SSE Riga · Lecture archive13 October 2026 · 18 slides · Read-only · v33 · student edition

Build something useful.
Let it run.

The lecture as shown at SSE Riga, with recorded examples. This archive does not connect to live intake or private presenter controls. Numbers belong to their displayed dates; live classroom counters are omitted; class ideas have their own archive.

Download the complete offline lecture · Learn the loop model

01 / 18 · FROM SSE RIGA TO BUILDING WITH AI

Build something useful.
Let it run.

01Use it.02Check it.03Design yours.

I sat where you sit. My first business: free student copies, paid for by ads.

You are
the customers.

1. Scan.
2. Type one skill or problem.
3. Open, rate and keep your page.

Keep one weakness in mind. We’ll use your page again.

Recorded startup-concept page
Workshop exercise · see the recorded example. The live intake is not part of this archive.

Jeff Burton.

Co-founder of Electronic Arts

  1. When did you know EA's customers really wanted what you built?
  2. What did you never delegate as a founder?
  3. Here are the first startups the machine made from this room. Pick one. Is it a business? What is missing?
  4. If you were 21 and in this room today, what would you learn and build?

While we talked,
it worked.

Pages delivered in class—
Classroom rating—

Recorded classroom demonstration. Live room counters and student submissions are not included.

The service delivers pages automatically. People built and maintain it.

05 / 18 · FIVE LOOPS · 1 / 5 Cloud · fixed workflow

The 60-Second Startup

The 60-Second Startup: actual outputA generated page is a prototype.
Use and payment test demand.1pages delivered · recorded 13 Oct, 12:56 RigaView the public example ↗

REMEMBERS Job state + page + instructions

01Request→02Draft→03Check→04Deliver→05Next request↻

Fail? Repair once, recheck, then stop if it still fails.

06 / 18 · FIVE LOOPS · 2 / 5 Cloud · fixed workflow

While You Sleep

While You Sleep: actual outputHourly research.
Publish only what passes.7ideas published · recorded 13 Oct, 12:13 RigaView the public example ↗

REMEMBERS Sources + posts + slot log

01Due slot→02Research→03Check→04Publish / skip→05Record↻

No qualifying post? Publish nothing this slot.

07 / 18 · FIVE LOOPS · 3 / 5 Local Codex · creative agent

The Long Run

The Long Run: actual outputRecorded local run.
Saved context drives the next page.74pages · recorded 11 OctView the public example ↗

REMEMBERS Recap + cast + last page

015-min attempt→02Read recap→03Create→04Inspect→05Save↻

Needs the Mac awake and Codex connected. A failed attempt adds no page.

08 / 18 · FIVE LOOPS · 4 / 5 Local Codex · creative agent

The Unfinished Atlas

The Unfinished Atlas: actual outputRecorded local run.
Each tile extends the same map.149tiles · recorded 11–12 OctView the public example ↗

REMEMBERS Coordinates + neighbouring tiles

015-min attempt→02Read neighbours→03Draw→04Inspect→05Save / upload↻

Needs the Mac awake and Codex connected. Failed checks add no tile.

09 / 18 · FIVE LOOPS · 5 / 5 Cloud · bounded agent turns

The class simulation

The class simulation: actual outputA class that talks,
refines ideas and votes.436messages · recorded 13 Oct, 12:48 RigaView the public example ↗

REMEMBERS Profiles + personal and shared history

01Tick→02Read context→03Choose action→04Enforce rules→05Save↻

38 fictional roles, one configured model. Rule checks are not fact checks.

10 / 18 · THE COMMON PATTERN

A task ends.
A service runs again.

A one-off task moves from request to work to done. A recurring service takes an order, makes, checks and delivers a result, then accepts the next order.

People still own the goal, the limits and the exceptions.

11 / 18 · OUR BUSINESS DESIGN MODEL

Three jobs to design.

GrowUseful page → share link → next user.

ServeMake → check → deliver.

Improve15 ratings → test new instructions → keep if better.

