A startup just raised $10m to fix the reason 88% of AI agent pilots never reach production — by cloning your actual software first

Arga Labs raised a $10m seed round led by General Catalyst, reported 26 August 2026, to build high-fidelity sandbox 'digital twins' of enterprise software like Salesforce and Workday so AI agents can be trained and tested against realistic conditions before going live — a direct market response to the well-documented pilot-to-production gap in AI agent deployment.

Arga Labs, a Y Combinator-backed startup founded in 2025, raised a $10m seed round led by General Catalyst — with Box Group, Emergence, Gradient and SV Angel also participating — reported by TechCrunch on 26 August 2026. What it builds is narrow and specific: full-scale sandbox environments that clone real enterprise software like Salesforce, Workday and email clients, permission systems and webhooks intact, so an AI agent can be trained and stress-tested against something that behaves like the real system before it ever touches production data.

That’s a direct answer to a problem this site flagged in detail earlier this month: Forrester and Anaconda research puts the AI agent pilot-to-production failure rate at 88%, with the top blocker being evaluation gaps — no reliable way to test whether an agent’s output is actually correct before it ships. Arga’s founders, CEO Phillip Li (previously at Amazon) and CTO Akira Tong (previously at Stripe and Goldman Sachs), are betting that most AI agent failures aren’t a model-quality problem at all — they’re a testing-infrastructure problem, because most teams are evaluating agents against stateless API mocks that don’t reproduce the permission edge cases, rate limits, and workflow quirks of the real software the agent will eventually run inside.

Why a funding round for test infrastructure is a signal, not just a data point

Venture money chasing “how do we safely test an AI agent before it’s live” is a tell about where the market actually is. A year ago, most investment in this space went to the agents themselves — better models, faster completions, broader tool access. Money now flowing into the unglamorous layer underneath — sandboxes, evaluation harnesses, digital twins — suggests the industry has broadly accepted that the model was never the bottleneck. The bottleneck is proving an agent won’t break something real before you let it near something real.

So what

If you’re a founder or CTO weighing whether an AI agent feature — customer-facing or internal — is ready to leave the pilot stage, “how was this actually tested against production-like conditions” is now a fair, specific question to ask any team building it, in-house or agency. It’s also exactly the kind of evaluation and governance layer that separates a working AI product from a demo that never ships. Our AI-assisted development work is built around that gap between pilot and production — get in touch if you’re trying to work out where your own agent project actually stands.

← All research

More research.

GitHub secret scanning cuts false alarms for AWS, Google and Alibaba cloud keys and adds a SpaceX detector

GitHub's 5 November 2026 secret scanning update tightens detectors for AWS, Google Cloud and Alibaba keys, relevant to teams building software with AI coding tools that can paste credentials into code.

Read ›

GitHub's Copilot code review can now be managed by API, with review effort levels and AI credit costs shown

GitHub's 6 November 2026 update lets organisations manage Copilot code review by API and see the review effort and AI credits each AI code review used, a sign AI-assisted development is becoming auditable.

Read ›

GitHub's Copilot impact dashboard now counts chat and VS Code agent use, and drops the 'passive users' label

On 6 November 2026 GitHub added Copilot Chat and the VS Code Agents window to its Copilot impact dashboard, bringing it to nine tracked experiences, and renamed 'Passive users' to 'Chat-first users'.

Read ›

Claude Code 2.1.319: a setting that denies permission prompts so AI agent runs don't stall

Claude Code 2.1.319 (6 November 2026) adds a permissions.denyInsteadOfPrompting setting and moves long MCP tool calls to the background, a sign AI-assisted development is being built for unattended runs.

Read ›

GitHub now reports how often developers act on Copilot code review comments

On 5 November 2026 GitHub added code review metrics, including an addressed rate, to the Copilot usage metrics API. For AI software development buyers, it is the first first-party number for how much AI review feedback is acted on.

Read ›

Dependabot can now keep AI agent skills and prompts up to date

GitHub's Dependabot gained Agent Package Manager (APM) support on 3 November 2026, treating AI agent skills, prompts and instructions as dependencies that need updating like any other software.

Read ›

Start a project

Building with AI? Let's talk.

Tell us what you are building. We'll give you an honest view of the right approach.

Start a projectSee our work