The EU AI Act’s high-risk classification is more nuanced than most summaries make it look.
Article 6(3) offers a narrow exception — but if the system profiles individuals, it’s high-risk regardless.
Which of the three is hardest in practice?
For anyone running AI compliance at a mid-market company:
You probably don’t have an agent inventory.
You probably can’t tell a regulator how each agent was classified.
You probably don’t know if you’ll be ready for August 2.
If that’s where you are — you’re not behind. Most are.
Nobody quite says this about AI compliance:
Providers are talking. Regulators are talking. Deployers — the companies answering for what their agents do in production — are silent.
Not because they don’t care. Article 26 gives them distinct obligations, and no one’s built for them
I’ve rebuilt Layr twice in three months.
V1 was an open-source observability SDK. V2, live today, is an EU AI Act compliance platform for mid-market companies deploying AI agents.
Solo, nights and weekends. Zero customers. Looking for the first three.
https://t.co/dGR3cowUpN
Wrote the honest story of building
layr-sdk — the pivots, the decisions,
the week one numbers.
Why I think AI agents need an
observability standard and what
I'm doing about it.
https://t.co/NGKmPUswKT
OpenTelemetry didn't replace your
observability platform.
It became the standard for how telemetry data is emitted.
Grafana, Datadog, — they all consume it.
That's exactly what Layr is, for AI agents.
Instrument once.
Send anywhere.
Own your data.
pip install layr-sdk
AI agents don’t fail loudly
they fail silently
everything looks fine
outputs look reasonable
but the decision was wrong
and most tools don’t explain why
That’s why we’re building Layr.
Every layer of the modern stack has
an observability standard.
Infrastructure — OpenTelemetry
Applications — OpenTelemetry
Databases — OpenTelemetry
AI agents — nothing.
That's what Layr is building.
pip install layr-sdk
shipped a small SDK to trace AI agent behaviour last week
~200 downloads already
biggest takeaway so far:
people don’t realise how blind they are until something breaks
Agents will only be as mature as the
infrastructure around them.
Models: ✅
Frameworks: ✅
Observability: ❌
That's why we built Layr.
What's stopping your team from putting
agents into production right now?
@ApplyWiseAi Hey, so the child spans under the parent action, so you get the full hierarchy without the noise.
Just the essentials per span. Tried to make as clean as possible by default.
What stack are you on?
Just shipped layr-sdk v0.1.0 — open source observability for AI agents.
Three lines of code. Works with Grafana, Datadog, and any OpenTelemetry backend.
pip install layr-sdk
https://t.co/QGBuqvmAg6
Building this in public. Feedback welcome✌️
@andreadominic Love this — infra via natural language
is exactly the kind of prod use case
Layr was built for.
What does your observability look like
when the agent does something unexpected?
DM me 👋
Honest question.
Are you actually running AI agents
in production right now?
Not planning to?
Not piloting?
Actually in production, taking real actions?
Reply and tell me what you're building.
Genuinely curious about where you’re at on the journey.
Day 2 of Layr.
185 downloads. Zero marketing.
Just three posts and a GitHub repo.
If you're one of them — what are
you building? Reply and let me know.
The Show HN is coming Tuesday 🤌
pip install layr-sdk
layr-sdk hit 84 downloads in its
first 24 hours with zero marketing.
Building in public.
If you installed it — reply and tell
me what you're building. I read everything 🤓
https://t.co/QGBuqvmAg6
AI agents are the new microservices.
You wouldn't run microservices without
observability.
But most teams deploying agents have
zero visibility into what they're
actually doing.
That's what Layr fixes.
pip install layr-sdk
https://t.co/dGR3cowUpN