Something massive is shifting in the tech landscape, and it’s happening right in our neighborhood. At a recent Malt AI afterwork in Brussels, our community focused on the most exciting question in development: How are builders actually deploying AI today?
Read on for some of the key learnings we gleaned throughout the event:
The engineering shift: how to transition from manual prompting to building autonomous agentic testing loops
The architectural play: why data sovereignty is an active structural defense, not just a compliance checklist
The process reality: how to avoid the trap of over-building free code and focus entirely on user adoption
The freelance advantage: why independent experts are executing these shifts months ahead of traditional corporate structures
By Romain Pattyn (Co-founder @ Codika)
Many developers treat coding agents like super-geniuses that need hand-holding. They write a quick prompt, hit enter, and wait. While heavy enterprise structures spend months debating specifications, tech freelancers are actively rewriting the development pipeline on the ground.
The most successful strategy today is to treat AI like a highly specialized, collaborative dev team rather than a single tool.
The winning methodology relies on deep preparation: sketching out database relationships, logic pathways, and workflows to build an end-to-end plan before writing code.
From there, you can deploy specialized sub-agents in parallel using Git branches. Granting your orchestrator agent access to a CLI (Command Line Interface), database, and system logs allows it to simulate real usage, seamlessly run autonomous tests, and refine its own work beautifully.
💡 Don't use AI to write code line-by-line. Use a centralized orchestrator to manage an entire team of parallel sub-agents, and close the loop with automated CLI testing.
One project ran continuously for eight hours during one night, but because we launched our sub-agents in parallel, it actually represented 52 hours of work delivered by the next morning. That is a massive gain in time."
By Jean-Christophe Cuvelier (Founder @ Wellmade)
Data privacy in Europe became a structural necessity (and not only a legal checkbox). Building modern software requires understanding the legal and operational landscape to protect sensitive data.
True data sovereignty relies on recognizing the overlap between the US Cloud Act and GDPR. If an enterprise hosts its data locally in Europe, but uses a cloud provider partially owned by a US parent company, US courts can still legally compel access to that data without European approval.
The solution isn't to abandon global cloud tools, but to architect for business continuity. The best architectural play is to build containerized workflows and explore open-weight models hosted on local, sovereign European infrastructure. If a major provider changes its pricing or drops a feature, your entire agentic workflow shouldn't vanish overnight.
💡 Hosting data in Europe is meaningless if your infrastructure provider falls under foreign legal jurisdictions. True sovereignty requires containerized, open-weight architectures self-hosted on native European infrastructure.
If your business depends entirely on an agentic loop, you need a sovereign backup plan. You can't afford to let a third-party pricing shift or regulatory block switch off your company overnight."
By Andrea Balducci (Fractional RevOps & AI Builder @ SiSu RevOps)
We’ve all seen those posts claiming that a single engineer replaced a whole team using AI. But where heavy traditional consulting firms sell expensive, multi-month replacement briefs, a freelance builder steps in, shadows the workflow, and builds a working prototype in days.
Andrea Balducci shared an insightful and transparent case study on automating sales workflows. Out of 12+ built projects, only 3 actually survived his team's real-world adoption check. His most successful deployments were brilliant, including an automated meeting prep agent pulling from LinkedIn, CRMs, and the web (scraping specific company career pages to track active hiring trends) to save teams 25 minutes per day.
The core takeaway? Code generation has become a readily accessible commodity. The high-value tech skills of tomorrow center around business judgment, ensuring we build the right tools and thoughtful change management that integrates naturally into a team's existing workflows.
💡 Stop building tools just because you can. AI should augment humans where relationships matter, and replace discrete tasks only where automated execution adds objective value.
Everyone asks if AI will take their job. Wrong question. Ask which tasks you'd happily never do again, then automate those. I'm fractional RevOps for B2B companies, and I got tired of recommending tools so I started building them. AI closed the gap between 'someone should build this' and 'I shipped it last night,' and it's the best teammate a solo freelancer never had to hire."
We wrapped up the evening by bringing three speakers to the stage for a panel discussion focused on governance, cost, and the future of engineering.
Right now, LLM tokens are heavily subsidized by tech giants looking for market share. But this pricing structure will inevitably normalize, and corporate cost models will shift.
As massive, centralized cloud architectures become more expensive to run, the industry is paving the way for Edge AI. Between data center strains, local compliance, and shifting government rules on data handling, the future of development is going to be running optimized, open models directly on local consumer hardware.
The cloud-only era is ending. Not because the cloud loses, but because the default flips. Frontier reasoning stays in the cloud, the routine 80% moves to small, specialized models running close to where the data already lives. Between EU data-sovereignty pressure and the fact that nobody knows if today's token prices are real or subsidized, betting your stack on a single deployment model is the actual risk."
The insights shared by our speakers match exactly what we are seeing across the broader European tech market. Traditional corporate structures are slowed down by heavy legacy debt, but freelancers are on the ground right now, holding the keys to rapid execution and defining tomorrow's corporate standards.
In the final section of our Malt Tech Trends (MTT) report, the data shows that company demand for AI projects is surging, with agent requests multiplying 60x. Freelancers are leading this charge as Malt's AI community has grown 229%, with developers rapidly mastering production-level frameworks like LangGraph (+1,007% growth) and MCP (+2,788% growth).
We are definitely not stopping here, and the rest of the year is going to be brimming with new initiatives! We will be hosting dedicated live events this second semester in Brussels, Madrid, Berlin, Lille, Grenoble, Lyon and London to present the full Malt Tech Trends report data, dive into these code architectures, and network. Keep an eye out for the registration links coming soon!
In the meantime, you can read the full report insights here 👇
