AI Tinkerers

Give an agent a world to operate inside, and the demo gets more interesting.

David Hague showed an autonomous researcher built around agent memory.

Robb Winkle showed a local multi-GPU inference server.

Voice, memory, local hardware, and interfaces shape the product

The inner loop is becoming the real battleground for coding agents.

Kyle Doherty showed an Attio agent workflow built around a company's own data.

Hana Benko showed forkable sandboxes for parallel bug hypotheses.

Rob Zuber showed CI-grade validation inside the agent's inner loop.

Validation and context belong inside the workflow.

Paris put containerized runtimes, live data, and agent failures in the same room.

Joe Heitzeberg fixed Ashley after it unsubscribed the organizer.

Daniel Phiri showed local agents using real-time analytics.

Thibaut Melen showed workflows compiled into YAML.

The event brought together Docker, ClickHouse, Ippon Technologies, and Tavily.

Failures. Data. Executable workflows.

The important question in Toronto was not whether an agent could act.

It was whether anyone could see what it was doing.

Ying Zhang showed a governance control plane for coding agents.

Soheil kianzad showed a bounded local VLM interaction loop.

Brandon Leung showed product data turning into evals and next actions.

James Henry showed concepts activated during LLM inference.

Different surfaces, same demand:

the agent should leave evidence behind.

The dangerous part of an agent is often the part that looks like a success.

Matteo Saponati stress-tested research agents against broken evals.

Heinrich Krupp showed local semantic memory in SQLite.

Yves Zumbühl showed logs turning into fix PRs for review.

The interesting work is the instrumentation around autonomy:

drift checks, history, and a human review point.

The first AI Tinkerers Seoul room was already asking the hard question:

what survives contact with a real user?

Heejin Jo showed a real-time interview copilot.

Daniel Ahiaba showed tool-calling for trading workflows.

Tashi Tengyal showed a notebook for making prompts more reliable and portable.

The useful part is the system around the model.

The hard part of an agent is what happens after the demo ends.

Nate Aune tested whether coding-agent harnesses improve the software.

Kushagra Kumar showed an auth boundary for long-running agents.

Naoto Shibata showed dashboards turning into reviewed next actions.

Durability, permissions, evals, and review are product surfaces.

The model is only one part of the system.

The @AITinkerers Tokyo Toranomon meetup felt more like a space for builders exchanging what they are actively testing than a traditional, agenda-driven event.

Balazs Opra walked through Building Adaptive AI Workflows, exploring how systems adjust when inputs and conditions change in real environments.

Yuichi Suzuki presented AI Assisted Decision Systems, which led to questions around where human judgment still plays a critical role.

Yusuke Endo shared Automating Knowledge Retrieval, sparking interest in how teams organize and surface internal information more effectively.

Lexa Baldwin discussed User First AI Interfaces, opening up a thoughtful conversation around usability and how trust forms in AI-driven products.

Leonard Lin demoed Scaling Agent Based Systems, prompting discussion about coordination and orchestration across multiple tools.

Ivan Seleznov focused on Reliable Model Deployment, highlighting the work required to keep systems stable in production environments.

Hiro Yokoki presented Observability in AI Pipelines, encouraging discussion about visibility and monitoring across AI systems.

Grisha Szep walked through Collaborative AI Frameworks, which resonated with teams building systems across multiple functions.

Changyu Hu closed with Designing AI for Real World Contexts, grounding the conversation in the practical constraints teams encounter when deploying AI.

It was one of those evenings where conversations continued between talks, with small groups forming around specific problems people were working through.

Thank you to the organizers Ryunosuke Tanaka, Drew Robbins, Stefania Druga, Mike Lanzetta, and sponsor Sakana AI for making the evening possible.

#AITinkerers #TokyoTech #AICommunity #Builders #HandsOnAI

@AITinkerers Chicago February meetup felt more like an open studio where ideas were being tested and shared than a structured event with a fixed agenda.

@nbonamy walked through how he is approaching applied AI in live environments, focusing on what changes when systems meet real world use instead of controlled conditions.
@InsatiableMinds shared a demo centered on how people interact with AI driven workflows, which led to a thoughtful discussion about usability and where trust in these systems actually comes from.

@shoebsmk brought attention to reliability and coordination across tools, showing how structure often matters more than raw model capability.

