200,000 Users Built Automations That Run Themselves

01 /The Concept

Two hundred thousand users arrived in eight months without a paid marketing campaign. They came to Creao AI to automate tasks they could describe but couldn’t code, and the product did something most AI workflow tools do not: it built the automation and then ran it.

Kai Cheng founded Creao in 2024 after a decade building production AI systems for over 250 enterprise clients. CTO Peter Pang was a research scientist on Meta’s Llama 3 team. Clark Gao, running go-to-market, previously built data teams at LinkedIn and Tencent. The company launched in September 2025.

The core architecture is what Cheng calls a closed loop: AI that both creates tools and executes them. A user describes a task conversationally — “pull new signups from our CRM, enrich with LinkedIn data, add a summary row to the tracking sheet each morning.” Creao’s Coding Agent writes the integration, tests it in a sandbox, and saves it as an Agent App: a persistent, schedulable unit of automation with its own memory. The task moves from description to autonomous production run without the user seeing a line of code or clicking through a visual workflow builder. Working patterns are saved for reuse across other tasks.

The $10 million latest round, led by Prosperity7 Ventures — the $3 billion diversified fund of Aramco Ventures — brings total funding to $25 million across three rounds in under a year.

02 /Why It Matters

The standard AI workflow tool in 2026 is a sophisticated version of the problem it claims to solve. Visual workflow builders require users to understand the structure of their automation before they can build it — which presupposes the technical context that non-technical users lack. Most AI assistants can describe or draft but not deploy: the output is a plan the user still has to execute. Creao’s architecture is the distinction that produced 200,000 organic users: the system goes all the way to execution.

The persistent Agent App concept is architecturally different from a static workflow. A Zapier or Make automation is a fixed trigger-action chain — it does exactly what it was wired to do, and when the upstream data format changes, it breaks silently. An Agent App in Creao is stateful: it tracks what previous runs produced, adapts when inputs shift, and saves patterns from successful executions for use in future tasks. That is not a UX improvement over existing tools; it is a different model of what business automation is.

The clearest proof point is internal: Creao runs its own SEO, content production, and marketing operations on its own platform, with a 20-person team operating at a capacity that traditional headcount accounting would attribute to organizations several times larger. A company that trusts its own automation platform with production operations is making a verifiability claim that most B2B SaaS cannot make.

03 /Opportunities

Full autonomy has a failure mode. An Agent App that generates incorrect output and runs on schedule without human review produces compounding errors — wrong data in the CRM, incorrect rows in the tracking sheet, API calls that hit rate limits and fail silently. Creao’s sandboxed testing is the first-pass mitigation; it does not cover production edge cases, particularly involving external APIs that change schema, rate limits that shift, or data sources that return unexpected formats.

The gap that falls out of this is monitoring infrastructure for autonomous AI systems. Not pre-deployment testing — production monitoring: a system that watches running Agent Apps, compares expected outputs against actual, flags statistical anomalies in results, and routes alerts into Slack or PagerDuty when an autonomous process is doing something the user would not endorse. This is a distinct product category from static analysis, integration testing, or traditional APM tools, because the comparison is between autonomous intent and autonomous output rather than between code and specification.

Every platform building toward full automation autonomy — Creao, Make AI, Relay, n8n with AI — will need this layer. Building it as an SDK that integrates with existing agent platforms positions it as infrastructure rather than competition. The specific entry angle: a lightweight production monitoring wrapper for scheduled AI automations, starting with Creao as the integration target and expanding across the category as autonomous agent platforms proliferate.