← TechBrig

Notes from the work

Short, practical pieces on AI automation, where AI pays off, building with Claude, and turning prototypes into products. Written by the people who do the work.

Nothing sends itself: why the approval step is the most important part of an AI system

The approval step decides whether an AI system can be trusted with real work. What should wait for a person, how to stop approvals becoming a bottleneck, and when a step can earn its way out of review.

Read →

You don't start with all of it: how to roll out AI across several locations

A multi-location business does not need a big AI program. It needs one piece, working at one location, before anything else. How to pick the first site, what travels to the next one, and when to stop.

Read →

Private events and catering: how a restaurant answers every inquiry

Event and catering inquiries arrive while the people who can answer them are on the floor. How a restaurant catches every one, replies with the right room and package, and keeps a person in charge of what goes out.

Read →

AI for restaurant groups: where the hours go, and what a system changes

AI helps a restaurant group most in the work around the food: inquiries, reviews, invoices, ordering and the nightly numbers. What a system does across several locations, what stays with your people, and what to build first.

Read →

How to implement AI in a small firm: the sequence that works

A plain answer for a firm of five to fifty people: pick one workflow with a clear before and after, name an owner, put the data in one place, build a first draft for a person to review, and measure one number. The order matters more than the tool.

Read →

Build or buy: when an off-the-shelf AI tool is enough, and when your firm needs a system built

A decision rule for firms weighing a subscription against a build: buy when the work is generic and lives in one tool, build when it crosses systems and carries your rules, and treat the model itself as something you rent either way.

Read →

AI for immigration law firms: where the hours go, and what a system changes

An immigration practice types the same facts into six USCIS forms, reads the same bank statements for the affidavit of support, chases the same documents, and answers the same question from every client. Here is what an AI system built for that practice does, and what it never does.

Read →

AI for accounting firms: tax season intake, and the documents that never arrive on time

Busy season is a document problem before it is a tax problem. What an AI system does for a CPA firm between the organizer going out and the return being prepared: classify what arrives, chase what has not, and hand the preparer a file that is ready.

Read →

What an AI harness is, and why it decides what the model can do

The model is a brain in a jar. Everything it does in your business, reading files, taking actions, checking its own work, stopping when it should, comes from the harness around it. Here is what a harness is made of and why the same model behaves so differently in two of them.

Read →

The overhang: what AI can already do versus what your firm uses it for

Three ideas from a recent OpenAI note on where AI is heading, read from the seat of a firm deciding what to do this quarter: the overhang between capability and use, spiky skills, and systems that work for hours without a person. And what each one changes about the first system you build.

Read →

Why most AI pilots die after the demo, and the two things that keep them alive

The demo goes well. Six weeks later nobody is using it. The failure is almost never the model. It is two decisions that should have been made before the build.

Read →

Where AI pays off first in a professional services firm

Law firms, accountants, recruiters and advisors run on the same four kinds of work. Three are worth automating now, and the order matters more than the tools.

Read →

What an AI automation agency actually does

The phrase covers everything from a chatbot widget to a full operations rebuild. Here is what the work is, what it is not, and how to tell them apart.

Read →

From a Lovable prototype to a product that passes diligence

The prototype got you customers and a term sheet. Now it breaks weekly and a reviewer is coming. Here is the rebuild that works, and the one that does not.

Read →

Claude or ChatGPT for business automation: how we choose

We build on both. The choice is rarely about which model is smarter. It is about documents, instructions, tools, data handling and where the model can run.

Read →

Rather talk about your own workflow?

Tell us what you have in mind. We'll come back with the highest-ROI path to get there.

Book a free callor write to us at [email protected]