For Kargas · 14 May 2026

Opta

Before we walk in together, let me show you what I've been building — and why.

Three acts Thirty minutes Then the summit
scroll ↓
Part I · Opening— why we're here
§ 01 — The Opening

Every chess game starts with the same problem.

You need to set up a position you can actually play from. Get that wrong and no clever move later will save you.

I had the same problem with AI.

No single model is good at everything. The one that's brilliant at writing is mediocre at code. The one that codes well forgets what you told it yesterday. You end up paying four subscriptions, switching between four browser tabs, copy-pasting context between them like a person trying to play four games of chess on four different boards at once.

So I stopped picking models. I built the position instead.

Opta is one AI you talk to. Behind the scenes, the right model handles each move — the writer when you write, the coder when you code, the researcher when you research — and they share the same memory of who you are and what you're working on. One identity. One memory. Different specialists for different positions.

That's it. That's the whole idea. Everything else we'll look at this morning is just how it works.

§ 02 — The chess idea behind everything

Position over move.

A 1200-rated player calculates. A grandmaster looks at the same board and sees something completely different. Click between the two views below.

What a single model sees

Calculate.

Search the next three moves. Look for a tactic. If Ng5 hits something, follow it up with Nxf7, bring the queen to h5, hope it works.

  • 1. Nf3 → g5
  • 2. Nxf7 (check?)
  • 3. Qh5+

Opta is built to play the position.
Not just compute the next move.

Most AI tools work in tactical view. You ask a question, they search for a clever answer. They don't know your work. They don't remember yesterday. They don't pick the right specialist for the job. They just calculate.

Opta works in positional view. Before the move plays itself, it sets up the position: which model is right for this, which memory matters, which tool does the doing, what's the context. The clever answer becomes inevitable, because the position was right.

If you remember one thing from this morning, remember that.

Part II · Middlegame— how it actually works
§ 03 — Pieces & Roles

A king alone isn't a game.

You need pieces with different strengths working together. AI is the same. There are three layers to it — and once you see them, the whole system makes sense.

Brain · the strategist

Where the thinking happens.

Cognitive engines — the part that reads what you wrote, understands what you mean, picks the right approach. Think of it as the king's mind: it doesn't move pieces directly, but everything starts here.

Body · the hands

Where the doing happens.

Tools, files, browser, terminal, the things that actually move. The knight, the bishop, the rook. The Brain decides; the Body executes. Most AI products have a Brain and almost no Body — they can write a beautiful answer but can't open a file for you.

Mind · the reflective mode

Where the learning happens.

Catches its own mistakes. Reads back what it did yesterday. Notices a pattern in your work and adjusts. This is what makes Opta one continuous thing rather than a thousand forgettable conversations.

§ 04 — The Apps

These are the boards I play on.

Same pieces, same memory, different positions. Four apps, each shaped for a different way of working. Here's what each one is for, plainly.

HQ High

The control room.

The window I keep open in the background while I work. Birds-eye view of everything Opta's doing — what's running, what got done today, the stats, the history. Comfortable. Easy. Not where the focused work happens — where I check in on it.

Terminal Proven

The cockpit.

A keyboard-driven AI workspace. As capable as the best terminal coding tools out there — but built for Opta's architecture from the ground up, not bolted onto someone else's. Where I do focused work that needs speed.

Deploy Medium

The studio.

The native desktop sibling of Terminal. Same depth, same AI, different shape — a calmer surface for the same work. You can hop between Terminal and Studio without losing your place; whichever feels right for the moment is the right one.

Gateway Medium

The front door.

The desktop app that handles signing in, keeping everything updated, and showing the system's health at a glance.

Two of these are about doing the work — Terminal and Deploy. They're peers, not a beginner/pro split. They share the same memory, the same models, the same project state; you pick whichever surface fits the moment. The other two — HQ and Gateway — are about looking after the work. HQ is the always-open view from above; Gateway is the front door.

