AI agents hackathon winners, enhance your debating skills, AI agents in Ops, marketing, customer success and more
interview of builders, behind-the-scenes, AI use case of the week, AI news, meet us IRL
đȘ Hey there, welcome to the second issue of âAnother one bites on Dustâ, where we share whatâs happening in and around Dust and some insider takes for AI builders, tinkerers and enthusiasts.
AI is fundamentally transforming both our personal and work lives, altering the way we act, the way we think, and the way we collaborate. This profound change requires new environments. We're building Dust to serve as the operating system for AI-driven companies. When we're done, work won't be the same.
We hope this email helps at least a bit to understand what to do/how to use AI in your lives, and if thereâs more that we could/should be doing, feel free to leave a comment, we read them all.
Long live the builders: AI agents hackathon winners
đ 25+ teams spent their Friday at Qonto's Paris office, hacking on agent-focused projects in all shapes and sizes. Special focus on the winners. If youâre curious about the finalists as well, hereâs a quick Linkedin post.
Most impactful use case (1st place): Simon Cleriot, Khaled Khebbeb, Jarek Ecke and David Aparicio.
Can you describe what you built?
Jarek: We built an agentic workflow that lets non-tech users fix code bugs automatically end-to-end by just describing what is not working (the rest is handled by the workflow).
Khaled: I'd call it the "headache saver"; an enabler for non-technical staff who have to maintain or interact with some tech.
We took the example of brand managers, PMs, marketing officers who might need to maintain some kind of "brochure" or company website with static contents. But it could be applied on any use case where the definition of "what should be" is simple enough to be detailed in a knowledge base or "KB" (like a Notion documentation) and where the definition "what is going wrong" is tangible.
One might call these descriptions as "functional specs" or "functional rules".
So starting from that, we worked on a multi-layered autonomous bug fixing approach, allowing anyone to interact with a chatbot that has knowledge of the KB and that can process the bug report. The next step would be sending a summary to an agent that has the ability to pull, analyze and write on the codebase to directly propose a fix to the reported bug.On the demo, we showcased a direct fix, with automatic release of the changes to production. But since we can write operations, the right path would be: opening a pull request to propose the fix, and deploying that change to some temporary and disposable environment for testing and QA.
David: Imagine a world where non-technical team members can seamlessly contribute to development projects without waiting for tickets to move into the sprint. Our Dust Agent bridges the gap between the two teams. Let's break down the silos!
Why did you decide to work on this?
Jarek: It's a real problem that is quite easy to solve in a PoC [proof of concept] but is not solved at scale yet.
Khaled: The impact here is time saving and it also enables autonomy. Non-tech staff would no longer require a busy dev to be available, and the dev won't need to delve into the KB to understand the specifics. They would instead just review a PR and save on both headaches and keystrokes. In this scenario, the non-tech person gains enthusiasm and a sweet feeling of accomplishment. Every person that has ever experienced this kind of interaction -i.e., the simple bug that takes
monthsweeks to be fixed- knows the frustration.
David: This comes from an observation we've made in all the companies we work with. Non-tech people can't develop, they don't know our tools (like Cursor, Windsurf or Roo Code) so they have to create a ticket, wait, wait and wait for that ticket to leave the backlog to get into a sprint and try a quick resolution.
Most impactful use case (2nd place): Axel Darmouni.
Can you describe what you built?
I built a product to help sick people get quicker to their nearest pharmacy.
To tell you the truth, when I am sick, I am usually feeling too bad to even budge a muscle: Iâd rather stay comfy in my bed than face the pain and get out of bed. The origin of this idea was to ask myself the question: âhow would I make myself move?â. I am usually a thought-to-action kind of person, so my answer was that âI need a goal. Even if itâs not a complete goal, I need a clear goal and for the process to be as fast as possibleâ. Therefore, I was going for it at first to create this app for myself. Given that it was a hackathon and I had low experience in App Development, I settled for a Web App in React instead. The two pushes I needed were the following:
Medication idea: Taken from the French âANSMâ database of over-the-counter medication
Identification of nearby pharmacies: Use of Google Maps API
The idea that I had in mind was also to test several tools I had under my radar but couldnât find time to test:
Using Structured Outputs with an Enum list for classification, which was done
successfully for diagnosis identification extracting a taxonomy from the ANSM database
Getting a feel of the Gemini models: I used to be a Claude enjoyer, but this vibe test allowed me to validate their strong performances
Using an MCP, more precisely the Maps MCP, which in the end ended being slightly less successful than making a workflow related to the maps API
Using Cursor for Frontend Development, which definitely converted me as a cursor fan: I even took the cursor subscription after this hackathon!
Going alone into this hackathon was both intentional and unintentional at the same time: I first registered with two friends of mine, who told me three days before that they could not make it. I asked myself whether or not I wanted to take a challenge and go for a project idea alone, and so I did! Which ended up being quite the pressure, but quite fun as well.
