What does 'autonomy' mean?
A panel of global AI experts debated the claim that “responsible governance that treats agents as autonomous decision makers will fail,” and 72% agreed.
Read the full study from @mitsmr and @BCG: sloanreview.mit.edu/article/…
3
3
21
2,534
Open source won the web once. What will it take to win AI?
Join host Marcus Rein along with Apache co-founder Brian Behlendorf Kali Villarosa, and Seher Shafiq from the #MozFest team for Owners Not Renters live, Tues, September 29th at 12pm ET.
We’ll be live here. Want to ask a question? Join the chat: streamyard.com/watch/p26zyuK…
2
3
16
2,687
You might be renting horsepower you don’t need to get the average task done. Closed models can cost ~5x more per task, but that premium only pays off with tasks that take 8-12 hours to complete. Whatever closed can do today, open does ~4 months later. stateofopensource.ai/
1
5
73
501,074
Owners Not Renters Livestream with Jad Esber and Raffi Krikorian nitter.net/i/broadcasts/1yKAPwNQY…
2
6
16
4,090
Subscribe to the Owners Not Renters newsletter: newsletter.ownersnotrenters.…
1
463
If you have a question for the Raffi or the Koodos team, join the chat here: streamyard.com/watch/ZUWnfaM…
1
423
The model layer’s commoditizing, this has been proven. What’s less talked about is that power and lock-in have moved up the stack to the harness and accumulated memory. Meanwhile, “open” itself splits into release rungs, license terms, deployment footprints, and provenance that don’t all move together.
Open weights can buy economic optionality, but computing and physical hardware costs are still the user’s responsibility.
Read @politico's summary of our SOSAI report and OSAI stack explainer.
politico.eu/sponsored-conten…
1
1
13
2,747
Mozilla retweeted
this spring, the best closed model scored 60 and the best open models 54. Aayear earlier, the leading open model scored 22. the frontier still leads on the hardest problems — and for the work most builders actually ship, where price, control, and deployability decide, the data does not say “promising.
read the full piece from @tech_eu tech.eu/2026/09/18/open-ai-c…
1
1
1,115
Middle powers can win by building "roads", and focusing less on building "engines." @InstituteGC builds on the findings of our recent report The State of Open Source AI to argue for investing in the AI harness.
Read about it here: institute.global/insights/te…
6
2
22
6,614
Reupping for the morning crew: big announcement from Montreal yesterday with @Mila_Quebec. All in on open source!
blog.mozilla.org/en/mozilla/…
4
2
29
4,635
Owning your AI shouldn't take a whole engineering team and months you can't spare. At ALL IN, Mozilla and @Mila_Quebec announced a new initiative to build an open source AI foundation layer — so any org can run advanced AI locally, on their own terms. 🧵
5
5
31
7,958
The goal: make owning your AI as easy as renting it.
Using open-source models was never the hard part. Turning one into something you can actually run in production — secure, reliable, plugged into everything else — takes an engineering team and months most orgs don't have.
2
2
9
1,076
The Government of @Canada is backing this work, doubling down on its commitment to open source AI.
Read more →
blog.mozilla.org/en/mozilla/…
4
580
Do you actually own your data when AI is the one reading it?
Join Mozilla’s CTO @raffi and @Jad_AE, co-founder and CEO of Koodos (the company behind Shelf) for episode 4 of the Owners not Renters livestream and a conversation about the intersection of personal data protections and AI usage.
Hosted by @Marcus_Rein_
📅 Tuesday, September 22nd at 11am ET
🔗 Add it to your calendar: streamyard.com/watch/ZUWnfaM…
6
14
3,318
If economic incentives for independent creation disappear, original reporting and human-driven knowledge risk being replaced by closed systems.
This is one of many reasons we must build sustainable revenue models for human creators in the age of AI, says Mozilla’s Mark Surman in an interview with @CBC
cbc.ca/news/business/ai-summ…
2
5
33
4,342
As of last month, open models are winning requests.
Google held #1 on OpenRouter for 51 straight weeks. On Aug 3, DeepSeek passed them. DeepSeek weekly requests grew 12.6x since Sep 2025. We predicted this in v1 of the report. Whoever wins tokens wins requests.
Closed providers' share fell from 54% to 44% since May. The model is a commodity now, and the lock-in has moved up a layer. v1.1 of the State of Open Source AI report is live today.
stateofopensource.ai/
5
5
27
4,394
Middle powers can win by building "roads", and focusing less on building "engines." @InstituteGC builds on the findings of our recent report The State of Open Source AI to argue for investing in the AI harness.
Read about it here: institute.global/insights/te…
1
11
4,356
Open models are ~50x cheaper than closed ones.
Mozilla CTO @raffi talks about the economics of open source AI, the "amnesiac" problem with agent memory, and why concentration risk isn't just about who's closed — it's about who's open, too.
Read/watch here: artifactstech.substack.com/p…
3
6
39
7,223