Wat zijn de openingstijden?

Maandag 09:00 - 18:00 Dinsdag 09:00 - 18:00 Woensdag 09:00 - 18:00 Donderdag 09:00 - 18:00 Vrijdag 09:00 - 18:00

ML6

Accelerate Intelligence. Together. We partner with leading organizations to achieve the extraordinary by developing exceptional people & cutting-edge tech.

03/09/2026

Voice AI in support isn't a switch between "fully automated" and "not used." It's a design question: what should happen while the customer waits?

In this clip, our AI Project Manager Jens Crokaert walks through how it plays out based on queue time alone.

→ Under 30 seconds, direct connect. No added friction.
→ Between 1 and 5 minutes, the AI collects the account number, issue type, and context first, so the handoff to a human is warm, not cold.
→ Over 5 minutes, the AI does a full intake and schedules a callback, so the customer never has to repeat themselves.

The agent picks up with a pre-filled ticket already in the CRM. Zero wait, zero repetition for the customer, and the agent walks in briefed, free to focus on the part of the call that actually needs their expertise.

🎥 Watch the full recording of the webinar, Voice AI in CX: From Pilot to Production with ElevenLabs: https://hubs.la/Q04wvcys0

02/09/2026

More applications. Fewer strong matches.

AI has made applying nearly free. CV builders and AI chatbots let anyone generate a polished application in minutes, and some candidates now let an AI agent handle the whole process. These applications can look relevant on paper, but often don't hold up against actual experience.

We saw it in our own data: applications at ML6 rose 45% from Q1 to Q2 2026, while candidates reaching the offer stage rose only 12%.

The obvious response — screening with AI — risks starting an arms race. Candidates optimize to get noticed, employers optimize to filter them out, and both sides end up frustrated.

At ML6, we don't use AI to screen incoming applications. Instead, we're testing additional steps that are harder to automate meaningfully.

Read the full article on how we're adapting our hiring process: https://hubs.la/Q04w25T20

01/09/2026

What actually gives an AI security agent the power to act?
Everyone asks which model is best at finding vulnerabilities. Our AI Engineer Sven Oehri argues that this is only part of the question.

In July, an AI agent of OpenAI reportedly breached part of Hugging Face's production environment while working through an OpenAI security benchmark. It wasn't trying to attack anyone. It was hunting for another route to the answer, and it found one, across thousands of autonomous steps.

What made that possible wasn’t the model’s intelligence alone, but its combination with the harness: the layer of memory, tools, permissions, and ex*****on access built around it. An LLM can suggest how to investigate a vulnerability. Inside a harness, it can read the source code, run it, and act on what it finds.

We see this play out in tools like OpenAI’s Codex Security. In the article, we break its adaptive workflow into eight phases:

◾ understanding the repo
◾ building a threat model
◾ mapping attack surfaces
◾ forming hypotheses
◾ tracing code
◾ running checks in a sandbox
◾ validating exploitability
◾ proposing a fix

Together, these steps close the loop between suspicion and proof. While traditional scanners often produce false positives, agentic workflows can investigate and validate findings before raising an alert.

This matters because the skills gap isn't closing on its own. 88% of security professionals say shortages have already caused a real consequence for their organization (ISC2, 2025), and 94% expect AI to be the biggest driver of change in cybersecurity this year (WEF, 2026).

One thing to try this week: stop asking which model powers your AI security tooling and start mapping its harness. What tools can it call? What systems can it touch? Who approves the sensitive actions?

🔗 Full breakdown, including the eight phase Codex Security workflow: https://hubs.la/Q04w25Zj0

How are you scoping access for the AI agents running in your stack?

27/08/2026

📍 ML6 is growing, and yes, also in Munich.

We opened the doors to our Munich office in 2026. Now, we’re ready to bring more brilliant people through them. We’re scaling the ML6 team across the board; from Sales and Project Management to Senior Engineering roles and our AI Platform team.

As Jasper puts it:
''Munich is an important part of ML6’s growth in DACH. The people joining us now have a real opportunity to take ownership and help shape how we grow in the region. An open environment where ideas are welcome and supported.”

Want to build next level AI and help shape ML6’s next chapter in Munich alongside Jasper, Luca and Konstantin?

👉 Explore our open roles and apply here: https://hubs.la/Q04vv1-k0

26/08/2026

UI testing tooling has existed for years. The bottleneck was always the time it takes to write and maintain the tests.

Our Engineer Violetta Nguyen built cua-uat, a CLI tool that puts Computer Use Agents to work on it. A Computer Use Agent looks at a screenshot, decides on an action like a click or a keystroke, and repeats the loop until the task is done, the same way a human tester would work through an app. One pipeline maps user flows in undocumented applications. One checks a plain-language description against a live site and flags failures with screenshot evidence. One turns a natural-language flow into a working Playwright test.

On a real test application, it caught 27 of 28 planted functional bugs. It also generated 14 Playwright tests that made it into an actual pull request.