Most creative use case: Gabriel Landman and Gary Klajer.
Can you describe what you built?
Our project, DebateGab, is a game-like platform designed to enhance debating skills. It enables users to engage in debates and having AI agents providing feedback and insights to refine their argumentation skills.
Why did you decide to work on this?
We found the idea of agents representing not ârolesâ but actual persons amusing, so we started thinking about a creative use case leveraging conversations with real or fictional characters. We are aware it wonât revolutionize the tech industry but was super fun to work on !
People found it fun, and several people asked us to publish it, so we extended the project and plan to make it accessible to a larger audience. But you can already try it out (on laptops only) right now (here). Use the credentials (newsletter, 123456), or create an account and enter an OpenAI API key if you want to have your own account.
We can't wait to hear your feedback!
đż You can watch below the demo of the 6 finalists. Gary and Gabriel start at 1min45 â Axel starts at 28min24 â Simon, Khaled, Jarek and David start at 47min33. Enjoy đż
If youâd like to learn even more about our winners -who they are, how they experience the hackathon itself, what theyâd do differently etcâŠ- you can read the interviews in full on our blog here.
Behind-the-Dust
đ€ Direct news and links about Dust from people inside and outside of the company.
Amelie highlighted how the integration of custom AI agents into Alan's marketing workflow has effectively eliminated the traditional trade-off between speed and quality, allowing Alan's marketing team to meet growing content demands while maintaining their high standards (here).
Thibault gave a snapshot about how MCP-powered Dust agents will complete tasks for you (here).
Pauline went one step deeper to explain how Dust is building the AI operating system (here). And as she wrote: âteams donât use Dust to automate existing processes only but to enable entirely new workflows that weren't possible before.â How exciting!
She also published our latest Product Update (here).
We've made it easier to access your knowledge directly in conversations, added OpenAI and Gemini's latest models, and released a beta version of our Salesforce Connection. We're previewing exciting new tools that expand what your agents can do.
Alban shared how weâve built a unified querying system that breaks down disconnected data silos -think: warehouses filled with business metrics, spreadsheets tracking team operations, and CSVs capturing ad-hoc analyses (here).
Qdrant talked about our own challenge to scale to 5k+ data sources (here).
Spendesk highlighted their partnership with us and how their employees -regardless of their technical expertise- are building AI agents that interact with internal documentation using natural language (here).
Lindsay, Customer Success Manager at Mirakl, showcased how she leverages the Dust platform to create AI agents tailored to each client, which analyze public information and internal data pulled from Zoom meeting notes and other sources (here).
HiTech Podcast recorded a whole show about how to create your own custom AI work companion using us. đ
Use case of the week
We got way more than one use case in this video. This is Maltâs (the European Upwork) one year journey into using AI agents -in 20minutes- by AnaĂŻs, Head of Data Platform. Enjoy đż
(and if you want to dive into even more use cases -in French- scroll down to the end and the video âMeet up IA & Opsâ)
Wild AI-world
In case you missed it, AI 2027 is a must-read. We havenât found better words than Yoshua Bengio to describe this document:
âI highly recommend reading this scenario-type prediction on how AI could transform the world in just a few years. Nobody has a crystal ball, but this type of content can help notice important questions and illustrate the potential impact of emerging risks.â
Another thing for sure: AI literacy is a new must-have, and here to stay! After Shopifyâs CEO memo, Duolingoâs CEO also announced they were now âAI-firstâ, and then, Fiverrâs CEO shared his email sent to his team (and applying equally to the whole freelance community): âAI is coming for your jobs. This is a wake-up call.â đ€Ż How many will follow suit?
Meet us IRL
Weâre stoked to partner with the first-ever {Tech: Europe} AI Hackathon in Paris starting this Saturday â join here â a full weekend dedicated to pushing the boundaries of whatâs possible with artificial intelligence.
Stripe Tour is coming to London next week -on May 28- so if youâre around, come say hi to Clement and Abboud.
Join Alban and Seb on June 2 for an exciting hands-on session where they'll take their Dust agents beyond the platform and integrate them directly into their daily workflow (and yep, that one is happening online!)
And we are back with another Engineering Night in Paris on June 11. You know what to expect by now:
If youâre still hesitating to join us, hereâs a little recap of our latest edition.
Engage Paris is a big event on June 24 entirely focused on Customer Success. Both Clement and AdĂšle will be there; the former with a 15min demo and the latter with a 45min workshop session.
French Touch đ«đ·
We have offices in San Francisco and Paris, but while our founders used to live in the US (as early Stripe and one as an early researcher at OpenAI), both Stan and Gabriel are French. Which means, you can hear them/us in âthe language of MoliĂšreâ (as we say in French!).
s/o to Elisabeth for recording the meet up đ you rock!