Still shipping
I am Shawn LoPresto. I build software, I write about building software, and I give away the tools I build along the way.
Two decades in DevOps and SRE, building the platforms and delivery pipelines that let other people ship, and now a turn toward the business side to put AI to work on real processes.
The throughline has not changed: make the reliable, automated path the easy one, and measure whether it actually helped. I am at my most useful between engineering and the business side of a decision, which usually looks less like strategy decks and more like finding the automation nobody had thought to ask for.
Where I have worked
Brought a travel business onto the cloud
AWS was still EC2, S3 and RDS, and releases went through ITIL change management, as they did everywhere then. I built the hosting platform for the marine business: PageSpeed as the CDN layer, HAProxy for load balancing, RDS MySQL, GlusterFS for global failover, and Jenkins running a fully automated CI pipeline into it, so every delivery was consistent.
It grew to host other TUI companies beyond marine, and I flew to England to train the UK teams on cloud and help them plan their migrations.
Standardised delivery, then scaled the SRE team
Already on AWS, and growing fast enough that delivery had taken shape team by team: releases went out from developers’ own machines, with different patterns even inside one team, and an enterprise product was about to launch. I standardised delivery on Jenkins, running on Elastic Beanstalk at first, then built Octopada, an in-house tool for managing infrastructure as code for every team’s apps on ECS. We used it to migrate licensing, research, the enterprise platform and every other system with no downtime, and went on to Jenkins as code and ChatOps, first in HipChat and later in Slack.
Not long after, we moved that infrastructure code to Terraform. It was young then, but already further along than anything we would build in-house and maintained by people whose whole job it was. Knowing when to retire your own tool for something like that is a skill of its own.
I also ran the OpenSearch services behind our flight-recorder system, drawing on performance-tuning habits from an earlier stint on Merrill Lynch’s Operational Stability SWAT team. The SRE team grew from 3 engineers to 25 as development and product scaled, at roughly one SRE for every eight engineers.
Putting AI to work on business processes
Back on the non-engineering side of the house, with the business processes in view. I work across Sales, HR, Legal, Security, Engineering, IT, QA and Customer Success to find the processes worth handing to AI, help each team define what they want it to do and what other automation it needs, and roll it out with them.
Then we measure: the ratio of agent work to human work, and the quality of what comes back, so the effort has to earn its keep. I have used Claude since it launched and OpenAI’s models before that. Helping a team see how efficiently it can operate is the best part of the job.
Work with me
I take on a small number of engagements in the same three areas I write about. They tend to start as one of these:
- An AWS or GCP estate that costs more than it should, or a migration nobody wants to own.
- A team shipping slower than it could: developer experience, patterns and architecture, reviewed with the code open.
- Getting Claude Code, the Agent SDK or another agent framework into a workflow that does real work — safely and measurably.
The fastest way to find out whether it fits is a short email describing the problem.
What is here
Blog is field notes — the things that cost something to learn — filed under Infrastructure, Development and AI. Marketplace is the Claude Code plugins I build for my own work and hand out for free. No newsletter; there is a feed.
New posts, no inbox required
No newsletter and no email address — just a feed.