The honesty I want to give you: those dots aren't marketing. Proven means I use it every day and it works. High means it's solid. Medium means it works for me but isn't ready for strangers. I'd rather show you the real state than a polished demo.

§ 05 — Mission

Why I'm actually building this.

There's a story I want you to know — because it's the reason any of this exists, and it's the thing I'll be representing all day today.

People are stacking subscriptions. Twenty dollars here, forty there, two hundred for the good one. They're paying for AI that runs in a data centre on the other side of the world while a perfectly capable computer sits idle on their desk.

The industry says the answer is bigger models. Trillion-parameter monsters that need a nuclear reactor to run. I think that's the wrong answer. The answer is better architecture.

A Mac Mini can play like a grandmaster — if it knows the position. The right specialist for the right job. The right memory at the right moment. The right tool already in hand when the move comes up. That's not a hardware problem. That's a design problem.

Opta is the design.

I'm building this because it shouldn't cost what it costs. And because the best AI any of us will use shouldn't have to live somewhere else.
Part III

Endgame.

Today · noon · Melbourne Town Hall
§ 06 — Who'll be in the room

A quick read of what we're walking into.

I've done the homework so you don't have to. Here's what I've actually been able to confirm about today, what I'm inferring, and what's still unknown. Click any card for more.

The event Verified
Open Summit AI
"Australia's largest agentic AI convention." Promoted heavily by Number (number.com.au), a Melbourne voice-AI agency.
Scale Verified
~1,800
Operators, founders, agencies. Tickets from AUD $249. Format: keynotes + panels + workstations + networking.
The vibe Verified
Operator-led
Pragmatic. Anti-hype. People who ship agentic AI, not researchers presenting papers.
The tagline Verified
AI for Business.
Live. In Person. The whole pitch is "demystify agentic AI for business owners." Newcomer questions are on-pitch.
§ 07 — How we play the room

Five principles. No scripts.

You're the chess player here, not me — when it comes to reading people. So this isn't a playbook. It's the position I want us walking in with. You decide the moves in real-time.

01

Listen first.

A good game is mostly watching your opponent set up. Most people at this thing will want to talk about what they're doing. Let them. The more they talk, the better we'll understand whether there's a real fit.

// the opening is about understanding the position, not making moves
02

Ask about their work.

People remember being asked, not being pitched. "What are you working on?" is the strongest opener in any room. Genuine curiosity is rare and people light up when they meet it.

// developing pieces before attacking
03

Mention Opta only when the position calls for it.

If someone's struggling with the exact problem Opta solves — say something. If they're not — don't force it. Pitching unprompted is the AI equivalent of attacking before your pieces are developed. You lose the position.

// forcing a move is how you lose the position
04

Trade contacts, not promises.

A name, a real moment, and an honest follow-up beat ten business cards and a vague "we should talk." If we leave with three real connections, that's a great day. Five is exceptional.

// quality of pieces > quantity of pieces
05

Leave them better than you found them.

If you can give someone a useful idea, an honest reaction to their work, or just a thoughtful question — do it. Even if there's nothing in it for us. Especially then. That's how you become someone people remember.

// the position you leave behind matters more than the move you played

That's the whole thing. No scripts. No "elevator pitch." If you want my one-line version of Opta for when someone genuinely asks, here it is: one AI that gets continuously better, built so the right specialist handles each thing — and so it can live on your own computer if you want it to.

But you'll find your own version. Yours will be better because you'll mean it in the moment.

§ 08 — After the bell

Let's go.

Whatever happens at noon, the position we built before walking in is the part that matters. The conversations we have today are openings. The follow-up next week is the middlegame. Whatever comes out of it is the endgame — and that's mostly luck and patience.

You're here because I trust your read of people. I'll handle the AI half. We'll figure the rest out together, the way we always do.

Thanks for coming, Kargas.
— Matthew · 14 May 2026
Back to the start ↑ The deeper version →

If you want to read more about Opta later, that link above is the longer write-up.