In late July, we ran our third Agentic Engineering meetup, and our second here in Lisbon. Big thanks to The Nest for having us back, and to everyone who came out!

After three events, a few things are becoming clear.

The demand for quality AI events is high

People in Lisbon are excited about AI and agentic workflows. They’re using AI to ship products faster, build businesses, and manage their personal lives (and apparently their dating profiles). The city is becoming a hub for founders, engineers, investors, and generally curious people, many of them working through similar problems. The growing turnout tells me people are hungry for a community, and for high-quality, in-person events where they can find it.

The lightning talk format works best

Three short talks keep everything focused, and leave enough space for people to actually meet each other and discuss. Some might see the strict 10-minute timer as limiting, but I see it as a feature. It forces speakers to practice their talk and only leave in what’s absolutely essential. We now have more people wanting to speak than we have slots, which tells me the bar is only going higher from here.

We can go a lot bigger

We’re going to continue running these meetups on a monthly cadence, likely landing on the last Wednesday of the month. Consistency is what turns an event into a community.

Building the largest AI community in Lisbon isn’t something one or two people can do as a side gig. Every event so far has shown us things we want to do better. So we’re building a team, and there are five areas where we need help.

How to contribute

1. Organize: Help with planning and logistics, check people in at the door, take photos, and record the talks.

2. Build: Ship the website and the other tools we need (like our own dedicated agent).

3. Communicate: Run socials, edit video, write the newsletter, and keep the Telegram group engaged between events.

4. Teach: Write technical content, give talks, and help us find and vet speakers.

5. Support: Connect us to sponsors, partners, and sources of public funding.

None of these require a huge commitment on their own, but together they add up to a lot. What I’m looking for is a small group of people who care about this as much as I do, and who get real ownership over building it.

The Talks

Apologies for the video quality on these. It’s exactly why we need someone dedicated to filming!

Faster Isn’t Free

Pia Steiner argues that AI gave back productivity, not hours, and explains why moving from the creative seat to the reviewing seat is quietly draining people.

• Reviewing agent output replaces earned dopamine with spike-and-crash hits that lower your baseline.

• Parallel tabs and agents erase natural pauses; nothing finishes, so background stress never stops.

• AI burnout never hits while you type. It hits at 9pm, Sunday evening.

• Losing one burned-out person who stayed costs a founder roughly six months of salary.

• We sandbox our agents obsessively, then run ourselves with no curfew or permissions.

Behalf.bot

Sean Tierney demos behalf.bot, the open-source personal agent he runs daily to ship software, handle life admin, and screen his dates.

• Loom videos convey ideas to an agent better than typing. His Loom Vision skill reads screen and audio together, then reports back as audio.

• The dating bot drives an Android emulator, writes openers, learns his type, books coffee dates.

• The same agent runs life admin: insurance claims, HOA negotiations, a partner buyout.

• Everything is open source at behalf.bot.

Sean’s X, GitHub, LinkedIn

Mind as Model

Rachel Papirmeister presents a puzzle benchmark that current AI can’t solve, and the human learning mechanism her user study uncovered inside it.

• Neural networks transfer skills well but adapt poorly to unfamiliar rules and environments.

• Humans learn without a clear reward signal, something her benchmark is built to isolate.

• Across 75 participants she found a third mechanism between exploration and exploitation: establish.

• Establish is hypothesis testing — you form, revise, and confirm rules while operating on a partial world model.

• Symbolic approaches offer interpretability and data efficiency; cognition points toward architectures with built-in hypothesis testing.

Rachel’s GitHub, LinkedIn

What’s Next

The next meetup is on August 26 at The Nest. If you’re building in this space, come to the next one. And if you want to help build this thing with us, reach out.

For as long as I’ve been coming to Berlin, it’s been the place to find the most interesting collection builders, hackers, cypherpunks, and founders. There’s a unique underground culture here that somehow cultivates fringe ideas and resistance to anything mainstream. Berlin is, for lack of a better term, a vibe.

I was expecting nothing less from the crowd at this week’s Agentic Engineering Berlin meetup. It’s Berlin Blockchain Week, so the room was a good balance of the usual crypto crowd, and AI enthusiasts. This turned out to be the perfect combination for the talks we had lined up.

The event was hosted at Full Node (the new one, for those who have font memories of the OG). Big thanks to our friends at Gnosis who graciously offered the venue.

The Talks

Harness the Harness

Juan Cruz Fortunatti shows how to create sensors that let coding agents see what they’re building, using a 3D rendering project and a conversational agent as case studies.

Main Takeaways

• A harness is everything wrapped around the model: tools, memory, context, guardrails.

• The build-check-correct loop only works when there’s a ground truth to optimize against, and front-end UIs and conversations usually don’t have one.

• Agents aren’t dumb, they’re blind, so build a sensor that turns what you see into something they can read.

• His Oracle pattern diffs a ground-truth render against an output and hands the agent a JSON of which polygons are wrong, not the image it can’t interpret.