Illustration, not live data. A design lens—not three required agents. Changes update instructions, not model weights.

12 / 18 · TWO WAYS TO ORGANISE THE WORK

Workflow or agent?

Fixed workflow

Code chooses
the sequence.

Startup: draft → check → repair.

Agent

AI chooses
an allowed action.

Simulation: talk, refine or vote.

Both can run unattended within a defined scope.

13 / 18 · RUNNING UNATTENDED NEEDS MORE THAN A MODEL

What keeps
the routine running?

TriggerA clock or event
RuntimeA worker that stays available
StateProgress saved between runs
BoundariesChecks, limits and a stop path

You set the goal, budget and stop rules. Monitor failures, cost and manual work.

14 / 18 · THE CHALLENGE

One loop.
Real users.

In four weeks, launch something people use every week.
Let its routine run without you.

The topic is open. The users must be real.

15 / 18 · SMALL STEPS, EVERY DAY

Your four weeks.

Week 1Ship.

Practise daily with AI.
First version. 5 users.

Week 2Run.

Every day.
10 real users.

Week 3Grow.

30 real users.
Try for the first euro.

Week 4Prove.

Learn from feedback.
Rehearse the demo.

30+ users outside your team, or 3 paying2 weeks running daily on its ownPeople return week by week

16 / 18 · EXERCISE 2 · DESIGN YOUR OWN

Describe one useful loop.

Every [trigger], we make [result]
for [people we can reach].

We check [your rule].
If it fails, we [your safe action].

Reuse your problem—or choose a better one. Add a quality check and a safe failure action.

17 / 18 · WORKSHOP EXERCISE · RECORDED CONTEXT

Your turn: send your ideas.

In class: seven minutes to send ideas. Then read the Top 3 and ask what makes number 1 good.

Browse the workshop ideas →

This copy does not connect to live intake, scoring or the projector board.

18 / 18 · JOIN THE CHALLENGE

What will you launch?

At 14:00 · with DmitrijsTurn a loop into
a running n8n workflow.Bring your best idea and your laptop.

Teams form Thu 15 Oct.
Final demo at the end of the course project.
SSE sets the date.

AI tool budget for up to 3 teams, from Danielius.

Everything: the lesson, examples and how to joinEverything: the lesson,
examples and how to joinxndanius.github.io/
autonomous-business-lab
Businesses that run while you sleepSSE Riga · 13 Oct 2026

FROM SSE RIGA TO BUILDING WITH AI

Build something useful.
Let it run.

01Use it.02Check it.03Design yours.

I sat where you sit. My first business: free student copies, paid for by ads.

You are
the customers.

1. Scan.
2. Type one skill or problem.
3. Open, rate and keep your page.

Keep one weakness in mind. We’ll use your page again.

Recorded startup-concept page
Workshop exercise · see the recorded example. The live intake is not part of this archive.

Jeff Burton.

Co-founder of Electronic Arts

01 / 04

When did you know EA's customers really wanted what you built?

The machine works while we talk.

While we talked,
it worked.

Pages delivered in class—
Classroom rating—

Recorded classroom demonstration. Live room counters and student submissions are not included.

The service delivers pages automatically. People built and maintain it.

FIVE LOOPS · 1 / 5 Cloud · fixed workflow

The 60-Second Startup

The 60-Second Startup: actual outputA generated page is a prototype.
Use and payment test demand.1pages delivered · recorded 13 Oct, 12:56 RigaView the public example ↗

REMEMBERS Job state + page + instructions

01Request→02Draft→03Check→04Deliver→05Next request↻

Fail? Repair once, recheck, then stop if it still fails.

FIVE LOOPS · 2 / 5 Cloud · fixed workflow

While You Sleep

While You Sleep: actual outputHourly research.
Publish only what passes.7ideas published · recorded 13 Oct, 12:13 RigaView the public example ↗

REMEMBERS Sources + posts + slot log

01Due slot→02Research→03Check→04Publish / skip→05Record↻

No qualifying post? Publish nothing this slot.

