AI coaching for teams
AI tooling for teams that ship real software.
Live coaching for engineering leads, small development teams, and the IT people who keep a business running. On your screen, in your repository, from a CTO who runs an AI-native shop on systems that have been in production since 1997.
Your team tried the tools. The first pull request was nine hundred lines nobody wanted to read.
The tools were not the problem. The missing rules were.
Write the rules once, put them in the repository, and the speed is real.
Judgment directs. AI accelerates.
Who this is for
You have engineers. They have not shipped with AI yet.
Or they have, and it went badly: a pull request nobody could review, tests that passed for the wrong reason, a connection string in a prompt. Now the team is split between people who think the tools are magic and people who think they are a liability. Both are half right, and the half that matters is the one you write down.
This is also for the person who keeps a small company or an agency running: the one who was told to “do something with AI” and would rather it be the right thing. Integrations, automations, the spreadsheet that quietly became a system. I have built those for twenty-five years, and I build them with these tools now.
Every session is live, on your screen, in your repository and your accounts. I do not send slides. You leave with something running and a rule written down for the next one.
Handling public money, resident data, or anything regulated? Read the rules first — they are the part that makes the speed safe.
Run a small business rather than a team of engineers? Ellen’s coaching is built for you, and she sends people to me when the problem turns out to be an integration.
The background
Twenty-five years of making teams faster without losing control
At Parametric Portfolio Associates, now part of Morgan Stanley, I grew an engineering department from six people to more than thirty and ran a DevOps group that released software 450 times a year. When I started, two releases in three failed. When I left, fewer than one in a hundred did. The difference was never a tool. It was rules the whole team could see, and a review step that lived in the workflow instead of in good intentions.
Before that I built the laboratory system that made an environmental testing lab one of eleven in the country certified by the EPA. Since 2019 I have been Meserow’s chief technology officer, and the firm now runs AI-native: Claude Code and Copilot in every engineer’s working day, on systems that have been in production since 1997, under written rules for what the tools may touch and what they may not.
Every one of those transitions had the same shape. Someone senior said the new way was reckless, someone junior said the old way was slow, and the team that came out ahead was the one that wrote down how to do the new thing safely and then did it every day. AI is that transition again, with one difference: the gap between a team with the habits and a team without them is a multiplier, and it compounds every month.
I still read every plan before I type yes. That is not caution for its own sake. It is the habit that lets the rest of the day move fast.
The kind of work
Three problems that have already been on my desk
Names you may recognise, and the shape of the problem behind each one. The shape is what you will bring me.
KING COUNTY
A production system, rebuilt three times
PRISMnx is the system of record for the county’s capital program, with more than $20 billion in active projects. We have rebuilt it three times as the platforms underneath it changed, and the AI tooling is the fourth. Our engineers use it there every day, under the rules on this page. The story is at meserow.com.
PARAMETRIX
Reading machine-generated code and deciding what to keep
As a subconsultant to Parametrix for King County’s wastewater division, we reviewed a vendor’s machine translation of an aging desktop system into C#, estimated what finishing it would really take, and specified the servers and the QA environment. On a second work order, the SharePoint architecture and Power Automate workflows behind a project management manual. Judging generated code is a job I did before the generator was an LLM.
UNIVERSITY OF WASHINGTON
A reporting database for a legislative study
The database behind Buy Clean Buy Fair, a pilot study for the Washington State Legislature run by the College of Built Environments’ Carbon Leadership Forum. Ellen and I are named co-authors on the published report. Every client, by sector.
Your coach
Patrick Escarcega
Chief technology officer at Meserow Design. I make the architecture and platform decisions on every system the firm builds, I set the rules for how our engineers use AI tooling and where they do not, and I still work in the code every day. In coaching I do the same thing on your screen: your repository, your accounts, your real backlog.
I coach the way I run my own team. You keep your hands on the keyboard, I explain what I am seeing and why, and nothing is applied until you have read the plan.
