Personal AGI

Live workshop · 26 Aug & 9 Sept

An AI agent that knows your clients, your projects, your judgment — and gets better every day you use it.

Not a chatbot you rent. A system that runs on your own knowledge, does real work — research, briefings, finished drafts, overnight — and belongs to you. A two-part live workshop for anyone whose work runs on what they know: founders, consultants, solo operators, professionals of every kind. You'll watch a real one work, watch it break, and see exactly what has to be true before you can trust it.

Sessions Wed 26 Aug + Wed 9 Sept, 12–3pm ET Seats 20 Price $600
Reserve a seat

Nothing to install to attend. No technical background needed.

Sixty seconds on what this is — and why almost nobody has one yet.

The gap

Same model, same subscription, two completely different outcomes.

You've seen the disappointing version: impressive for ten minutes, then it hands you something plausible and wrong, and you go back to doing it yourself.

Meanwhile, some people run their working lives on exactly the same model:

The difference isn't talent, and it isn't prompting. It's that their AI knows their work — their clients, their history, their way of doing things — and that knowledge shows up at the right step.

That's not a better product you can rent. It's a system you build once, own outright, and that compounds every day you use it.

What actually happens in the room

Three demos, each designed to fail first.

A claim that size deserves scepticism, so nothing here is slides. Every idea gets taught the same way: run a real task against a real system with a real flaw in it, watch the failure, then fix it and run it again. The failures are the useful part — they're what teach you when a system can be trusted.

1 — The confident wrong number

Ask the system something that needs arithmetic across dozens of records. It answers immediately, in fluent full sentences, and it's wrong. Then we change one thing about the setup and it writes and runs its own code to get the right answer. If AI has ever quietly mangled a number for you, this is why — and this is the fix.

2 — The employee who misunderstood you

A written instruction that's just slightly ambiguous. The system does the wrong thing, competently. We correct it live, fold the exception back in, and run it again. This is what managing an agent actually looks like day to day.

3 — The confidently stale answer

Two records disagree about a client's budget. The system picks one and states it as fact. We show what has to be true about a system before you can trust anything it tells you.

Then we run the same thing on a real business belonging to someone in the room.

The curriculum

Two half-days, seven hours, and a week in between.

Session one is where you see the whole architecture and understand why it works. Session two is where yours gets working.

Session one · Wednesday 26 August

Why a curated system behaves differently

The architecture, in plain termsA rented model, your own context, and the harness that wires them together. Why the model is the commodity and the context is the asset. What "personal AGI" means once you strip the marketing off it.
What a library actually looks likeA guided tour of a real one — projects, people, decisions, a year of accumulated work. What makes a good page, what makes a useless one, and why every fact carries a source and a date.
Break/fix 1 — the confident wrong numberWhere computation belongs. Judgment and taste go to the model; arithmetic, dates and anything exact go to code. We watch it fail one way, then work the other. This is the failure behind most of the AI mistakes you've already had.
Break/fix 2 — the employee who misunderstood youSkill files: one job, written in plain English, clear enough that a new hire could follow it. We run an ambiguous one, watch it do the wrong thing competently, then correct it and fold the exception back in. This is the loop you'll live in.
Break/fix 3 — the confidently stale answerMemory hygiene. Provenance, contradiction handling, and pruning. What has to be true about a system before you can trust what it tells you — and why an uncurated one is worse than no system at all.
Live, on a real businessThe same demos run against an actual attendee's context, built from what they send in at registration.
Your build orderExactly what to install, in what order, and what to write first. Plus an honest account of what week one feels like.
The two weeks between

You build. I'm on hand.

Two group office hoursOne in each week. Drop in with whatever's broken. Most blockers turn out to be the same three or four, so everyone learns from each one.
Email supportStuck on a Tuesday night and the next office hour is days away? Email me — I reply within one business day. That gap is where people quietly give up, so it isn't left uncovered.
Private sessionsFor the Working tier — two 60-minute sessions on your own clients, your own context, your own setup.
Session two · Wednesday 9 September

Getting yours running

Setup triageLive troubleshooting, screens shared. This is scheduled time, not an afterthought — the assumption is that a good part of the room will arrive stuck, because that's what happens.
The failure clinicYour real failures from the week, diagnosed in front of everyone against a three-question framework: was the context missing, was the instruction ambiguous, or was exact work handed to judgment? Almost everything that goes wrong is one of those three.
Work that finishes while you sleepRecurring jobs. A morning briefing, a Friday summary, overnight research. Plus the rule for when a job is safe to automate and when automating it will just produce a week of garbage.
Ownership and portabilityGetting it into a repository you control. Why the same files in someone else's repo are a completely different future, what travels with you between jobs and clients, and how to check that yours actually would.
Never do the same work twiceThe discipline that separates people who compound from people who dabble: ending every task by turning it into something reusable.

The tools

Five things, and most of them are free.

Nothing proprietary, nothing you buy from me, no platform to get locked into.

Claude CodeThe harness. The agent that runs on your own machine and does the work.$20–100/mo
ObsidianWhere your library lives. A friendly window onto a folder of plain text files.Free
Git & GitHubYour undo button, your off-machine copy, and the thing that makes it provably yours.Free
MarkdownThe only language you write in. It is plain English with the occasional hash symbol.Free
A terminalUsed lightly. You'll be told exactly what to type, and why.Free

The concepts matter more than the tools. Claude Code is what I use and what I'll demo, but the same architecture runs on Codex, OpenClaw and others. Nothing you build is trapped: your library is plain text files in a folder you own, readable by anything, for as long as text files exist.

What you leave with

You won't be installing anything during the sessions. You watch, then set up afterwards on your own machine — with help. That's deliberate: ten people installing software live is how workshops lose an hour and everyone's first impression.

