A personal note: On Tuesday, 28 April 2026, I posted on LinkedIn:

“In between courses, I built Understories, a prototype at the intersection of communications and AI. It’s a research pilot. Open-source, research-backed, human-centered.” Here’s an update on the last month.

To everyone who engaged: thank you.

Caroline - 25 August 2026

—————————————

1 . Data and updates

The visits are steadily growing. A few data points about the website content that were not planned.

• Journal: a new section on the navigation bar. The monthly update has so far only been sent by email to the community of users, practitioners, and AI tech people who engaged with the tool or with me. I eventually decided to have it on the website too, inside a new section, Journal.  It will host the Reading List in batches of four entries, the priority being to send it to the subscribers by email first.

2. The tech and its surprises

I’m engineering Understories myself, and joined a 4-week bootcamp to do so. The transition towards a more EU and open-source stack is complicated, and I chose pragmatism for now as the ground is moving very fast: 

• The coding environment: At the bootcamp, we use Cursor (acquired by Elon Musk’s SpaceX, 14 August). I can’t explore two new complex AI coding editors at the same time. I heard of Cline and Void as open-source alternatives for advanced developers, which I am not for now.

• The RAG: Understories is based on retrieval-augmented generation. Agents need to be harnessed. Interestingly, Anthropic has released a blog Patterns and problems in emerging multiagent systems. A take: “Coordination doesn't naturally emerge from stronger intelligence nor alignment at the individual level.” It questions orchestration in hindsight.

• The model: Understories uses Claude as an LLM, and I had in mind ultimately to use French Mistral AI’s models, though when testing, I had reservations. It turns out MistralAI is pivoting from an EU Frontier Lab to a neocloud and compute provider, hosting third-party Chinese models (Sifted, 14 August 2026). The LLM is a strategic question that is open now.

3 . Perspective from the outside

Open-source and AI? A French perspective. I talked AI to one open-source suite builder I’ve known for more than 15 years from La Villette, the Tech Museum of Paris, which organises “Les Samedi du Libre”. Linux, Ubuntu, Debian, a long journey:

• Open-source is captured by big tech. GitHub was acquired by Microsoft*. Where do you see this going? There are proprietary solutions that offer free access to the source code or low-cost paid solutions. What matters is management via Git, data backup, and community-led maintenance. Corporations have also enabled open-source software to fund itself. In 2024, Mozilla brought in over $680.41 million including $482.22 million in royalties from partners such as Chrome / Google. These worlds are not separate. 

• At “La Villette”, the open-source community rejects AI. But the venue also hosts a “Café AI”. These two worlds are physically cohabiting. The Parisian open-source scene is diverse: there are varied approaches, and projects coexist: launching them and keeping them going takes hard work, but also a lot of luck. Luck is what makes encounters, opportunities, and unexpected collaborations possible. 

• Do you use AI in your day-to-day life and for your open-source distribution? For my open-source distribution, I made translation upgrades with AI. I use AI for development too, setting up an accessibility version (for the visually impaired), or writing code in a language I wasn’t familiar with. The AI fixed my submission and made it work. I use ChatGPT, Grok and Gemini. In areas such as voice recognition, AI is less convincing.

Update: Cursor has launched a beta version of Origin on 17 August, inviting developers to host their code there, mirroring GitHub.

—————————————

On Ethics and Environment

“We cannot drink DATA,” read signs held by activists protesting Google’s planned AI data-centre hub in Visakhapatnam, India. The proposed $15 billion project would be built in a water-stressed city, near a wildlife sanctuary home to leopards and pangolins. It raises a question of justice and participation: who gets to shape AI’s development, and who bears its environmental costs? If local people do not benefit, why build it there at all? Reported by Reuters and Firstpost.

—————————————

4 . What’s next

• Reopening the Test Phase: work in progress, very complex and a bit painful when you’re not a developer, but that’s possible… I plan to move forward in September, and have it reviewed by an experienced AI Engineer. I will collect testimonials from the upgraded version rather than the pilot.

• Methodology Paper: The methodology relies on the code architecture and is being updated with the code. It's an interesting exercise to build the theory and the practice at the same time.

• Ethical reflection, open-source: anyone interested in joining the discussion, please reach out.

  ——

The monthly update is shared by email first to users, researchers, AI and comms practitioners I happen to meet while building Understories.

I publish every other Thursday a "newsletter" you can subscribe to: The Reading List. It's made of short explainers, key findings, and data points from selected reports as they go live, plus exclusive content.