A Field CTO Perspective

Inside the AI Harness: The Substrate Around the Chat

Tue, 13 Oct 2026 00:00:00 +0000

Written by a practitioner who has spent three years building AI apps, harnesses and code development tools. See About the author at the end.

Why the harness matters

In 2026, what an AI agent can actually do depends less on the model and more on the software wrapped around it. That wrapper is the harness.

A language model on its own is a function: text in, text out. It cannot read a file, run a test, remember yesterday, or check its own work. Everything that lets it do those things lives outside the weights.

The Rules Nobody Wrote Down

Tue, 29 Sep 2026 00:00:00 +0000

For twenty years, good software depended on people remembering which parts of a program were allowed to talk to which. AI coding tools don’t remember. So the rule has to be written down where a machine can enforce it.

A growing share of new code is written by AI. One peer-reviewed study estimates that by early 2025 about 29 percent of new Python functions pushed to GitHub from the United States were AI-written, the highest rate the study measured, up from about 5 percent in 2022.

The deterministic half of neuro-symbolic AI already exists. It's called network verification.

Sun, 26 Jul 2026 00:00:00 +0000

Everybody’s suddenly got a take on neuro-symbolic AI. Inference for pattern recognition, symbolic logic paired in for reasoning you can actually trust, and the EU AI Act’s enforcement phase lighting a fire under all of it because “the model said so” stopped being an acceptable audit trail.

The pitch is real. Inference is powerful and unreliable, a statistical guess dressed up as an answer. Deterministic, mathematical models are reliable and narrow. Combining them, having the deterministic side check inference’s work, is turning into the actual path to AI you can trust in production. What’s less talked about: in networking, the deterministic half of that equation isn’t something you’d have to build. It already exists, and it’s been hardening in production for 7 years.

Delivering APIs for AI

Thu, 09 Jul 2026 00:00:00 +0000

The rise of AI agents and autonomous systems is fundamentally changing how we design and deliver APIs, shifting from human-centric interfaces to AI-native integration patterns that require new architectural thinking.

The Inflection Point

Technology innovation has repeatedly transformed how we build systems. From the fourth-generation programming languages (4GL) of the 1990s that abstracted database operations to today’s AI-powered code generation, each wave has forced us to rethink fundamental assumptions. We now stand at another such inflection point: the transition from human-interface (HI) APIs to agent-interface (AI) APIs.

Breaking the Capability Trap

Fri, 26 Jun 2026 00:00:00 +0000

A Systems Approach to Safe, Fast Network Change

A whitepaper on applying System Dynamics to network change management, and the business case for pre-change simulation

Executive Summary

Network change is one of the riskiest things an enterprise does to itself. A single routing or firewall edit can quietly break an application dependency, pull reachability from a whole region, or take the business offline for hours.

The usual response (freezes, heavier approval boards, rarer maintenance windows) feels like risk reduction but isn’t. Research from MIT Sloan’s System Dynamics group shows it does the opposite. Each control added after a failure quietly raises the odds of the next one, and the organization spends years tightening the loop on itself.

A brief look at Network Automation

Tue, 06 Jun 2017 00:00:00 +0000

There is an incredible amount of hype these days discussing automation especially in the realm of networking. Some would say that this is relatively new but I would argue that we have been struggling with automating the network for many decades, it is however becoming critical today due to the massive dependencies on our communications facilities and shifting application architectures towards distributed computing.

As defined by Oxford the definition of automation:

Monitoring the Control Plane

Sun, 28 May 2017 00:00:00 +0000

Primer

Within every customer, provider and transport network, there exists a host of applications that collect, disseminate and make decisions so that application services can communicate. These applications are also known as routing protocols and are at the core of how the Internet functions. The most common routing protocols used today include: Border Gateway Protocol (BGP) ₁, Open Shortest Path First (OSPF)₂ and IS-IS (Intermediate System - Intermediate System) ³.

Facebook Fabric Networking Deconstructed

Mon, 28 Jan 2013 00:00:00 +0000

Facebook engineering has recently introduced their next-generation data center network architecture planned to be operational in the new Altoona facility. Introducing data center fabric, the next-generation Facebook data center network

In this post we are going to look at the design proposed and break down some of the reasons why this design was necessary although maybe still not ideal.

Facebook, the killer app.. literally

Within Facebook’s data center the network is a critical component driving this social networks ever increasing appetite to connect people together. The challenges unfold on two axis, the first being the explosive growth in the user population and its increasing demand and access bolstered by modern mobile devices and the second related to the exponential growth of machine to machine traffic required to aggregate and compose information which we all know and love. Due to the nature of Facebook applications, machine to machine communications is “several orders of magnitude larger than what goes out the Internet”

About

Mon, 01 Jan 0001 00:00:00 +0000

Gary Berger is a Field CTO and principal engineer. He writes about network automation, control-plane observability with eBPF, and data center fabric design.

Developer, engineer, father, musician.

AI Research Index

Mon, 01 Jan 0001 00:00:00 +0000

Open-source AI projects I’m tracking, grouped by area: local inference on owned hardware, multi-agent orchestration, coding agents, memory and reasoning, agent tooling, and agent sandboxes.

Experiments

Mon, 01 Jan 0001 00:00:00 +0000

Smaller things I’ve tried in the open: prototypes, workshop repos and ideas that may or may not grow up. The finished work is on the Projects page.

Projects

Mon, 01 Jan 0001 00:00:00 +0000

Some of the things I’ve built and put in the open, mostly tools for working with AI agents, APIs and networks. Everything else is on GitHub.