AI tooling consultation
AI that does the work, not just the talking.
Two things: learning to use AI properly yourself, and getting it into the product you are building. Loops that run on their own, tools an agent can actually use, the rules of your codebase written down, and your developers taught how to ask. Especially useful if AI has not clicked for your team yet.
What I set up
The pieces, in plain language.
Six things, installed around your codebase and handed over working. You keep all of it — it lives in your repository, not in an account I control.
Work that runs without being asked
An agent set on a schedule or a trigger: sweeping the ticket queue overnight, watching a deploy, keeping a long job moving. You describe the job once and it runs on its own from then on.
Giving the model hands
A model that can only talk is a chat box. Wired to your own systems — the database, the ticket tracker, the deploy — it can do the task instead of describing it. That wiring is the work, and it is where the value is.
The written-down rules, in the repo
Plain markdown files that tell an agent how this codebase works, what the conventions are, and what never to touch. They turn scattered tribal knowledge into something every developer and every agent reads the same way.
Your team’s procedures, packaged
The multi-step jobs your team does the same way every time — a release, a review, a new project scaffold — written once so anyone, or any agent, can run them properly on the first try.
Asking in a way that works
Most disappointing AI output is a briefing problem, not a model problem. I show your developers what a good ask looks like, and where to stop trusting the answer.
Deciding what the model gets to see
Quality comes from what you feed it: the right records, at the right size, from your own data. This is the difference between a demo and something a business can rely on.
A worked example
I run my own day on this.
Several small apps I built for myself, each one built with AI and each one now part of how I work. None of them existed a year ago, and none of them would have been worth the hours by hand — that shift is the whole point, and it is what I set up for other people.
A control room for my agents
I run several coding agents at once, so I built the place that keeps track of them: what each one is working on, what it produced, and which ones are waiting on me.
A dynamic island for my Mac
The little status strip Apple gives the iPhone, rebuilt for my desktop — what is building, what just finished, and what needs a decision, glanceable without switching windows.
A nudge to get up and do push-ups
The job is sitting down, which is bad for you, so this one interrupts it — a reminder to stand up and knock out a set. Small thing, but it is the difference between moving during the day and not.
Somewhere for every loose end
Tools that catch the things I mention in passing and hold onto them until they are done, so follow-ups stop falling through the cracks.
In production, not in theory
Two products with my AI work inside them.
Ruta AI
Service management platform · team product
An assistant answering from the company's own records, taking actions with a person confirming them, and reachable from anywhere in the app — on the web and in the field.
Read the case studyThe Tap to Tick money coach
iPhone expense tracker · my own product · coming soon
Ask a question in plain English and get an answer built from your own numbers — and only the numbers that bear on the question ever leave the phone.
See the appWhat I will tell you straight
- It will not replace your developers, and I will not pretend otherwise.
- Anything irreversible stays behind a person confirming it.
- A cheaper model per task where a cheap model is enough, so the bill stays sane.
- If a job is better done by ordinary code, I will tell you that instead of selling you a model.
Want your team set up properly?
Consultation hourly fee: $75/hr, booked in two-hour blocks — over your choice of Zoom, Google Meet, FaceTime, or whatever you already use. Build work is scoped and quoted per project instead.