AI transformation · Claude and ChatGPT for your team
How we move a team onto Claude and ChatGPT.
We time your team's most repeated jobs in week one, build a skill for each with the person who does it, teach it in the room, and measure the time again against that baseline.

4,793
45.4
10 structured insights on this week's usage
Operations: strong skill uptake, light on context reuse.
Actionable: add pricing rules to the quote skill.
Who this is for
A team already paying for AI, with no one owning it.
The seats are bought. The pilot happened. A few people found their own uses and the rest went back to how they worked before, because nobody built anything for their job and nobody measured whether it helped. That is the team we move over, and it is almost never a technical one.
We call the method an AI operating system: the goals, skills, training and evals that make AI part of how a team works. It runs on the Claude or ChatGPT organisation you already pay for, and everything we build in it is yours.
Already paying, nobody owning it
Your team has Claude or ChatGPT seats. A few people use them well, most open them twice a week, and nobody can say what the licence has returned.
Small and mid-sized teams
Small enough that we can sit with every person whose job we build a skill for, and large enough that one recurring workflow per person adds up to real time.
New Zealand or Australia
We are Auckland based and teach in the room. We run the training day on site across New Zealand and Australia, and the follow-up sessions remotely.
Goals · Skills · Training · Evals
The method in five steps
Every step produces something you can check: a timed baseline, a skill with a person's name on it, a second timing, a proposed fix. Nothing in it depends on the team being technical, and every result is measured against your own week-one baseline.
- Step 1
Name the goals and time the baseline
Up to three goals. In week one we time how long each job takes today, so every result is measured against your own week-one baseline rather than a number from a slide. The default target is twice as fast.
- Step 2
Pick one workflow per person
Each person names the job they repeat most. Not the most impressive one: the one that eats their week. That is the workflow we build first.
- Step 3
Build it into a skill, taught in the room
We write the skill with the person who does the job, in their words, on their real work, and teach it on the training day. They leave with it running.
- Step 4
Measure the time again
The same job, timed the same way, against the week-one baseline. If it is not faster we keep working on the skill until it is, and we say so plainly when a job was the wrong pick.
- Step 5
A daily eval proposes the next fix
Every day we read what the team actually did with the skills and propose the next change. The team approves it, we ship it, and the skills keep getting better after we have left the room.
What you get in the first month
Six things, each with a name, each of them yours. The skills live in your own Claude or ChatGPT organisation, the baseline and the second timing are written down, and the Sentry AI engineer who built them stays on.
- A goals registerUp to three goals, each with a baseline timed in week one and the measure we will report in: minutes a week, hours a month.
- One skill per personBuilt with the person who does the job and owned by your organisation. A folder, not a saved prompt: the job, how your business does it, and the rules it reads from.
- A training day in the roomClaude or ChatGPT taught on your team's real work, with every skill live by the end of the day.
- Your business written downThe context the model cannot know: your products, your terminology, your tone, how a job is done here. Every skill reads from it.
- The time, measured againAt the end of the month each goal is re-timed against its baseline and reported as a number you can check.
- The daily eval, runningA read of what the team did each day and the next proposed fix, with a named Sentry AI engineer who stays on to ship it.

Teams we have moved over
Figures from the systems we run, as of October 2026. Each is a floor, and each client result is measured against that team's own week-one baseline. Seat by seat, a rollout is on Claude for business, and the programme round by round, two workflows at a time, is the AI adoption programme.
- 185+people connected to the AI systems we run
- 1,500+skill runs a month across client teams
- 45skills in a single client's library
- 7organisations we run AI for today

A full-day process now takes about an hour.
a NZ retail client's weekly shipment process went from a full day to about an hour on the legacy database connection, and every skill is theirs.
See the full case study
40+ hours a week returned to a five-person team.
Creative, Meta buying and the monthly P&L, each built into a skill with the person who runs it. Four-hour jobs down to one, measured against the week they started.
See the full case studyNot a deck · Not a one-off build
Most agencies ship and leave. We deliver every day.
A strategy deck tells you what could be done. A one-off build hands you one thing and goes. Both leave the team exactly where the pilot left it: a few enthusiasts and a licence nobody can account for. The method is different because the work is timed before and after, the skills are built with the people who use them, and we stay on as an embedded partner focused on compounding growth.
It is also not for everyone. If what you want is software built and handed over, with nobody inside the business using AI on their own work, we are the wrong choice and will say so on the first call.
A strategy deck
Tells you what AI could do for the business and leaves the doing to you. Nothing is timed, nothing is built, and six months later the seats are still barely used.
A one-off build
Ships one tool and leaves. It works on the day, then the business changes, the model changes, and nobody owns the fix.
This method
Starts with a timed baseline, builds one skill per person with them, measures the time again, and stays on with a daily eval proposing the next fix.
When we are the wrong choice
A team that wants software built and handed over, with no one inside the business using AI on their own work. We can point you to people who do that well.
Where it leads
The training day is the start. A team that has its skills running usually wants two things next: more of the team in the room, and the knowledge the skills read from kept as something the business can ask.
AI workshops
Half-day and full-day sessions in the room for a team that wants to learn Claude and ChatGPT on its own work before committing to the full method.
See the workshopsSecond brain for your business
Once the skills are running, the knowledge they read from becomes a living record of how the business works, and the team can ask it what only your business knows.
Read about the second brainOpen-weight models
What we are building towards: the same skills running on open-weight models inside your own boundary, for teams where the data cannot leave the region.
Where that is headingQuestions we get asked
What is an AI operating system?
The name we give the method: the goals, skills, training and evals that make AI part of how a team works. Up to three goals with a baseline timed in week one, one recurring workflow per person built into a skill with them in the room, the time measured again against that baseline, and a daily eval that proposes the next fix. It runs on the Claude or ChatGPT organisation you already pay for and the skills are yours.
Our team is not technical. Is that a blocker?
No, it is the normal starting point. We build each skill with the person who does the job, in their words, and teach it in the room. Nobody writes code, and nobody has to remember a prompt: they open Claude or ChatGPT and the skill for their job is already there.
How is this different from an AI training course?
A course teaches people to prompt and leaves. We start by timing the work, build the skill for each person's own job, teach it on that job, and measure the time again against the week-one baseline. The value sits in the skills and the measurement, not in what somebody remembered from a session.
We tried ChatGPT and it never stuck. Why would this be different?
Because nothing was built and nothing was measured. Without a skill written for the job, everyone starts from a blank box and gets a different answer, and without a baseline nobody can say whether it helped. One workflow per person, built with them and timed, is what makes it stick.
Claude or ChatGPT?
We work across both and start with whichever your team already pays for. Most of our teams run on Claude, which is where the training day is built, and we will tell you plainly if a licence you hold is not worth renewing.
How soon do we see time back?
The first skills land in week one, and that is usually where the first hours appear. We measure every result against your own week-one baseline rather than a number from a slide, so you can check it yourself. The default target is twice as fast on each goal.
How much of our team's time does it take?
One training day in the room, then short sessions with each person while their skill beds in. We watch the work rather than run a classroom, so most of the effort sits with us.
What happens after the first month?
We stay on. The daily eval reads what the team actually did and proposes the next fix, and a named Sentry AI engineer ships it with you. Most agencies ship and leave; we deliver every day and focus on compounding growth.
Start with one training day.
Your team in the room, on their real work, with a skill each by the end of it. Everything after is measured against your own week-one baseline.