Session two is the clinic

Most workshops end and you're on your own.

This one assumes you'll get stuck, because everyone does. That's why half of session two is scheduled troubleshooting and clinic time rather than more content — and why the two weeks in between come with a live office hour each week and email support, rather than a link to a help centre.

Who this is for

The architecture doesn't care what you do for a living. It cares whether your work accumulates — whether you carry clients, projects, people, decisions and hard-won judgment around in your head and in scattered notes.

People in the room will include founders, consultants, agency owners, solo operators, researchers, lawyers, accountants, product and marketing people. The demos use a consultant's library because it's a shape most people recognise, but nothing in the method is specific to consulting.

A good fit if

  • Your value is your judgment and what you've accumulated
  • Your work compounds — clients, projects, decisions, relationships
  • You've tried the AI tools and found them shallow
  • You'll spend a couple of hours setting things up afterwards

Not a good fit if

  • You want a done-for-you service
  • You're looking for prompt templates
  • You want to attend and do nothing afterwards

No technical background needed. You will see real code appear on screen. You will not write any.

The part most courses won't tell you

Week one feels like a toy.

The system is thin, you're fixing more than you're saving, and the obvious conclusion is that it doesn't work. That's the normal curve, not failure — and it's exactly where most people quit.

That's precisely why this is two sessions two weeks apart with support running through the gap, rather than one impressive afternoon.

Format

Session 1Wednesday 26 August · 12:00–3:00 pm ET · live online
BetweenTwo group office hours, one each week, plus email support — replies within one business day
Session 2Wednesday 9 September · 12:00–3:00 pm ET · live online
Seats20
RecordingYes — yours to keep
Running cost afterA Claude subscription, roughly $20–100/month. Nothing else.

Pricing

Three ways in.

Standard
$600
20 seats

Both sessions, both office hours, email support across the two weeks (replies within one business day), the sample library, recordings, and the community channel.

Reserve a seat
Working
$1,250
4 places

Everything in Standard, plus two private 60-minute sessions on your own business — your clients, your context, your setup, built with you.

Take a place
Private team
$5,000
Up to 10 people

The same workshop run privately for your company, on your schedule.

Get in touch

Founding cohort. This is the first run. Everyone in it keeps lifetime access to future cohorts, gets the self-paced course free when it launches, and gets early access to the tool being built from this.

Guarantee. Attend session one, and if you don't think it was worth the ticket, email me and I'll refund you in full. No form, no conversation.

Who's running it

Rui Luis
Rui Luis · Burlington, Ontario

Rui Luis

I've started three companies over the last thirty years and taken one of them public. Long enough to know the difference between a tool that demos well and one that survives contact with real work.

For the past three years I've built nothing but AI systems. And I should be straight with you about something, because it's the whole reason this workshop exists: I have no coding background. I never trained as an engineer. What I have is a keen interest and the same tools I'm going to put on screen. Two of the things I've built with them are live right now:

Aether — an AI-native platform for consulting firms. I architected and shipped the MVP solo: four agents, connectors for Slack, Gmail and Google Drive, and a retrieval engine that cites its own sources.

Tend — every relationship tool is a contact list with a timer, nagging you to check in with people for no particular reason. Tend flips that. You tell it what you're trying to achieve — land your next three clients, close a raise, make a key hire — and it organises your network around that goal: who actually moves it, what your next move is with them, who to stop spending time on, and who's missing entirely. Every suggestion cites the conversation it came from, so it isn't a black box.

I also build custom agents and automations for consulting, recruiting and creative firms.

And I run the exact architecture in this workshop, every day. My library lives in Obsidian. Around thirty skill files. A routine that reconciles what I actually shipped against my tickets and writes the decisions into a daily note without being asked. Jobs that produce finished work while I'm asleep.

That's the part that matters here. I'm not teaching a framework I read about, and I'm not an engineer showing you engineering. I'm a business person who built all of this with the system I'm about to show you — including the places it breaks.

Questions

Do I need to install anything beforehand?

No. Turn up and watch.

What will it cost me to run afterwards?

A Claude subscription, roughly $20–100/month depending on how hard you push it. No tool purchases, no platform fees.

Is this just prompting tips?

No. Prompting is the part that stops mattering once the system is right. This is about what the system knows, and when.

Mac or Windows?

Either — you don't install anything to attend, so it makes no difference on the day. I demo on a Mac, and you'll get a tested step-by-step for both when you set up afterwards. Windows needs a couple of extra steps at install; everything past that point is identical.

Isn't this only for technical people?

No. You'll see code appear; you won't write it. The parts you write are plain English.

Where does "personal AGI" come from?

The term is Garry Tan's — president of Y Combinator — from his talk "Personal AGI Is How You Stay Under Your Own Power". Worth watching: it's the manifesto, aimed at startup founders and engineers. This workshop is the how-to for everyone else — the same architecture built live, broken live and fixed live, with support while you build yours.

What if I can't make a session live?

Both are recorded. The office hours and the clinic are where the live value is, so try to make session two.

I'm in a different timezone.

Both sessions run 12:00–3:00 pm Eastern — that's 9am Pacific and 5pm in the UK. Ask before booking if that's unworkable; if there's enough demand elsewhere I'll run one at a friendlier hour.

Almost nobody has one of these yet.

Not because it's hard — because nobody has sat them down and shown them what a working one looks like, and where it breaks. That's the whole workshop. The model gets cheaper every quarter; what it knows about you is the part that compounds, and the part you own.

Sessions Wed 26 Aug + Wed 9 Sept, 12–3pm ET Seats 20 Price $600
Reserve a seat