Demian Brecht

Keep the workflow, swap the coding harness: using MCP as a portable boundary around bounded, durable agent applications

Runtime QA is judgment from end to end, so let the agent own the investigative loop. Separate its judgment from reviewable mechanics, stop for the decisions that carry consequences, and leave the last word with a person.

The most important question in AI architecture isn’t which model you use. It’s who owns the control loop. That choice affects cost, reliability, and how easily you can swap providers. In this article, I look at three categories of workflows and explore why the nature of the problem should determine the harness, and why code, rather than the model, should often drive the loop.

LLMs are stochastic by design. Reliable agent workflows come from surrounding that chaos with deterministic code, not from wishing the model were predictable.

Reproducing Android-to-API security findings means juggling emulators, certificates, and proxy config you touch a few times a year. Here's how I automated the setup away.

A self-described tech curmudgeon on how generative AI rekindled the urge to tinker, explore, and build outside the comfortable set of tools built up over the years.

Two Salesforce engineers describe separating a Django API framework from a legacy application and navigating the process of releasing it as open source.

Originally published in Stack Overflow Blog.

Your feature shipped, passed review, and then someone flagged the latency. A walkthrough of profiling Python to find where the time actually goes before you start optimizing.

Answering the funny looks: why I still reach for vim over an IDE: its ubiquity across *nix systems, the power of command-line debuggers, and a deliberately minimal .vimrc that travels anywhere.

Separating critique of the code from critique of the person who wrote it — a realization about code review that reshaped how I think about my career.

Originally published in Salesforce Engineering.

Lessons from a first-time speaker on preparing, practicing, delivering, and learning from technical presentations.