@shaun_lawrie closed by breaking down the architectural thinking behind maintaining stability while still allowing room for experimentation.

It was one of those evenings where people stayed back just to keep the conversation going.

Thank you to the organizers @Kdeshpande3 , @landon20s , @WilDublin , and sponsor @DubClub_win for making the evening possible.

#AITinkerers #ChicagoTech #AICommunity #Builders #HandsOnAI

On Saturday, February 28

OpenClaw Global Unhackathon runs at the same time in 29 cities across 6 continents.

This isn’t a demo day.

You show up with an idea.
You test it in public.
You break things.
You fix them.
You leave with something real.

Find your city aitinkerers.org

Joe Heitzeberg sat down with Sam Hesson (Meta AI) to talk about what it really takes to move AI from idea to production.

From experimentation to reliability, they break down the practical side of building scalable systems.

If you’re building in AI, this one’s for you.

Watch the full podcast here: youtu.be/scJEpcO62_Y?si…

#AITinkerers #AI #SoftwareEngineering #MetaAI

From Years of Tinkering to a $15M ARR Breakout 🚀

What happens when a "deep tech" company on the brink of spinning down finally hits the perfect product-market fit?

In our latest @AITinkerers "One-Shot," Joe Heitzeberg sat down with Eric Simons, founder and CEO, to hear the incredible story behind Bolt.new.

After seven years of building core infrastructure, the team pivoted to create the world’s first web-based "vibe coding" tool. The result? They went from $700k to over $12M ARR in just a matter of weeks.

If you’re a founder, engineer, or PM trying to navigate the "vibe coding" revolution, this is a must-watch.

Watch the full interview here: youtube.com/watch?v=o1AWLg…

#AITinkerers #VibeCoding #Bolt #StackBlitz #SoftwareEngineering #AI #StartupPivot #ProductManagement

@AITinkerers Ginza: Live Demos from Builders Shipping AI

We had an incredible time at the @AITinkerers Ginza Meetup in Tokyo!  This selective, hands-on gathering brought builders shipping AI in production for live, code-forward demonstrations and high-signal technical conversations. The energy and quality of demos were exceptional. It was inspiring to see so many practical systems and thoughtful technical conversations in one room.

A huge thank you to the organizers - Stefania Druga, Mike Lanzetta, Drew Robbins and Ryunosuke Tanaka for curating such a selective, high-signal meetup that brought together an incredible group of builders. We’d also like to extend our gratitude to our sponsor, @auth0 , for supporting this session and enabling builders to connect, share, and grow together.

We’re excited to carry this momentum forward and continue fostering meaningful collaboration across the AI Tinkerers community.

#AITinkerers #AICommunity #DemoNight #LiveCode #AIBuilders

Tel Aviv Builders: 2025 went out with a bang (and a lot of code).

The @AITinkerers Tel Aviv End-of-Year Demo Party was the definition of "High Signal." We gathered nearly 200 vetted practitioners at the AI21 Labs HQ to skip the holiday fluff and dive straight into the technical breakthroughs that defined the year.

As always: Demos > Decks. We closed 2025 with rapid-fire lightning talks that demonstrated why Tel Aviv is one of the world's strongest AI hubs. The pattern is clear: Builders in this room are moving beyond "Hello World" and into the messy, complex reality of production-grade agentic architectures and sovereign AI.

A massive thank you to our host and lead sponsor, @AI21Labs , for providing the perfect space for the community to "pop the hood" on their stacks.

And huge props to our partners @CopilotKit , MCP-UI, @tavilyai , and Deeplearning.ai for backing the builders.

To the Tel Aviv organizing team: Thank you for your relentless curation. You’ve built more than a meetup—you’ve built a room where the signal-to-noise ratio stays at a strictly technical maximum.

2026 is for the ones who build. Code talks. Hype walks.

#AITinkerers #TelAviv #GenAI #AgenticAI #MLOps #WorkingCode #AI21Labs #CopilotKit #Tavily

Seattle Builders: The Holiday Science Fair wrapped up 2025 at Foundations.

We traded the standard meetup format for a high-signal "Science Fair" setup—deep-dive stations, real-time code reviews, and a heavy dose of hot cocoa and cookies.