Working safely
Nothing from production goes into a prompt.
The first hour of The Half Day is spent on this, and it is the hour that pays for the rest. These are the rules Meserow’s own engineers work under. You leave with your team’s copy, in the repository, where a new hire reads it on day one.
01 / ACCOUNTS
Business tiers the company owns
Training on your inputs excluded by contract. Your admin, your logs, your off switch. Never a personal login, and never a tool I administer for you.
02 / DATA
Schemas and fixtures, never production
No customer records, no secrets, no connection strings in a prompt. The AI gets the shape of the data, not the data. Secret scanning on, and the ignore file honoured.
03 / THE REPOSITORY
Rules live in the repo, not in people’s heads
An instruction file the tool reads on every task: what it may run on its own, and what waits for a human. Deploys, migrations, deletes and commits wait.
04 / A HUMAN SIGNS
AI drafts. A named engineer reads every line.
Plan before apply. Tests before merge. The person who merges it is the person who answers for it at two in the morning, so that person reads it.
For public agencies. Meserow has built and operated systems for King County since 1997, including the software of record for a multi-billion-dollar capital program. For an agency, generated code and AI conversations may be public records, so we decide where they live before anyone types. We will map these rules to the AI guidance your oversight body already uses, and I will join a call with your IT, security, or public-records staff before you sign anything. Set that up at meserow.com.
Three ways to work together
Pick by the size of the problem.
All three are live, on your screen. The prices come from the same hourly rate Meserow bills engineering at; nothing is marked up for being called coaching. The seats are capped so I can keep doing the work I coach.
The Hour
$250
One problem, sixty minutes
“Is this tool safe for us?” “Our first AI pull requests were a mess; show me why.” “Read the rules I wrote.” Bring one thing and we work it.
- On your screen, in your accounts
- No preparation needed: a screen share and a question
- A short written note afterwards: what we decided and why
The Half Day
$1,000
Four hours. Four a month.
Your team’s AI tooling set up properly, the rules written into the repository, and one real change shipped through them before we hang up.
- You brief me first: the repo, the tools, what went wrong
- Accounts on business tiers; your team’s rules in the repo
- One real change: AI draft, human review, tests, merge
- A week of Slack or email afterwards while it beds in
The Quarter
$2,000/mo
Three months. Two teams at a time.
For a team lead who wants the habits to stick. We work your real backlog together until the team runs them without me.
- A two-hour working session every other week
- Slack or email between sessions
- Your backlog, your codebase, your standards
- Month three: you run it, I watch
The Half Day
You hang up with it shipped.
A setup that is “mostly done” is not done. The team goes back to the old way on Thursday, and the half-finished rules become the reason nobody trusts the tools. So the session does not end at the setup. It ends when a real change has gone through it.
01 / BEFORE
You send, I read
The repository layout, the tools you are using, the three most recent pull requests, and what went wrong the last time. I read all of it beforehand, so no live minute goes to “tell me about your stack.”
02 / THE SESSION
Four hours, building
The first hour is accounts and rules: business tiers, the instruction file, what waits for a human. The other three are one real change from your backlog, end to end: AI draft, your review, tests, merge. Your hands, your keyboard, your repository.
03 / AFTER
A week on call
Slack or email for the week after, while the team lives with it. The second and third changes are where the questions come, and that is when a short answer saves a long detour.
When coaching is not enough
The Quarter is me coaching your lead. The Embed is one of us on your team.
Some teams need more than habits. They need a senior engineer in the codebase for a quarter, shipping to their standards and leaving the habits behind. That is the Embed, and it is where The Quarter leads when the backlog is bigger than the coaching. I will tell you which one you need in The Hour, and I will not sell you the bigger one when the smaller one will do.
One afternoon, and your team has shipped a real change with AI tooling, under rules it can read.
Start with an hour. Ask for a Half Day when you want it built with you.
Book an hourPrefer to talk first? Send a note and I’ll reply myself.