AgentGlass

Make websites and edge devices reliable for AI agents. Measure whether your business is visible, usable, and actionable with one Agent Readiness Score.

[AgentGlass @AgentGlassInc]

asked what the next five years look like, @michaelh_0g gave three. physical AI. sovereign AI. agent commerce.

every talk after it fit under one of the three, which is how you know a frame is doing real work.

@SidraMiconi (Founder @AgentGlassInc)

filmed + cut by Sidra Miconi

[AgentGlass @AgentGlassInc]

the failure mode isnt a bug. its an agent that breaks into the database, changes the records, deletes the logs, fakes the work, then asks to get paid.

@michaelh_0g on why append-only systems matter: you can back-trace what the agent actually did.

@SidraMiconi (Founder @AgentGlassInc)
filmed + cut: Sidra Miconi
https://t.co/Du89BmnFz0

[AgentGlass @AgentGlassInc]

five moments from the House of AI robotics + demo day 🔥 @0G_labs put on at El Prado:

— an AI-run airport that worries @0G_labs’ own CEO
— LLMs called “crap technology” on stage
— a $500 robot programmed by someone who says he can’t code
— an AI model that called all four World Cup semifinalists
— why code review may be the new bottleneck

featuring:
@michaelh_0g (CEO @0G_labs)
@mark_turrell (founder, unDavos)
@eshanchordia (founder, @impulseai_)
@sachelik (co-founder, Om Labs)

@Weisiyuen (Chief of staff @ 0G_labs)
@vargs_g (Strategic Partnership Lead @ 0G_labs)

@SidraMiconi (Founder @AgentGlassInc)
filmed + cut: @SidraMiconi
https://t.co/Ssln6eioHc

[AgentGlass @AgentGlassInc]

@FReza1984 @0G_labs thanks for sharing your thoughts Fajar.

[AgentGlass @AgentGlassInc]

five moments from the House of AI robotics + demo day 🔥 @0G_labs put on at El Prado:

— an AI-run airport that worries @0G_labs’ own CEO
— LLMs called “crap technology” on stage
— a $500 robot programmed by someone who says he can’t code
— an AI model that called all four World Cup semifinalists
— why code review may be the new bottleneck

featuring:
@michaelh_0g (CEO @0G_labs)
@mark_turrell (founder, unDavos)
@eshanchordia (founder, @impulseai_)
@sachelik (co-founder, Om Labs)

@Weisiyuen (Chief of staff @ 0G_labs)
@vargs_g (Strategic Partnership Lead @ 0G_labs)

@SidraMiconi (Founder @AgentGlassInc)
filmed + cut: @SidraMiconi
https://t.co/Ssln6eioHc

[Sidra Miconi @SidraMiconi]

the robotics and demo day from @0G_labs (CEO @michaelh_0g) at el PRADO. one fireside, two panels, four demos from their Apollo accelerator's first cohort. I cut the whole thing. best moments below

@michaelh_0g (CEO @0G_labs)
@Weisiyuen (Chief of staff @ 0G_labs)
@vargs_g (Strategic Partnership Lead @ 0G_labs)

@SidraMiconi · @AgentGlassInc
filmed + cut: @SidraMiconi
https://t.co/V8r39CWwWh

[AgentGlass @AgentGlassInc]

a company is planning to use AI to run an airport.

@michaelh_0g, CEO of @0G_labs:

“I would be very, very worried and scared.”

his argument: once AI moves into mission-critical infrastructure, verification and recourse stop being optional.

@michaelh_0g (CEO @0G_labs)
@Weisiyuen (Chief of staff @ 0G_labs)
@vargs_g (Strategic Partnership Lead @ 0G_labs)

@SidraMiconi · @AgentGlassInc
filmed + cut: @SidraMiconi

[AgentGlass @AgentGlassInc]

This is an underrated part of agent infrastructure: delegation has a cost model.

An agent being able to call another agent is useful. But if the delegated path consumes substantially more context, credits, or session capacity without improving completion rate, orchestration itself becomes the bottleneck.

“Can it delegate?” and “Should it delegate?” are different questions.

[AgentGlass @AgentGlassInc]

@SidraMiconi @shrez_royal09 @ecfchess

[AgentGlass @AgentGlassInc]

@SidraMiconi @shrez_royal09 @ecfchess

[AgentGlass @AgentGlassInc]

@CodeHappyX @SidraMiconi @shrez_royal09 @ecfchess “Small part” is exactly how long-horizon support tends to look from the outside.

[AgentGlass @AgentGlassInc]

@CodeHappyX @SidraMiconi @shrez_royal09 @ecfchess Sponsorship is easy to underestimate because the result shows up years later wearing someone else’s name.

[AgentGlass @AgentGlassInc]

@SidraMiconi @shrez_royal09 @ecfchess The nice thing about chess is that “small” has never meant unimportant. Position, timing, and what you enable next matter more than size.

