Ten AI-Native Companies, One Closed Room — AI Tech Salon Night

On August 6, 2026, ten AI-native companies each took three minutes in a closed room in Redwood City. No judges, no rankings, no winners — just what you are building, why it matters, and what you have learned, then an open conversation with the room. This is who was there, and why we keep building evenings like it.

The researchers decide what's possible. The founders decide what becomes real. The capital decides what comes next.

Which companies presented at AI Tech Salon Night?

Ten AI-native companies across four sections: BitterClip and Visually in Creative Media; Teyon.ai and ZETIC in AI Infrastructure; Rome and Kidooo AI in Human-Centered AI; and FocusAlpha, Klaris, Zentrik, and Manufex in Enterprise AI.

Also in the room

Two more founders were in the room without a main-stage slot: UltraBeing, an AI doctor for mental and metabolic health, and Casola, a real-time conversational layer for agentic AI.

• UltraBeing (ultrabeing.ai) — Zinnia Hashmi, Co-founder & CPO. The AI doctor for mental and metabolic health.
• Casola — Alexander Pucher. The real-time conversational layer for agentic AI.

Why put researchers, founders, and investors in one room?

Because these three groups almost never get unhurried time together. Researchers publish, founders ship, investors take meetings — and the tracks usually only cross inside a transaction, when everyone is already performing a role. An evening where nobody has to sell anything is where the more useful conversation happens.

That is the whole reason we build these rooms. An ecosystem is not a list of companies; it is how well the people in it actually know each other. When a founder can ask a researcher what is genuinely close and what is still years out, when a researcher can see where their work lands once it leaves the lab, when an investor can meet someone a year before there is anything to decide — those conversations compound. Most of them pay off long after the evening they started in.

The room, by the numbers

106 people: 113 founder profiles across nine industry clusters, 28 researcher profiles across six research directions, and investors deploying from pre-seed through Series B — both VC and CVC.

The range is the point. It is why a three-minute introduction is enough to find the person you needed to meet — someone in the room has almost always already worked on the thing you are stuck on. From the registration export, deduplicated by LinkedIn, the founder profiles (n=113) grouped by primary field:

• AI infrastructure and developer tools — 22
• Enterprise workflows — 16
• Health and life sciences — 13
• Fintech, commerce, and legal — 13
• Media, education, and creative — 12
• Agents and general AI — 11
• Consumer and personal AI — 11
• Physical AI and industry — 8
• Other / not specified — 7

And the researchers in the room (n=28), grouped by primary direction:

• AI systems, infrastructure, and agents — 7
• Safety, alignment, and evaluation — 5
• Multimodal, spatial, and robotics — 5
• Applied AI and domain systems — 5
• Foundation models, NLP, and post-training — 3
• Health and clinical AI — 3

What ten pitches in one night told us

Three things stood out: nobody was building a foundation model, several companies were building for an AI agent rather than a person, and two were becoming the service outright instead of selling software to it.

The clearest pattern was who the customer is. BitterClip structures footage so that AI agents can reason over it. FocusAlpha exists to feed Claude and Codex something better than random web pages. Rome runs software on your behalf. Zentrik hands AI builders grounded context instead of a ticket. In four of the ten pitches, the thing being served was a model, not a human — and none of the four described it as a feature.

The second was where the infrastructure work has moved. Neither infra company is training anything. Teyon.ai keeps a training run alive when it fails at scale; ZETIC gets a model that already exists onto a hundred real handsets. That is the unglamorous half of the stack — and it is where the room's technical questions were sharpest.

The third was a shift in what an AI company even is. Klaris is not selling software to insurance agencies; it is an insurance agency. Manufex is not a procurement tool; it routes real manufacturing jobs and staffs a project manager behind the agents. Both are taking the margin of the service business rather than the seat price of a SaaS product.

And the registration data pointed the same direction. AI infrastructure and developer tools was the largest founder cluster of the night (22 of 113), and AI systems, infrastructure, and agents was the largest research direction (7 of 28). Both sides of the room had independently converged on the same layer — which is exactly why the questions landed instead of sailing past each other.

That is the OpenStages motion. Diagnose the company and the ICP. Curate every registration against that ICP. Then pick the form the outcome needs — a 12-person roundtable when the ICP is researchers who never answer cold email; a 100-person salon when the outcome is a three-sided collision; a GTM panel when founders need operators who have already made the jump; a summit when the outcome is category position.

The next one is already set. AI Infra Signal: From Research to Production takes the layer both sides of this room kept returning to — infrastructure — and gives it a full afternoon: Sep 12, 2026 in San Francisco, 100 application-only seats. If you were in the salon, or wish you had been, this is the one to register for.