Juan's X, GitHub, LinkedIn

One Chat Box, Many Agents

Federico Kunze Küllmer breaks down how Moneda built a cheap, context-aware financial support agent for internal use, benchmarked against Intercom’s AI.

Main Takeaways

• Most support agents fail not because they’re built wrong but because they only have FAQs and a knowledge base, never context about your actual problem.

• The architecture is a multi-agent flow: a) input guard, b) an orchestrator that classifies the question and reads user sentiment to set the reply tone, c) support or companion sub-agents, d) an output guard and a tripwire.

• Build for adversaries, since the tripwire exists to stop scam and refund-extraction attempts where an under-context agent hands money back it shouldn’t.

• Cost dropped from $1 per resolved conversation to 5-15 cents through model tiering with “good-enough” models like Kimi, prompt caching, and moving FAQs into a database with embeddings.

Federico’s X, GitHub, LinkedIn

AI Agents & Crypto: What Could Go Wrong

Jean-Loïc Mugnier walks through real incidents of AI agents losing funds on-chain and lays out a framework for securing agentic systems that touch crypto.

Main Takeaways

• The risk isn’t just the crypto or the model, it’s the whole stack: infrastructure, on-chain data, the protocols, and any bug becomes a direct financial loss.

• Lobster Wilde’s trading agent misread digits in an interface and sent 52 million tokens (~$500k). The deeper problem is you can’t explain why an agent decided what it did.

• aixbt lost 55 ETH (~$100k) not to a model error but to a vulnerable control panel that let an attacker submit a fund-approval request.

• Transaction simulations aren’t safe by default. Jean-Loïc’s research with the Ethereum Foundation found a safe-looking preview that drained a wallet because contracts and liquidity can change between signing and execution.

• Treat everything as malicious (zero trust), and rather than filtering what the agent reads, constrain what it can do with policies and on-chain enforcement contracts that can revert a transaction before it settles.

Jean-Loïc’s X, LinkedIn

We want to bring the Agentic Engineering meetup to other cites in Europe, and beyond. If you’re interested in stewarding a local community, please reach out!

Something interesting has been quietly happening in Lisbon.

A dense community of AI builders has been settling here. Many of them are in the same group chats sharing links, best practices, and progress on their projects. But here's the thing. Most of them have never met in person.

This week, we changed that.

We brought 50 AI builders together for the first Agentic Engineering Lisbon meetup. It exceeded my expectations, and I was blown away by how many smart, talented, and creative people we have in this city.

What struck me is how starved this community has been for a real-world version of itself. I saw people huddled around laptops showing off their agentic workflows, founders sharing what they're building, people swapping notes on the same roadblocks, and real friendships being made. That was the whole bet with this event.

The Goal of Agentic Engineering Meetups?

I have very fond memories of the meetups from the early days of the web, mobile and crypto (now going back 15 to 20 years). They were simple, often in bars, and attracted a eclectic crowd of entrepreneurs, geeks and generally curious people.

People would bring their laptops, show others what they were working on, and leave inspired to build something. And often, making lifelong friends in the process.

This is what Agentic Engineering wants to bring back.

Lisbon Meetup #1 Talks

What The Hell Is An Agent?

Misha Kolesnik describes seven pillars that make up an agent stacked together, with orchestration as the line that separates an "agent" from a "function call."

Main Takeaways

• An agent isn't one thing. It's seven: API, sessions, tools, tool-execution, persistent resources, harness, orchestration.

• Tool-execution, as in letting the agent build and run its own tools at runtime, is what made Claude Code resonate with so many developers.

• Orchestration is the line between "agent" and "function call."

• Each pillar evolves independently. That's the strength.

Misha’s X, GitHub, LinkedIn

Me, My Slop, and I

Alex Lajarre shares two real stories about agent-generated slop in open source, and the layered back-pressure system he built to defend against it.

Main Takeaways

• The agent never says no. You have to. Especially as a maintainer.

• Two failure modes: PRs your agent filed without you noticing, and complexity an agent made cheap to ship.

• Defense is layered: your judgment inside, repo memory and drift guards in the middle, credential boundaries on the outside.

• Don’t build a universal sandbox. Defend the irreversible operations.

Alex’s X, GitHub, LinkedIn

Agentic Phone Call Demo with ElevenLabs & Twillio

Sebastien Couture gives a live demo of a phone-call skill making a restaurant reservation.

Main Takeaways

• Architecture: main agent → "make phone calls" skill → call subservice (ElevenLabs + Twilio) with a scoped prompt.

• Cost: less than 20¢ per call. Twilio is the expensive piece at ~7¢/min.

• Privacy by partitioning: the call subservice runs in a separate process which sends only the information relevant to the call to ElevenLabs.

• The system self-improves with edge cases surface in transcripts, the main agent updates the call rules.

Sebastien’s X, GitHub, LinkedIn

What's Next for Agentic Engineering?

We're going to keep going.

Next is Agentic Engineering Berlin on June 16. Lisbon edition #2 in the next few months. Longer term, I'd love to see this grow into a proper gathering for the agentic engineering scene across Portugal, and Europe.

If you're building in this space, come to the next one.