[AgentGlass @AgentGlassInc]

@SidraMiconi @demishassabis @shrez_royal09 “The interesting part is the recursion: chess funded the computer that helped start an AI career, AI came back and changed chess, and now the capital created along the way is flowing back into the next generation of players. Closed loops are underrated.”

[AgentGlass @AgentGlassInc]

"Manufacturing machines are really good at doing exactly what they're told. So if
something goes wrong, the machine has no idea."

Mirdain Systems watches the job instead. You upload the file, they parse where every
part should be and what it should look like at any moment, and a camera compares that
to the live stream. When something diverges the machine can act on it — skip the bad
part, cancel the run, whatever saves the material.

Raising a pre-seed, pilots in conversation...
@mirdainsystems @hasnainbaig06 @Jahenderson05
@fdotinc

[AgentGlass @AgentGlassInc]

Off Season II at @fdotinc last Friday: ~100 teams, six weeks of building.
Our founder @SidraMiconi spent Friday night filming what people built during six weeks at @fdotinc.

10 demos from a room of ~100 teams.

The full cuts are coming!
@ai_singhal → @runmirrors
@pedro_nagamine + @bellujrb → @usebido
@autobrik + @alvax64 → @robonet_
@HeapSmasher → @reachpad
@hasnainbaig06 + @Jahenderson05 → @mirdainsystems
@001TMF + @MukherjeeArvaan → Proteus-Bio

[AgentGlass @AgentGlassInc]

Proteus Bio's premise: the future of medicine isn't a blockbuster drug for three million people, it's a molecule for one person.

They're building a platform that learns continuously across long-horizon drug discovery tasks, and they were running a live campaign at the booth — a cancer that's treatable in most people, but where a pregnancy produces a very similar protein, so a normal
therapeutic risks serious off-target effects. Full lab-in-the-loop design.

Asked what he liked most about the program...

"I'm fascinated about biology. I think it's the least understood and most valuable part
of our lives. It governs how long we live, how well we live, how we can live."

@001TMF https://t.co/IyXGmUWNwA

[AgentGlass @AgentGlassInc]

@CodeHappyX @fdotinc @SidraMiconi @ai_singhal @runmirrors @pedro_nagamine Thanks Infinity @CodeHappyX .
@SidraMiconi will be posting the teaser.
Glad you liked the recap!

[AgentGlass @AgentGlassInc]

200+ views within hours, sponsors sharing it, and people encouraging you to take it toward production—that’s a strong signal.

ChartGate started as a hackathon build, but it’s clearly resonating beyond the room. Keep going. 🔥
ChartGate: https://t.co/hHVkRgCQRa .
https://t.co/hHVkRgCQRa

[AgentGlass @AgentGlassInc]

@KrutiVanatwala Exactly. The constraint forced teams to build against reality, not a polished sandbox.

Live data, citations, changing conditions, and a hard deadline made the demos feel much closer to real agentic systems than typical prototypes. 🔥

[AgentGlass @AgentGlassInc]

@SidraMiconi @ycombinator Awesome 👏

[AgentGlass @AgentGlassInc]

Demo:
https://t.co/WiznifQpwd

Built on @Medplum1 (FHIR-native audit/provenance), @DeepgramAI (voice AI), @stedi (eligibility), and @usemoss (sub-10 ms semantic search) — for the @ycombinator × @Medplum1 Agentic Healthcare Hackathon.

[AgentGlass @AgentGlassInc]

Live from @YCombinator Hack 🔥
Giving agents write access is where “mostly right” stops being good enough.

ChartGate adds the missing control layer: models can propose actions, but deterministic policy decides what actually reaches the medical record.

This is what production-grade clinical agents should look like.
-
Built on @Medplum1 (FHIR-native audit/provenance), @DeepgramAI (voice AI), @stedi (eligibility), and @usemoss (sub-10 ms semantic search) — for the @ycombinator × @Medplum1 Agentic Healthcare Hackathon.
https://t.co/4mRn9QSpJl

[Sidra Miconi @SidraMiconi]

At @YCombinator’s Hackathon right now 🔥

Hospitals are racing to give AI agents write access to medical records. But the permissions protecting those records were designed for humans. They ask whether an app can write—not whether this specific change is safe.

So I built ChartGate: a write-policy layer for clinical AI agents.

The agent can only propose a chart change. A deterministic policy engine—no LLM in the decision path—then:

commits safe changes
holds risky ones for clinician co-sign
rejects unauthorized writes

If an unverified caller tries to reroute a patient’s prescriptions, the agent may understand the request perfectly and prepare the update.

Doesn’t matter.

ChartGate evaluates the write, not the words.

LLMs propose. Policy disposes.

Full demo in the comments 👇
-
Built on @Medplum1 (FHIR-native audit/provenance), @DeepgramAI (voice AI), @stedi (eligibility), and @usemoss (sub-10 ms semantic search) — for the @ycombinator × @Medplum1 Agentic Healthcare Hackathon.