OpenClaw crossed 300,000 GitHub stars in April. On 19 May, Google announced Gemini Spark: a 24/7 agentic assistant that watches your Gmail, edits your Docs, drives your Chrome and runs on Google’s metal whether your laptop is open or not. Six weeks. That is the new clock.

If you read that as a David-and-Goliath story, you’re already a cycle behind. This isn’t an upset. It’s the default lifecycle of any breakout AI-tooling OSS project in 2026, and you need to price it into every dependency decision you make this quarter.

What actually happened

OpenClaw is Peter Steinberger’s MIT-licensed personal AI agent. It runs on your hardware, listens on your messaging apps, executes shell commands, drafts your work. The repo has 374,000 stars and 77,900 forks at the time of writing. Wikipedia’s entry records 247,000 stars by 2 March 2026. That is another ~125,000 stars in roughly ten weeks. Tencent and Z.ai shipped OpenClaw-based services. Chinese developers wired it to DeepSeek and WeChat. Mac Mini stock dried up in some markets because hobbyists wanted always-on hardware.

Then Google announced Spark at I/O 2026 on 19 May. It does roughly the same job. It watches an inbox, drafts status updates, browses the web, runs recurring tasks. You assign work to it by emailing a dedicated Gmail address, “much as they would message a human colleague.” It plugs into Gmail, Docs, Slides and Chrome out of the box, supports MCP for external services, and is gated behind the Google AI Ultra tier at $100 per month. Sundar Pichai called it “a personal AI agent that helps you navigate your digital life.”

Two products. Same shape. Different substrate. OpenClaw runs on metal you bought. Spark runs on metal Google rents you and bundles with a $100 subscription you probably already pay for something else.

This is the pattern, not the exception

You have watched this film before. MongoDB raised, got cloned by AWS DocumentDB, changed its license in October 2018. Elastic got cloned by AWS OpenSearch and switched to SSPL in January 2021. Redis got cloned by ElastiCache and changed terms in March 2024. Valkey was forked under the Linux Foundation within weeks and is now the default in AWS and Google managed caches. HashiCorp went BSL in August 2023; OpenTofu was production-stable and adopted by CNCF inside 18 months.

The OSS-to-hyperscaler-clone interval has been compressing for a decade. MongoDB took six years from launch to AWS clone. Elasticsearch took roughly the same. Redis took longer. OpenClaw took six months.

That compression is the actual news. The window where a breakout open-source AI tool gets to enjoy its own success before a hyperscaler clones it with distribution baked in is now measured in weeks, not years. If you’re building OSS, that’s your runway. If you’re choosing OSS as a load-bearing dependency, that’s your warning.

What this means for your stack

Most teams pick agentic tooling on three criteria: does it work, does the community look healthy, and is it free. None of those criteria survive a hyperscaler clone. They tell you what the tool is today. They tell you nothing about whether the project still has a maintainer in 18 months when Spark, Antigravity and whatever Microsoft ships next have eaten 80% of the addressable market.

The criterion that survives is this: what does this project have that a Google-grade clone cannot replicate in a quarter? Three answers tend to be defensible. A vertical wedge (the tool does one specific thing better than a generalist ever will). An integration moat the platform vendor cannot reach (data sources, on-prem systems, regulated workloads). Or a community structure that owns the roadmap (foundation governance, plural contributor base, not one founder).

OpenClaw has some of the third. Steinberger moved the project to an independent foundation when he joined OpenAI. That helps. It does not solve the distribution problem, which is that 90% of knowledge workers already have a Workspace account and zero of them have a self-hosted Mac Mini configured.

The contrarian punchline

Spark launching is not bad news for OpenClaw. It is validation. Hyperscalers do not clone projects that don’t matter. The relevant question is whether OpenClaw built the things that survive a Google-grade clone in the time it had, or whether it spent the runway accumulating stars.

For the founders and CTOs reading this: stop treating “Google launches a competitor” as a tail risk in your dependency calculus. It is now the base case for any AI-tooling OSS project with momentum. Pick your tools the way you pick your vendors: with explicit lock-in and disruption math, with an exit plan that doesn’t depend on the project staying in its current shape, and with a candid view of whether the moat you’re relying on actually exists.

We wrote about the VC-funded version of this pattern in March. Same arc, different actor. The hyperscaler version is faster — and you don’t get a Series A press release as a warning.

If you’re picking your agent stack right now and want a second opinion before you commit, we’ve seen this pattern enough times to be useful.

Sources: Google introduces Gemini Spark at I/O 2026 (TechCrunch) · Gemini Spark launch (The Next Web) · Google’s Gemini Spark agentic assistant (Engadget) · The next evolution of the Gemini app (Google) · OpenClaw on GitHub · OpenClaw on Wikipedia · OpenClaw becomes GitHub’s fastest-growing project (SOO Group) · AWS and Open Source: It’s Complicated (Redis) · HashiCorp adopts the Business Source License