FIVE LOOPS · 3 / 5 Local Codex · creative agent

The Long Run

The Long Run: actual outputRecorded local run.
Saved context drives the next page.74pages · recorded 11 OctView the public example ↗

REMEMBERS Recap + cast + last page

015-min attempt→02Read recap→03Create→04Inspect→05Save↻

Needs the Mac awake and Codex connected. A failed attempt adds no page.

FIVE LOOPS · 4 / 5 Local Codex · creative agent

The Unfinished Atlas

The Unfinished Atlas: actual outputRecorded local run.
Each tile extends the same map.149tiles · recorded 11–12 OctView the public example ↗

REMEMBERS Coordinates + neighbouring tiles

015-min attempt→02Read neighbours→03Draw→04Inspect→05Save / upload↻

Needs the Mac awake and Codex connected. Failed checks add no tile.

FIVE LOOPS · 5 / 5 Cloud · bounded agent turns

The class simulation

The class simulation: actual outputA class that talks,
refines ideas and votes.436messages · recorded 13 Oct, 12:48 RigaView the public example ↗

REMEMBERS Profiles + personal and shared history

01Tick→02Read context→03Choose action→04Enforce rules→05Save↻

38 fictional roles, one configured model. Rule checks are not fact checks.

THE COMMON PATTERN

A task ends.
A service runs again.

A one-off task moves from request to work to done. A recurring service takes an order, makes, checks and delivers a result, then accepts the next order.

People still own the goal, the limits and the exceptions.

OUR BUSINESS DESIGN MODEL

Three jobs to design.

GrowUseful page → share link → next user.

ServeMake → check → deliver.

Improve15 ratings → test new instructions → keep if better.

Illustration, not live data. A design lens—not three required agents. Changes update instructions, not model weights.

TWO WAYS TO ORGANISE THE WORK

Workflow or agent?

Fixed workflow

Code chooses
the sequence.

Startup: draft → check → repair.

Agent

AI chooses
an allowed action.

Simulation: talk, refine or vote.

Both can run unattended within a defined scope.

RUNNING UNATTENDED NEEDS MORE THAN A MODEL

What keeps
the routine running?

TriggerA clock or event
RuntimeA worker that stays available
StateProgress saved between runs
BoundariesChecks, limits and a stop path

You set the goal, budget and stop rules. Monitor failures, cost and manual work.

THE CHALLENGE

One loop.
Real users.

In four weeks, launch something people use every week.
Let its routine run without you.

The topic is open. The users must be real.

SMALL STEPS, EVERY DAY

Your four weeks.

Week 1Ship.

Practise daily with AI.
First version. 5 users.

Week 2Run.

Every day.
10 real users.

Week 3Grow.

30 real users.
Try for the first euro.

Week 4Prove.

Learn from feedback.
Rehearse the demo.

30+ users outside your team, or 3 paying2 weeks running daily on its ownPeople return week by week

EXERCISE 2 · DESIGN YOUR OWN

Describe one useful loop.

Every [trigger], we make [result]
for [people we can reach].

We check [your rule].
If it fails, we [your safe action].

Reuse your problem—or choose a better one. Add a quality check and a safe failure action.

WORKSHOP EXERCISE · RECORDED CONTEXT

Your turn: send your ideas.

In class: seven minutes to send ideas. Then read the Top 3 and ask what makes number 1 good.

Browse the workshop ideas →

This copy does not connect to live intake, scoring or the projector board.

JOIN THE CHALLENGE

What will you launch?

At 14:00 · with DmitrijsTurn a loop into
a running n8n workflow.Bring your best idea and your laptop.

Teams form Thu 15 Oct.
Final demo at the end of the course project.
SSE sets the date.

AI tool budget for up to 3 teams, from Danielius.

Everything: the lesson, examples and how to joinEverything: the lesson,
examples and how to joinxndanius.github.io/
autonomous-business-lab