Dominik Tornow

The 1st Table Mountain Database Management Seminar
Oct 9, 2025

An overnight hike — DBMS philosopher walk and talks, card games, South African braai under starry skies — with TigerBeetle, professors Philippe Bonnet and @peterabcz.bsky.social, and @dominiktornow.bsky.social

#TMDBMS

From brainstorming transactional applications with the @tigerbeetle.com team to drafting blog posts with Table Mountain as backdrop

Cape Town is a whole vibe

In the next chapter of Systems Engineering for Agentic Applications:

When agents leave the terminal, agentic applications become distributed applications.

We'll cover

➡️ Coordination
➡️ Recovery
➡️ Progress

And why you should never hang your agent off a TCP connection 🔪

Distributed Async Await

✅ Simulation tested
❌ Formally verified

Interested in some live coding ... wait, no ... specifying a distributed protocol?!

Who’s up for a webinar to formalize @resonatehqio.bsky.social protocol stack. From scratch. In quint.

Let me see hands

Chapter 2 gets a more formal treatment of Agents, environments, autonomy, alignment, etc

Helpful or heavy?!

MCP: real innovation or literally just a function call?!

Where do you stand in the protocol debate?

The second chapter of Systems Engineering for Agentic Applications just dropped

This chapter covers

➡️ Agents & Environments
➡️ Goals & Plans
➡️ Context Engineering

+ We build an agent to organize our desktop

Bonus: Is MCP just calling functions by name?!

bsky.app/profile/domi...

The second chapter of Systems Engineering for Agentic Applications will drop tomorrow

We'll cover

➡️ Agents and Environments
➡️ Goals and Plans
➡️ Autonomy and Alignment
➡️ Context Engineering

+ We build an agent to organize our desktop

agenticapplications.substack.com

You cannot step into the same river twice
－Heraclitus, Greek Philosopher

An agent is a tuple (model, tools, system prompt, history)

A = (M, T, s, h)

After each interaction, the history changes

A' = (M, T, s, h+i)

Is A' still A? When is A' still A

#PhilosophyFootnote

bsky.app/profile/domi...

Agents don't have a plan

Often, we prompt agents to plan out loud. But agents do not commit to a plan.

The proposed plan becomes part of the history, the context, biasing but not binding future decisions.

Why? Agents are turn based. Every step is a new decision

The #P99CONF speaker gift just dropped.

Box says: Get swag, not lag

I say: 10/10 swag box

Come see me squash Heisenbugs with deterministic simulation testing

bsky.app/profile/domi...

liveBook · Manning liveBooks are enhanced books. They add narration, interactive exercises, code execution, and other features to eBooks.

Hey Iker, super happy to hear you are enjoying the book so far ♥️

I believe Manning's preferred way to report errors including typos is the Manning's livebook discussion forum:

livebook.manning.com/discussion

Thanks for taking the time!!!

Thank youuuuuu

Please let me know if you have any questions, feedback, or comments

From pixels to paper, from Manning to mailbox, Think Distributed Systems is finally in my hands.

Thank you everybody for your encouragement and being on this journey together ❤️

Prompt Engineering, formalized

Prompt Engineering has felt intangible to me. Hype but no precision

In Chapter 2, we define Prompt Engineering as a search problem that maximizes a fitness function on an agent’s trajectory

Too abstract?! Too dense?!

Don't worry. We build this from the ground up

Engineering without a dependable mental model is just guesswork

In Chapter 2, we fuse theory with practice and formalize goals, tasks, trajectories, alignment, and even prompt engineering.

Timeless mental models you'll use forever

In Chapter 2 of Systems Engineering for Agentic Applications, we build our own MCP.

Not another MCP server but our own protocol

We'll discover

- What is MCP
- What does MCP do
- Why would we even need MCP

Resonate Dead simple — Killing the complexity of distributed systems.

Yes, you can use @resonatehqio.bsky.social to implement reliable long-running blocking or non-blocking tool calls

Join our discord at resonatehq.io and we can chat about options and architecture

Durable Execution, distilled

Distributed Async Await is so simple, the API fits on @resonatehqio.bsky.social's landing page

That's not just a feature, that's a flex

As far as I understand, even with http3, a tool call has to conclude on the same (logical) connection

Hot take

MCP's Achilles heel? Requiring a persistent TCP connection

How do you handle long running tool calls?!

Chapter 2 covers a lot of ground:

→ Goals and subgoals
→ Tasks and subtasks
→ Autonomy
→ Alignment
→ Open loop agents
→ Closed loop agents
→ Agent equivalence
→ Agent compatibility

The foundations for building systems, not toys

Your desktop is a mess?

In the next chapter of Systems Engineering for Agentic Applications we build an agent to organize our desktop

A simple task and a perfect lens for understanding how agents work and the systems engineering challenges they present

One of the hardest problems in agentic applications is alignment:

An agent is aligned with the user if the agent ranks possible states the same as the user

No frameworks

In the 2nd chapter of Systems Engineering for Agentic Applications, we’ll build an agent from the ground up

Why?

To study agents from first principles and identify the systems engineering challenges we’ll face at scale

Hey, I am happy you are enjoying the book so far. I am still writing the book, one chapter a month, so unfortunately I cannot publish all at once 😅

The first chapter of Systems Engineering for Agentic Applications just dropped

This chapter covers

1⃣ AI foundations
2⃣ AI agents and agentic applications
3⃣ Code samples exploring AI APIs and simple agents

systems-engineering-for-agentic-applications.resonatehq.io/chapters/cha...

The first chapter of Systems Engineering for Agentic Applications will drop tomorrow

We will cover

➡️ AI foundations
➡️ AI agents and agentic applications
➡️ First contact with AI APIs and AI agents