The ecosystem is moving fast. The tooling is getting good. The adoption is real.
The security conversation is mostly quiet.
That gap is going to matter. And I'd rather we start talking about it before it becomes a problem.
What are you seeing?
But here's the thing that stayed with me the whole evening - everyone in that room is building MCPs. Almost no one is talking about what happens when those MCPs go wrong.
Security. Authorization. Blast radius. Audit trails.
The talks were genuinely good:
β Viktor from Superhuman on shifting complexity from probabilistic models into deterministic MCP interfaces
β Christophe from Slite on progressive tool calling and disambiguation layers
A presentation in progress at a tech event in Berlin, showing an audience attentively watching a screen with engineering content.
A speaker presents to an audience, discussing MCP servers, with a slide visible behind highlighting agent management tools.
Spent last night at MCP Connect Berlin.
People are now actively building MCP servers - not just exploring the idea.
The ecosystem is moving from passive consumption toward active engagement.
If you're building in open source, this is the moment. Europe's figuring out how to invest in the commons. π
www.linuxfoundation.org/research/wo...
β Digital sovereignty
β Security improvements
β Breaking tech monopolies
Germany's Sovereign Tech Agency. The EU's Cyber Resilience Act. These aren't just policy changes - they're reshaping how we think about sustaining open source.
And here's the kicker: Few companies invest directly in sustaining the OSS projects they depend on - even though those that do report strong returns.
We're consuming without contributing back.
But something's shifting. Governments across Europe are starting to view OSS as strategic:
The Linux Foundation just released their 2025 Europe Open Source report and one finding caught my attention.
European companies are using open source everywhere. But most don't have formal strategies or Open Source Program Offices.
Applications close March 16. And if your project advances, there's a $250k Sustain Track for follow-on funding.
The internet needs more builders focused on this. If that's you, apply. π
www.mozillafoundation.org/en/what-we-...
β Tools that verify information and reveal algorithmic influence
β Systems that track government decisions and make them actionable
β Privacy-preserving coordination tools for activists
If you're building tech that strengthens democratic processes, this is for you.
Announcement for the Democracy x AI Cohort, offering $50,000 funding to tech projects reshaping democracy. Global applications open.
Mozilla Foundation just opened applications for something I'm genuinely excited about.
The Democracy x AI Cohort. 10 projects. $50k each. 12 months of support.
But here's what makes this different:
They're not funding ideas. They're funding working prototypes that are already being used.
If you're learning something new, resist the urge to delegate to AI. Use it to explain, clarify, and fill gaps. But try and proofread the code yourself.
Your future debugging self will thank you.
Full blog π
www.anthropic.com/research/AI...
That's the very skill you need most to catch AI-generated errors.
That's exactly what suffers when you rely on AI too heavily.
How you use AI matters. High performers asked follow-up questions, sought explanations, and used AI strategically - as a learning tool.
Flowchart outlines a study design with two groups: treatment with AI and control without, detailing task durations and quizzes.
Graphs show effects of AI assistance on coding speed and quiz scores. Comparisons highlight differences in task time and skill development.
Anthropic tested 52 developers learning a new Python library. Half used AI assistance, half coded by hand.
The result? AI users scored 17% lower on post-task quizzes. That's nearly two letter grades. The biggest performance gap wasn't in basic recall. It was in debugging.
Moderated a panel at FOSDEM with browser folks from Mozilla, Google, Opera, and Microsoft.
We talked about where browsers are headed in 2026 and how developers can actually influence what gets built.
Recording is up:
fosdem.org/2026/schedu...
A speaker in a Firefox hoodie presents to an engaged audience in a modern venue, with a projection displaying "What is tracking?".
A speaker in a teal hoodie presents to an engaged audience at a meetup, with a screen displaying a talk title about technology.
Last night's Mozilla meetup in Berlin was incredible! π
Over 60 people, pizza, and amazing talks on Firefox anti-tracking and in-browser AI with WebLLM/WASM. Playing host to such an engaged crowd was a blast. β€οΈ
If you'd like to give a talk on the web topics, reach out! π¦
Lola Odelola's talk asks a simple question: "What do you see?"
It's about centering screen-reader users and treating alt-text as a creative medium, not just a compliance task.
Accessibility doesn't limit artistry. It expands it for everyone.
Worth your time:
heypresents.com/talks/what-...
@oxide.computer published a really thoughtful framework on using LLMs at work.
No bans or mandates. Just five clear values:
Responsibility, Rigor, Empathy, Teamwork, Urgency.
They're specific about what works and what doesn't.
Responsible AI adoption β
rfd.shared.oxide.computer/rfd/0576
Quick heads up: first-come, first-served at the door (we cap at ~50 people). See you there!
π Feb 19 | 18:00 - 21:00
π Firefox's anti-tracking deep dive with Manuel Bucher
π€ In-browser AI with WebLLM + WASM with BarΔ±Ε GΓΌler
RSVP: www.meetup.com/berlin-mozi...
The Mozilla meetup is coming back to our lovely Berlin office! π
Haven't been yet? Come join us THIS THURSDAY for some really cool tech talks on browser privacy and offline LLMs using WASM π¦
@mozilla.org is launching Mozilla Pioneers: a paid program for technologists and builders shaping the next web.
What makes it different?
β Flexible scheduling
β Time-bounded commitment
β No career gambling required
Applications close Feb 16 π
blog.mozilla.org/en/mozilla-...
this is a gentle reminder to go out on a walk. πΆ
no strategy, no learnings here, just walk.
get your body moving.
blood flowing.
kbye.
I built something silly for Valentine's Day π
Your valentine can try saying "No," but the button runs away. Eventually, they have no choice but to say yes. π
Is it practical? No.
Is it fun? Absolutely.
Try it (or send it to your valentine):
valentine.pran.sh
Interop 2026 is here! π
Apple, Google, Igalia, Microsoft, and Mozilla are tackling 20 new focus areas in 2026!
The part I love? MDN documents every single one with clear examples and browser support data.
Read what's in focus:
hacks.mozilla.org/2026/02/lau...
AI with an actual off switch just dropped in @firefox.com 148.
You get to choose:
β Which AI features you want
β Or turn it all off entirely
While other browsers decide for you, we're giving you the controls.
Your browser, your rules.
blog.mozilla.org/en/firefox/...
The web platform doesn't belong to any single vendor. It's the one truly decentralized development environment we have. All browser engines are open source. The opportunity to shape what 5+ billion people use every day is right there.
β Share use cases on the developer signals repo
β Contribute to Web Platform Tests
β This is where it gets exciting. BROWSERS WANT YOUR INPUT. Your use cases and pain points directly shape browser priorities.
How to actually get involved:
β Submit proposals for Interop 2026 (opens Q3)
β Take State of HTML/CSS/JS surveys
β Even when browsers agree on standards, real-world interoperability is messy. Different OS combinations, hardware variations, and edge cases mean the platform is always evolving.