The biggest lesson: input quality mattered more than the model. A vague flow description costs an agent a working test, since it has to guess, and that guess compounds. Tighter descriptions paid for themselves in faster, more reliable runs.

If you work in QA, where does most of your test-writing time go: the writing, or keeping tests alive as the UI changes?

Read the full article: https://hubs.la/Q04vkF2M0

25/08/2026

𝗣𝗥𝗘𝗦𝗦
Anthropic just became the first AI company to watermark all of its output

Every text Claude generates now carries an invisible marker. On the surface, this is about the EU AI Act, which requires AI content to be detectable. Google built similar tech back in 2024 and OpenAI is still holding off. Anthropic is the first to roll it out everywhere, publicly, right now.

Our own Ryan Ott, machine learning engineer at ML6, spoke to Kristof Van der Stadt (Trends DataNews) about what's really behind the move. AI companies are running low on fresh, human-written text to train on. Training a model on its own generated output only makes it worse over time. A watermark lets a company quietly filter its own text back out of the pile before that happens.

So the AI Act may have set the deadline, but the incentive to actually do it might be coming from somewhere else entirely.

Read the full article [Dutch]: https://hubs.la/Q04v9Nm10

25/08/2026

What happens when AI leaves the lab and enters the wild?

As part of ML6 for Good, we partnered with the AMES Foundation, a non-profit organization dedicated to protecting endangered wildlife and building financially sustainable conservation models across vast protected areas in Africa.

Together, we transformed drone surveillance into an AI-enabled, real-time detection system, built on Google Cloud, capable of identifying potential threats, triggering instant alerts, and supporting anti-poaching teams when every minute counts.
This project goes beyond technology.

It’s about applying machine learning, computer vision, and cloud infrastructure responsibly, in environments where scale, ethics, and real-world impact truly matter.

👉 Read the full case study to explore how AI can actively support wildlife conservation, not just in theory, but on the ground.
https://hubs.la/Q04tP0nK0

21/08/2026

Voice AI can catch a name in seconds. Numbers are harder.

In this clip from our webinar on voice AI in CX, we get into what happens when a caller has to read out a customer number or ID, not just their name. Take the Belgian national registration number: always 11 digits. The system checks the format first. Only if that's correct does it check the number against the database.

Small design choice, real impact: lower latency, and it fails fast instead of failing expensive.

🔍 This is one detail out of a full hour on getting voice AI from pilot to production. If you're weighing where the real complexity hides, worth the watch: https://hubs.la/Q04tK4Y30

20/08/2026

Diversity brings people together. Inclusion ensures every voice is heard.

At ML6, we believe true inclusion is built through psychological safety, continuous feedback, and a strong sense of belonging. From employee engagement surveys and coaching conversations to community-driven initiatives, we work every day to create an environment where people can thrive.

Innovation starts where people feel they belong. Here's our approach.
https://hubs.la/Q04tGzzK0

19/08/2026

Most companies think their AI adoption problem is a usage problem while it isn't. People are using the tools. The real problem is that the knowledge they build up while using them disappears.

An engineer spends three weeks figuring out exactly how to get an AI agent to work well on a specific codebase; which context to feed it, which workflows actually hold up, what not to let it touch. That knowledge is genuinely valuable. And then the chat session ends, the engineer moves to another project, and all of it evaporates. The next person starts from zero. We call this context evaporation, and in our experience it's the actual bottleneck behind uneven AI adoption, not tooling or training.

In our latest blog, Senior Data Engineer Georges Lorré lays out the fix: context as code.

Instead of letting AI knowledge live in Slack threads and chat history, it goes into versioned markdown files that sit in the repo alongside the code itself — an https://hubs.la/Q04tw5z00 that tells any agent what a project is and what it should never touch, reusable https://hubs.la/Q04tw45h0 workflows for recurring tasks, AI review instructions that evolve with the codebase, and the specs and ADRs most teams already have but haven't pointed an agent at yet.

The blog also gets into the harder, more structural question: how do you make this scale across an entire organization rather than living project by project? That's where our internal tool Nimbus comes in - a "company floor" of standards that every new project inherits by default, kept current through a mechanism called reinit, and pushed toward staying fresh (rather than going stale like most documentation) through something we're calling context harvesting.

It's a genuinely practical read for any engineering org that has noticed its AI adoption feels uneven — strong in pockets, invisible everywhere else — and wants to understand why.

Read the full article: https://hubs.la/Q04tw1cH0

Adres

Esplanade Oscar Van De Voorde 1
Ghent
9000

Openingstijden

Maandag	09:00 - 18:00
Dinsdag	09:00 - 18:00
Woensdag	09:00 - 18:00
Donderdag	09:00 - 18:00
Vrijdag	09:00 - 18:00

Website

http://www.ml6.eu/

Meldingen

Wees de eerste die het weet en laat ons u een e-mail sturen wanneer ML6 nieuws en promoties plaatst. Uw e-mailadres wordt niet voor andere doeleinden gebruikt en u kunt zich op elk gewenst moment afmelden.