Thesis
Criteria
Approach
The Founder Experience
Companies
Team
News
Pitch Us
Thesis
Criteria
Approach
Companies
Team
News
Pitch Us
Back to thesis

Agent-Native Infrastructure

The production layer other builders' agents run on.

This is the most consensus-heavy category in our portfolio.

It's the picks-and-shovels layer — identity, payments, orchestration, trust, observability for AI agents. The thesis is straightforward: as the world moves to agentic systems, you need infrastructure built specifically for agents. The loop exists here but operates at the infrastructure level — more agents in the ecosystem means more interactions through your pipes, more edge cases in your orchestration, more signal in your trust framework.

The catch is that this is where every mega-fund is already playing, and the valuations reflect it. At seed, we need conviction that our specific company gets crowned before the market realizes it's obvious — so we're highly selective here.

THE GATES WE APPLY AT IC

The core question: is this a genuinely new primitive, or an existing infrastructure pattern rebranded for agents?

Payment rails for agents aren't different from payment rails for humans unless the company solves a structurally new problem — agent-to-agent settlement, autonomous authorization, machine-speed compliance. If it's Stripe with an "AI" badge, it's not agent-native. Beyond that: Can a seed-stage company win this against well-funded incumbents — has it found a wedge Stripe, Datadog, or Okta can't reach in six months? And do we see winner-take-all dynamics in this specific sub-layer, one that will consolidate rather than splinter into ten equivalent providers?

In the portfolio

Orbifold AI

Orbifold AI

Enoch Zhu's team builds the training-data infrastructure physical AI needs to work. The LLM era trained on free internet data; physical AI has no such luxury — every hour of usable robot training data must be physically captured and pose-recovered to sub-15mm accuracy. Orbifold produces sub-5mm pose-accurate data at 5,000 hours per week, 3x tighter than the standard, with an eval-to-curation flywheel where every delivery improves the next. DeepMind, NVIDIA, and World Labs — the labs whose research defines this data type — all chose to buy from Orbifold rather than build it.

Founders

Enoch Zhu

Location

Palo Alto, CA

Industry

Partnered

2024

Bonfire Lead

Brett Queener

Stage

Early Stage

Founder fit

You may be a fit if builders are already choosing you as the default for a critical layer of the stack, and your usage compounds.