full-time work at fourteen and presenting to executives as a teenager
• Being one of very few trans founders in venture rooms, and the tension between visibility and being treated as a token
• Elevator pitches that change with the audience
• Detection and response after a problem has already occurred
• Common Vulnerabilities and Exposures becoming untenable when tools like Mythos surface hundreds of findings per project per day
• Kubernetes and vendors selling yet another layer while the foundations underneath are misaligned
• Secure defaults as the path of least resistance for teams that just need a cluster that works

Alex's mission is to make secure computing the default. Today you work hard to get a secure environment, and she's building Edera to invert that. What stays with you is how personal that work is for her. The path from a small Alabama town into those rooms is not separate from the product. It's why the default being broken bothers her enough to build a company around fixing it.

GENERAL PODCAST LINKS
Watch: https://www.youtube.com/@alexasinput
Read: https://alexasinput.substack.com/
Listen: https://creators.spotify.com/pod/profile/alexagriffith/
More: https://linktr.ee/alexagriffith

LEARN MORE ABOUT THE HOST
Website: https://alexagriffith.com/
LinkedIn: https://www.linkedin.com/in/alexa-griffith/

FIND OUT MORE ABOUT THE GUEST
LinkedIn: https://www.linkedin.com/in/azenla/
Bluesky: https://bsky.app/profile/alex.zenla.io
Edera: https://edera.dev/
GitHub: https://github.com/edera-dev

RESOURCES
Edera docs: https://docs.edera.dev/

AI agents have created more possibilities for engineers than ever before. But what does daily life actually look like for the builders who've gone all in?

In my very first panel episode, I sit down with Adam Anzuoni from Cursor, Taylor Dolezal from Dosu, and Peter Bell from Gather.dev. Three builders running agents every day for real work, not demos. Peter runs nine API plans across three Mac Minis, with built-in adversarial review. Adam manages cloud agents from his phone. Taylor is building the context infrastructure that makes agent knowledge portable across teams.

We get into deterministic pipelines, skill systems that become their own technical debt, and what all three kept coming back to: attention is now the bottleneck. When agents can do everything, deciding what deserves your focus is the actual hard problem.

Three setups. One shared constraint. A conversation worth hearing.

Topics discussed:

• Attention as the real bottleneck when agents can do everything
• Deterministic pipelines vs. agentic orchestration — when to use scripts and when to use agents
• Peter's system: nine API plans, three Mac Minis, adversarial review, self-improving context
• Adam on Cursor cloud agents and managing builds from his phone
• Taylor on context distribution — making accumulated knowledge available to ephemeral agents
• The skill maintenance problem and why agent systems become their own technical debt
• The ADHD-like productivity loop that agent-driven work creates
• Plans matter more than prompts — all three panelists converged on this
• Intermediate artifacts as the key to quality output
• Product mindset as the engineer's next high-value skill
• Sandboxes, governance, and why "approve, approve, approve" puts your hard drive at risk

General podcast links

Watch: https://www.youtube.com/@alexasinput
Read: https://alexasinput.substack.com/
Listen: https://creators.spotify.com/pod/profile/alexagriffith/
More: https://linktr.ee/alexagriffith

Learn more about the host

Website: https://alexagriffith.com/
LinkedIn: https://www.linkedin.com/in/alexa-griffith/
X: https://x.com/alexa_griffith_

Find out more about the guests

Adam Anzuoni
LinkedIn: https://www.linkedin.com/in/adamanz/
Website: https://www.adamanzuoni.com/
Cursor: https://www.cursor.com/

Taylor Dolezal
LinkedIn: https://www.linkedin.com/in/onlydole/
Website: https://onlydole.dev/
Dosu: https://dosu.dev/

Peter Bell
LinkedIn: https://www.linkedin.com/in/peterfbell/
Gather.dev: https://gather.dev/
O'Reilly Book: Scaling AI Adoption in Engineering

Resources mentioned in this episode

Cursor: https://www.cursor.com/
Dosu: https://dosu.dev/
Gather.dev: https://gather.dev/
Anthropic Claude: https://www.anthropic.com/
The Phoenix Project (book reference by Taylor)
Kelsey Hightower productivity survey (referenced by Taylor)