Everyone has agents now.Few have 33 years of software development to aim them with.
I'm Jon Richards. I build software with AI coding agents, across databases, infrastructure automation and multicloud platforms, and I help teams adapt how they work. A lot of people find this moment unsettling. I've never had more fun.
I help teams move from occasional AI use to a dependable practice:
shared AGENTS.md standards;
plan-to-ticket workflows;
parallel agents in isolated worktrees;
cross-model review and CI gates;
agent-safe credential handling;
a shared team agent with written delegation rules;
hands-on training.
I've run all of it in production, not just in demos.
Software development
I build products and platforms with that same practice:
multicloud services and data pipelines on AWS, GCP, Azure and OCI;
databases and migrations;
release pipelines;
web and mobile apps.
Selected Work
Case studies
A shared team agent: about 1,000 requests in six weeks, 65% from five teammates
A coding agent in a client engineering team's Slack. Engineers ask it for PR reviews, root-cause analysis, fixes and builds. It works under written delegation rules:
allowlisted requesters;
a pre-authorized small-fix policy with six scope limits;
requester attribution on every PR;
a CI re-gate before any merge.
One morning it took a release-blocking upgrade defect from report, through reproduction, to a fix with a regression test, merged after human approval.
I built a personal work-intelligence system run by Claude Code.
Sources: it pulls from Gmail, Google Calendar, Google Drive meeting transcripts, Slack channels and DMs, Jira and Bitbucket.
Outputs: structured notes, an action and decision tracker, and a daily briefing in Slack.
The workboard: its personal app is an always-on local web app where I reprioritize work. My edits flow back into the next briefing, and pull requests and tickets close themselves when their system of record says they're done.
Ad-hoc queries: between briefings I just ask it things: “What is my client asking of me today?” or “Which of my PRs cover the Node 24 upgrade? Add these two reviewers to both.” It answers from everything it has gathered and makes the change.
Designs become dependency-safe tickets, are audited with fresh eyes, then are worked one agent at a time. An agent escalates to a person whenever it's blocked rather than guessing. (319 closed, 32 escalated)
Three agents, one app: pbmatch
An iOS, Android and web app built by three coordinated agents in separate worktrees, sharing one ticket queue. UI changes are verified across iOS, Android and web at normal and large text sizes, with dated evidence records and automated UI flows.
Thirty-three years in, and having the most fun yet
I started in database kernels at Informix, writing a threading library in assembly and building storage and logging subsystems. At Oracle I spent 26 years automating infrastructure:
VM image assembly (two patents);
bare-metal provisioning for edge devices;
rack-scale backup and restore;
finally, the automation that builds new cloud regions.
In 2025 I went all in on agentic engineering and started GorgeGuy Solutions. Since early 2026 I've consulted for a multicloud SaaS company, building core services for a multicloud product across AWS, GCP and Azure. I also introduced the team's shared agent and agent practices.
Technology has always moved fast. For my whole career, keeping up felt like drinking from a fire hose: you chose the topics that mattered and kept learning. This year is different. Agents have changed the practice itself, not just the tools: how we plan, build, review and ship. Nobody has the playbook yet. Many people in our profession find that frightening. For me it's the most fun I've had in 33 years. We're at the very beginning, and the people doing the work get to decide what good practice looks like. That's where I want to be: trying things, keeping what works, and sharing it.
What I've learned: agents remove the cost of writing code in an unfamiliar language. They don't remove the need to know the domain well enough to say what to build, how to break it into pieces an agent can finish and verify, and whether the result is right. Thirty-three years of systems work is the part I bring.
