Tools change quickly. Teams need transitions.
A new AI tool also changes handovers, responsibilities and habits. Introducing it well creates room to learn and makes its everyday value visible.
19 / ENTRIES
Journal
A personal mini journal on AI systems, interfaces, open source and work designed to function in everyday life.
The prompt is part of the attack surface.
When an AI system reads external content and uses tools, context becomes a security concern. What matters is which permissions that content can effectively acquire.
Non-deterministic does not mean untestable.
AI responses can vary. Product expectations still need a stable yardstick: clear rules, meaningful test cases and a deliberate release decision.
Trust is an architectural property.
People who entrust a system with something personal trust a data flow, not a promise. For sensitive content, local processing is the foundation, not a bonus.
Machines are reading too.
Documentation has two audiences. This website is my experiment in writing for people and machines from the same source.
What print teaches about AI systems.
In print, quality is secured before the run — with proofs, sign-offs and clear criteria. That is exactly the playbook AI systems need in operation.
Let's hope they don't mess it up.
NVIDIA plans to acquire Hugging Face. It could strengthen open AI—if an open ecosystem does not become a controlled distribution channel.
Between people and systems.
On my first day as a Forward Deployed AI Engineer, this is not a field report but a look at the work I expect to encounter.
Prototypes can vibe. Products need engineering.
AI makes the first working draft easy. Security, clarity and reliable operation determine whether it becomes a product.
Local models are not a fallback.
Not every task needs the largest model. Local LLMs create choice, controllable costs and more democratic access to AI.
When attackers become agents.
AI agents can accelerate both attack and defence. Accountability, guardrails and observability therefore matter more.
From hackathon to beta.
A deadline made Webwoven work. Real players revealed what it would take to turn it into a product.
Good technology needs a good shell.
A reliable core becomes technology people can actually use through good UX, DX and clear design.
AI makes expertise more valuable.
As AI accelerates generation, experience, evaluation and technical judgement become the decisive bottleneck.
The limit lies outside the model.
Power grids, cooling and efficiency help determine how far AI can actually scale.
Open source is infrastructure.
Openness is not a label to me, but the basis for understanding, independence and building together.
Not the model. The workflow.
Models change quickly. Value lives in context, rules and a workflow built to survive the change.
AI has to earn its place.
AI belongs only where it solves a real problem better and its value remains transparent.
From a nightmare to a product.
somnara began with a question on holiday and grew through real feedback into a system people trust with something personal.