Arga Labs

VerifiedUpdated 2026-09-23

Arga Labs builds stateful digital twins of SaaS services so AI agents can rehearse multi-app workflows. The company raised a $10 million seed round led by General Catalyst after joining YC's Spring 2026 batch.

Scores run 0 to 100. Potential: how likely the firm is to become a real winner, judged across the six dimension scores and held consistent with them, whether or not anyone has heard of it. Scoop: how unlikely a well-read tech investor is to have heard of it yet. High means it has not reached the outlets that reader follows, whatever the firm's quality or age. How our top potential scores did.

ShareX LinkedIn

UBC fencer Phillip Li and pro-gamer-turned-quant Akira Tong met in calculus at ages 17 and 14 and now sell service twins that test multi-app AI agents after hitting validation limits at Amazon and Stripe.

Snapshot

What they build: Instantly creates digital twin environments for safe testing of AI-generated code before production.
Who buys: Software engineering teams and AI developers.
Comparables: Replit for sandboxes · Cursor for dev tools
Sub-industry: AI agent testing infrastructure

YC-backed with focus on scaling AI dev workflows.

How we found them

Surfaced in issue: Jul 6, 2026
Announced: Jun 18, 2026 (demo day)
HQ: United States (YC cohort)
Discovery track: launched
Discovery context: Showcased digital twin tool for AI agent testing at YC Spring Demo Day.

Originating post ↗

Dimension scores

Analysis

Phillip Li dropped out of UBC after neuroscience research and Canadian junior fencing. Akira Tong finished his UBC degree at 19 following high school and added roles at Stripe, Goldman Sachs, and Google. Both ran into the same testing failures while building agents at Amazon and Stripe.

The product supplies per-service twins that retain mutable state, webhooks, and timing quirks. Teams reset scenarios across multiple services without touching production. Early customers include Slash, Monaco, and Weave. The company reports 100,000 twin instances in the first 16 weeks and lists 30 live twins.

General eval platforms and basic mocks break once agents cross services and accumulate state. Arga targets this gap with a $10 million General Catalyst round and a narrow focus on high-fidelity rehearsal rather than broad observability.

What they do

Arga Labs provides stateful digital twins and on-demand sandboxes of enterprise services (e.g., Salesforce, Stripe, Slack) so engineering teams can safely test and train multi-app AI agents and generated code against realistic APIs, permissions, webhooks, and state before production.

Why now

Surge in capable foundation-model-powered agents (post-2025 improvements) that must take consequential actions across real enterprise SaaS, where traditional unit tests/CI, mocks, or staging environments fail on state accumulation, permissions, cross-system ambiguity, rate limits, and resetability.

Founding story

Phillip Li and Akira Tong met in first-year calculus at UBC (Phillip ~17, Akira 14). Both later encountered the same pain point, testing AI-generated code or agent actions in complex, multi-system environments, at Amazon (Phillip) and Stripe (Akira). Founded in 2025, initially focused on DevOps agents to “kill on-call,” the company pivoted to the validation/simulation layer for AI coding and enterprise agents.

Competitive position

Positioned in the emerging “agent testing/training infrastructure” lane against general eval/observability tools (e.g., LangSmith-style platforms) and traditional mocking/staging solutions (WireMock, custom enterprise test envs). Differentiator is stateful, high-fidelity digital twins of third-party services enabling realistic multi-app workflows and reset/replay at scale.

Defensibility

Primary moat is accumulated expertise and implementation depth in faithfully replicating complex, evolving SaaS behaviors (auth, state, webhooks, edge cases) across dozens of services, hard to replicate quickly without equivalent reverse-engineering effort. Secondary: customer-seeded scenarios/data lock-in and switching costs once workflows integrate twins; potential first-mover/brand in this specific high-fidelity twin primitive.

Network density

High founder concentration from shared UBC/personal network; small early team (4, 6 total) with strong personal ties (e.g., early supporters/angels include UBC-adjacent or founder-network names like Sepand Dyanatkar, Landseer Enga).

Early team signals

Key relationships include YC (P26/Spring 2026 batch, primary partner Pete Koomen); seed led by General Catalyst with Box Group, Emergence, Gradient, SV Angel; earlier pre-seed/YC participation from Comma Capital, Scribble Ventures; advisor Natalie L. (former Lightspeed Partner). Early customer signals include YC-related and small AI/dev-tool teams (e.g., slashapp, Monaco, WorkWeave).

Traction

Confirmed $10M seed (Aug 2026, General Catalyst-led)
Self-reported 100k+ twin instances over 16 weeks pre-announcement
Live product with demo, CLI, SDKs, and catalog of 30+ twins
Early named usage by slashapp, Monaco, WorkWeave, YC entities, and small AI cos (respan.ai, aemon.ai etc. on site)
TechCrunch coverage and YC profile provide independent triangulation
Pricing tiers suggest usage-based SaaS model with paid plans starting ~$1k/mo
GitHub activity and X posts show ongoing shipping (e.g., ArgaBench release)

Opportunities

Rapid enterprise adoption of action-taking agents creates pull for safe testing infra
Potential integration/distribution via major SaaS providers or agent frameworks
Regulatory tailwinds around AI accountability/safety evals
Pricing/usage collapse in foundation models enabling more agent experimentation that needs validation
Expansion of ArgaBench or OSS components for community/developer adoption

Risks / red flags

Unverified technical claims around “100% fidelity in backend functionality” and sub-12-hour cloning for “any SaaS” lack independent benchmarks beyond the company’s own ArgaBench (40-task multi-app eval)
Early traction metrics (100k+ twin instances) are self-reported internal usage rather than confirmed ARR or large enterprise logos
Very small team (4–6) for ambitious per-service emulation work creates execution/key-person risk
Young founders with strong but short/non-traditional pedigrees (dropouts, internships) face scaling challenges in a crowded, hype-driven agent-infra space
Potential regulatory exposure (AI safety/testing requirements) not visibly addressed
Customer base appears early/peer-network heavy (YC-adjacent small teams)

Hidden bets

High-fidelity, stateful service emulation (full API/MCP/CLI/UI twins preserving auth, permissions, webhooks, mutable state, timing/failures/retries)
Natural-language scenario seeding
Browser-based test runs with evidence capture
PR-triggered automation
Multi-twin “scenarios” spanning services
MCP/tool-calling compatibility and CLI wizard for .env swapping

Key relationships

YC (P26/Spring 2026 batch, primary partner Pete Koomen)
General Catalyst (seed lead)
Box Group, Emergence, Gradient, SV Angel
Comma Capital, Scribble Ventures (pre-seed/YC)
Natalie L. (advisor, former Lightspeed Partner)

Founders

Phillip Li, Co-founder & CEO

Education: UBC (Cognitive Systems/Data Science/neuroscience research track; dropped out in final year)

Key projects: Canadian Junior National Team fencer, Neuroscience research at UBC

Prior: Amazon / Intern-level

𝕏 @PhillipLii

Akira Tong, Co-founder & CTO

Education: UBC Bachelor’s (graduated at 19 after skipping high school), Tsinghua University (BCom/CS track)

Key projects: Pro player for Identity V (quit due to finals)

Prior: Stripe / Software Engineer Intern; Goldman Sachs / Quantitative Analyst / ~July–Sep 2024; Google / Research Fellow and Software Product Sprint; Boeing / Software Development Intern; UBC Trading Group / Quantitative Researcher

𝕏 @akkkkiira