Selective by design, the room was dense with 100+ practitioners. We saw raw work-in-progress builds on:

Real-time voice agent architecture
Quantitative LLM analysis
CUDA-to-Metal porting for edge inference

A massive thank you to our host, Foundations, and our sponsor @auth0 (by @okta ) for fueling the builders and backing the hands-on innovation that defines this community.

Props to the Seattle organizing team—Joe Heitzeberg and Killian Lucas—for maintaining the signal and curating a space where code speaks louder than hype.

Demos > Decks. 2026 is for the ones who build.

#AITinkerers #Seattle #HolidayScienceFair #WorkingCode #LLM #MLOps #Auth0

One-Shot #20: Andrew Baker on the Open vs. Closed Source Frontier.

In the latest episode of One-Shot, Andrew Baker joins Joe Heitzeberg to dismantle the false dichotomy of the current AI landscape. They dive into the trade-offs of performance, privacy, and the "sovereignty" of building on open-source vs. the raw speed of closed API-driven models.

The discussion cuts through the hype: the "moat" isn't the model you use—it's how you orchestrate it. Closed source is for the sprint, but open source is for the sovereignty.

Watch the full deep dive here: youtube.com/watch?v=PQlsgb…

#AITinkerers #OneShot #OpenSourceAI #MLOps #LLMs #TechnicalLeadership

Dubai Builders: The @AITinkerers Paper Club just concluded a surgical dissection of the Transformer.

We upheld the rule: Curation > crowds. The room was strictly limited to 15 of Dubai’s most active AI Architects and ML Engineers for a deep dive into the 'Attention is All You Need' paper.

This was a session on implementation, not vague theory. We went past the abstract and straight into the code, examining:
1. Self-Attention mechanisms
2. Positional Encoding
3. A practical code walkthrough in Python/PyTorch.

Thank you to Chandrashekhar Ramaprasad for curating a room so focused on technical depth.

The foundation of modern AI is built with code, not speculation.

#AITinkerers #Dubai #Transformer #AttentionIsAllYouNeed #PaperClub #DeepLearning

AI Tinkerers London Hackathon: 11 hours of pure, relentless building!

On December 7th, the curated builders at the Knowledge & Retrieval Hackathon by @Redisinc pushed the boundaries of retrieval and shipped working systems focused on agentic retrieval, vector search, and complex context management. Real-time code, not slides.

Massive thanks to our technical partners for backing the builders: @OpenAI , @Prolific , @join_ef , @elevenlabs , and @vibekanban . Special appreciation to @Redisinc for powering the event with critical caching and vector infrastructure.

And props to organizers Louis Knight-Webb, Luke Harries and the expert panel of judges—your rigor keeps the signal high.

Selective by design. For the ones who build.

Global Builders, The ElevenLabs Worldwide Hackathon is just hours away.

On December 11th, the world's most selective AI builder community unites for one night of intense, code-first creation across 33 cities.

Theme: Conversational Agents. Your mission: Build something that speaks, listens, and acts with real user intent. We are pushing voice-to-voice, multimodal agents, and real-time inference to the limit.

$200K+ in prizes on the line.
Teams must be locked in by the kickoff.

If you have secured your spot, prepare your stack. If you haven't applied, check your local chapter immediately—spaces are limited and require in-person attendance.
Sign up here: lnkd.in/dGq9-UGM
This is the frontier of voice AI. See you in the terminal.

Powered by: @elevenlabs with support from global partners including: Anthropic, Stripe, n8n, Bolt, BLACKBOX.AI, CodeRabbit, Clerk.com, and Miro.

#AITinkerers #ElevenLabsHackathon #ConversationalAgents #VoiceAI #GlobalHackathon

AI Tinkerers Bengaluru #11 is a wrap.

No panels. No pitch decks. Just builders shipping code.

This weekend, the signal-to-noise ratio was high. We saw live demos on DFD automation with LLMs, agentic alert-to-PR debugging, and GraphRAG multi-hop retrieval. Real systems, not theory.

Huge thanks to Cashfree Payments for backing the builders and hosting us at The Bay.

Props to Dhiren Mathur @runtimeHorror for leading the chapter. You help us keep the bar high.

To the founders and engineers who showed up: you’re the reason this is the room that matters.

We build. We ship. See you at the next one.

Want to get involved? Search for the AI Tinkerers chapter in your city and join the community: Welcome aitinkerers.org

#AITinkerers #Bengaluru #GenerativeAI #LLM #Builders