I build AI to amplify, not replace. Transformation is the work.

I lead AI adoption and enablement. Sometimes that means building the thing myself.

Work

Six systems. Five in production, one handed over.

Five of them build real work on their own, and then stop. Nothing reaches a client or a reader until a person has read it and said yes. Three were built under client deadlines; two run on my own work, where I prove the standard first. The sixth is the operating model that puts the same rule inside a whole company.

Case 01 · CLIENT DELIVERY

Client delivery

Where this applies

Turning expert knowledge into training at volume, without letting quality drift as the volume climbs.

In financial services: Product and compliance training rebuilt every time the rules change, with the compliance officer holding the gate that releases it.

Ten agents. Three human gates. Twenty courses.

Ten AI workers built twenty Finance and Marketing and Sales courses for a boutique consulting firm that trains studio owners to open, grow, and scale. Each one could only answer from the firm's own material, so every sentence traces back to something they had already written. Three times along the way the work stops until a human expert signs off, because review belongs inside the process rather than bolted on at the end.

The human gate: three review points.

20courses shipped

Includes what broke in production

See how it works

Case 02 · COURSE PRODUCTION

Product

Where this applies

Any content operation where publishing a wrong claim is expensive to undo.

In legal and regulated marketing: Client-facing guidance checked line by line against primary sources, with every flagged claim a lawyer's call to fix, cut or keep.

It fact-checked the course I wrote by hand.

A system that carries a course from a bare topic to a finished, published site in eight steps. One of those steps does nothing but check each claim against the source it came from. I pointed it at a course I had written by hand and already published.

The human gate: every finding is my call.

13of 90 claims flagged as wrong

Includes what broke in production

See how it works

Case 03 · THE CONSULTING TEAM

Product

Where this applies

Decisions that need several specialists to disagree in writing before anyone signs.

In professional services and deployed engineering: Proposals and architecture options drafted by specialists, challenged by a critic, and released only under a named human decision.

It refused to let me approve it.

Twenty-two AI specialists that argue with each other in writing before anything reaches me, so I read the disagreement instead of taking a verdict on trust. On a live run, one of them caught a teammate quoting a source that did not actually support the claim.

The human gate: it stays shut until I answer.

22specialist agents

Includes what broke in production

See how it works

Case 04 · THE JOB SEARCH

Personal OS

Where this applies

High-volume triage where the reading can be automated but the sending cannot.

In operations and intake: Claims, vendor or applicant queues: read everything, verify against the system of record, and stop before the irreversible step.

Reads everything. Submits nothing.

Every weekday at 6:15 a system reads the new postings, verifies which are actually open, grades the fit, and builds the complete application. Then it stops, because the send belongs to a person.

The human gate: approve, then submit.

0sent without me

See how it works

Case 05 · THE PUBLISHING ENGINE

Personal OS

Where this applies

Any publishing queue where brand or regulatory risk sits on the last click.

In finance and legal marketing: Campaign copy drafted in the house voice, links and claims checked, published only after a named approver signs that exact piece.

Drafts in my voice. Ships in my time.

An end-to-end publishing pipeline with one non-negotiable: nothing goes out until a person approves that exact piece. The queue is the safety, and the speed.

The human gate: the approval queue.

32posts through the gate

Includes what broke in production

See how it works

Case 06 · THE OPERATING MODEL

Client delivery

Where this applies

An organization that has already bought AI seats at scale and now has to decide what is safe to build, share and scale.

In financial services: Thousands of licensed staff building their own assistants, with anything touching customer data or a credit decision stopped for formal review and everything else governed by platform defaults.

The rollout had energy. What it didn't have was an operating model.

About 4,000 ChatGPT Enterprise licences were in use, and our audit found more than 3,500 GPTs people had built themselves; six had been through formal review. Contracted through Tribe AI, I co-designed the governance with the client and helped six teams build 17 working skills.

The human gate: guest data is the first question.

17working skills co-built with six teams

Includes what broke in production

See how it works

Honest numbers · from the system logs · September 10, 2026

20courses shipped to a client in 6 weeks of build
90claims checked against primary sources in one audit
32posts drafted, checked and published through one queue
~70roles read and verified in a single morning run
252tests green in CI on the agent platform

Full sources on each case-study page.

What this moves

What changes. Not what I promise.

These are the things the systems actually change. The numbers further up are mine, measured on my own work. Yours would be your own, measured with you.

You wait on a decision, not on the workDrafting, checking and packaging happen overnight. You arrive to finished work waiting on a yes, instead of a backlog waiting to be made.
Mistakes get caught before anyone sees themClaims are checked against the sources they came from before a person ever reads the draft, so corrections happen quietly rather than in front of a client.
Your experts judge instead of huntThe system finds the problems and hands them over. Your expensive people spend their time deciding what to do about them, which is the part that actually needs judgment.
Every decision leaves a recordWho approved what, when, and why they said yes. If anyone asks six months later, whether an auditor, a new hire or a post-mortem, there is something to read.
More work, same teamThe work carries volume that would otherwise need more people, while the person accountable for it stays a real reviewer rather than a rubber stamp.
The next person can pick it upEvery deliverable ships with the recipe that made it, written in the client's own words, so the work never becomes something only one person can run.

Human-centered by training, not by tagline.

PROSCI ADKAR certified
PhD (ABD), Education
IDEO Design Thinking trained
Ten years of enterprise transformation

Human-centered is a practice I was trained in, not a word I reached for.

Building a team that keeps people at the center?

Best fit: enterprises putting AI into real workflows, with named owners and gates they are willing to enforce. Not a fit: unattended automation with nobody accountable for the output.

Book a Discovery Call Email Tony

30 minutes. No pitch deck. Same gates I run on my own work.