Is Instance (the YC S2026 robot-eval startup) profitable?

No. Instance has 2 employees as of July 2026, no disclosed revenue, no named paying customers, and offers only a hosted demo at demo.instancelabs.ai rather than a priced product, per YC's company directory dated 2026-07-25.

Could AI replace Instance's core product?

Instance's product is itself an AI classifier judging robot task success, and it already faces commoditization: UC Berkeley's open-source AutoEval published the same success-detection and scene-reset concept in March 2025, over a year before Instance launched in July 2026.

Who are Instance's biggest competitors?

Robocurve (launched July 23, 2026, open-source eval framework), One Robot (world-model simulation evals), Cortex AI (YC F2025, human-in-the-loop data/eval marketplace), and the free academic AutoEval system from UC Berkeley published March 2025.

Instance is privately held, founded by two MIT graduates named Claire and Lucy, and is backed by Y Combinator as part of its Summer 2026 (S2026) batch; no other investors or funding amounts are publicly disclosed as of July 2026.

Does Instance have any real customers yet?

No named customer logos appear in any public source as of July 26, 2026. The company operates in demo/business-development mode with a hosted demo product and no documented public API, SDK, or disclosed pricing.

Will robot companies outsource evals to Instance instead of building in-house?

Unproven as of July 2026. Robot foundation model labs like Physical Intelligence, Figure, and DeepMind Robotics have the resources to build eval tooling internally, and rival Robocurve's own July 23, 2026 launch framed most labs as currently evaluating in-house.

YC Summer 2026 · United States

Instance

Two-person YC startup building automated success detectors for robot policy evals.

Losing its protection

Assessed 27 July 2026 · Evidence as of 27 July 2026

The structural protection is eroding.

On the recordSince 27 July 2026 · Checkable from H2 2026 · Every call on the record Checkpoint: First paid production rollout with a robotics lab, or a seed round announcement

The call

Instance is racing to prove robotics labs will buy eval infrastructure rather than build it themselves, against free prior art and well-funded in-house builders.

Instance makes software to automatically test whether robots complete tasks correctly, using AI to watch videos and decide if something worked. The problem is that a free, open-source version already exists from UC Berkeley, and the companies most likely to buy Instance (like Tesla and Boston Dynamics) are the same ones best equipped to build this themselves for free, with no reason to pay someone else. The one thing that could save Instance is if these robotics companies decide outsourcing testing is easier than doing it in-house, but that's the opposite of what happened in the AI lab world, where companies kept their testing tools private.

Why

AI pressure: Frontier VLMs (OpenAI, Anthropic, Google) and open-source AutoEval. Frontier VLMs improve at video judgment every quarter, and a free Berkeley tool already does automatic success detection and resets. Could AI do the core job? Yes, likely.

Bigger players: Robot foundation model labs like Physical Intelligence, DeepMind Robotics, Figure, Tesla. These labs have the money, the robots, and the incentive to build eval tooling in-house instead of buying it. Could a giant copy it? Yes, easily.

What protects it. Being first to market with a working demo, with no data lock-in or signed customers yet to back it up.

What would change this call

Instance announces 2-3 paying robotics labs routing production rollouts through its platform. Would show labs prefer buying eval infra over building it in-house.
OpenAI, Anthropic, or Google ships a native task-success verifier endpoint in their VLM API. Would commoditize Instance's core classifier and shrink its window.
A major robot foundation model lab publicly announces its own in-house eval tooling. Would confirm labs are replicating the in-house pattern, not outsourcing.

Key numbers

2 employees listed in YC directory (2026-07-26)
UC Berkeley AutoEval published 16 months earlier, open source (2025-03-31)
No public API or SDK, only a hosted demo (2026-07-19)

Named risks

Core success-detector primitive already exists as free, open-source AutoEval since March 2025. Severity: high
Natural buyers (robot foundation labs) have resources and incentive to build eval infrastructure in-house. Severity: high
Multiple near-identical YC-adjacent competitors launched within the same weeks in July 2026. Severity: medium
Product is a fine-tuned VLM classifier exposed to rapid frontier VLM improvement. Severity: medium
No disclosed revenue, pricing, public API, or named customers as of July 2026. Severity: medium

Rivals, as read in July 2026

Rival	Direction	Evidence
AutoEval (UC Berkeley, open source)	losing to	Free open-source success-detector and scene-reset system published 2025-03-31, predates Instance by 16 months
Robocurve	contesting	Launched open-source eval framework July 23, 2026, ten days after Instance's own launch
One Robot	contesting	YC-backed, offers simulation-first world-model alternative to real-world eval rigs
Cortex AI	contesting	YC F2025, bundles human-in-the-loop rollouts and evals into a data marketplace

Questions people ask

Is Instance (the YC S2026 robot-eval startup) profitable?: No. Instance has 2 employees as of July 2026, no disclosed revenue, no named paying customers, and offers only a hosted demo at demo.instancelabs.ai rather than a priced product, per YC's company directory dated 2026-07-25.
Could AI replace Instance's core product?: Instance's product is itself an AI classifier judging robot task success, and it already faces commoditization: UC Berkeley's open-source AutoEval published the same success-detection and scene-reset concept in March 2025, over a year before Instance launched in July 2026.
Who are Instance's biggest competitors?: Robocurve (launched July 23, 2026, open-source eval framework), One Robot (world-model simulation evals), Cortex AI (YC F2025, human-in-the-loop data/eval marketplace), and the free academic AutoEval system from UC Berkeley published March 2025.
Who owns Instance?: Instance is privately held, founded by two MIT graduates named Claire and Lucy, and is backed by Y Combinator as part of its Summer 2026 (S2026) batch; no other investors or funding amounts are publicly disclosed as of July 2026.
Does Instance have any real customers yet?: No named customer logos appear in any public source as of July 26, 2026. The company operates in demo/business-development mode with a hosted demo product and no documented public API, SDK, or disclosed pricing.
Will robot companies outsource evals to Instance instead of building in-house?: Unproven as of July 2026. Robot foundation model labs like Physical Intelligence, Figure, and DeepMind Robotics have the resources to build eval tooling internally, and rival Robocurve's own July 23, 2026 launch framed most labs as currently evaluating in-house.

Timeline

2026-07-25YC directory lists Instance at 2 employees, HQ San Francisco
2026-07-23Rival Robocurve launches open-source robot eval framework via Launch YC
2026-07-19apis.io lists Instance as API provider with vendor accuracy claims
2026-07-13Instance publicly launches via Launch YC, hosted demo goes live
2026-03-25AutoEval open-source code published on GitHub
2025-03-31UC Berkeley publishes AutoEval, prior art for autonomous robot eval

Revision history

27 July 2026

First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.

In the same band

Cosmic Robotics Pango Atlas Discovery