Runlayer Says Rippling Cloned It. Guard Your Demo.

Runlayer sells a secure MCP gateway, the plumbing that lets AI models and agents pull in outside data without leaking anything. On July 28, TechCrunch reported that Runlayer filed suit against Rippling, the HR software company, alleging trade secret misappropriation, unfair competition, and breach of contract.

The short version reads like a founder's bad dream. Rippling evaluated Runlayer as a prospective customer. The trial ran for what the complaint calls "nearly a year of intensive engineering collaboration." Runlayer shared its product roadmap and, eventually, its actual source code. Both sides signed a mutual NDA, and Rippling signed a trial agreement with a standard clause forbidding it from copying Runlayer's IP.

Then the deal died over price. Shortly after, a person the complaint describes as a Rippling insider texted Runlayer CEO Andrew Berman that an internal project was underway to build "essentially a clone" of the product. Rippling confirmed to TechCrunch that it's launching its own MCP gateway and denies using Runlayer's IP, calling the claims fabricated.

A court will sort out who's right. The lesson for you doesn't need to wait that long.

Because they can, and because the math often tells them to. A company like Rippling employs hundreds of engineers. When a vendor quotes a six-figure annual contract for a product those engineers believe they can rebuild in two quarters, someone in the room will run the build-versus-buy spreadsheet. That's not villainy. It's procurement.

AI infrastructure makes this worse. The MCP standard itself is open source; Anthropic released it in November 2024. So the protocol layer is free, and what vendors like Runlayer sell is the hardened wrapper around it: access control, audit, agent management. Wrappers feel copyable to a confident engineering team, even when they're not.

And here's the uncomfortable part: a deep trial teaches the prospect exactly how hard the problem really is. Every architecture discussion, every edge case you solve on a call, every roadmap slide shrinks their estimate of the rebuild cost. The better your solutions engineering, the better their internal spec gets.

So assume every large prospect is also a potential competitor. Not out of paranoia. Out of arithmetic.

An NDA is a seatbelt, not a force field. It matters enormously after a crash and does nothing to prevent one.

Runlayer had the paperwork. A mutual NDA, plus a trial agreement with an explicit no-copying, no-derivatives clause. That paperwork is why the lawsuit exists and why a firm like Sullivan & Cromwell took the case. Without it, there'd be no case at all. So yes, sign the agreements, every time, before the first deep call.

But look at what the paperwork didn't do. It didn't stop the year of collaboration from transferring know-how. It can't reach into an engineering team's memory. Trade secret cases are slow, expensive, and hinge on proving what the other side actually used, which is brutally hard when the defendant claims independent development. Rippling's statement, that its product uses "only our proprietary information," previews exactly that defense.

Treat legal protection as your floor. The real protection is deciding what never gets shared in the first place. Contracts compensate you for damage. Disclosure discipline prevents it.

Think of disclosure as a ladder, and make the prospect earn each rung.

Rung one: the demo. Show outcomes, not internals. Anyone with a signed NDA can see this.

Rung two: a scoped pilot in their environment. They see behavior, logs, and results. They don't see how it works inside.

Rung three: architecture discussions. This is where know-how starts leaking, so it should come only after commercial terms are roughly agreed, a letter of intent, a term sheet for the deal, or at minimum a written price range both sides accept.

Rung four: source code. For most startups the honest answer is never. If a customer needs assurance the code exists and is maintainable, offer a code escrow arrangement through a third party, or a paid audit under tight supervision. Handing your repo to a prospect's engineering org, as Runlayer says it did, converts your product into their reference implementation.

Write the ladder down before your next enterprise deal, because in the heat of a big logo chase, you'll be tempted to skip rungs. Everyone is.

A trial without an end date isn't a trial. It's free consulting.

Three rules keep evaluations honest. First, time-box everything. Sixty or ninety days with a named decision date. If they need more time, they can ask, and the ask itself tells you something about their seriousness.

Second, talk price before the deep work starts, not after. Runlayer and Rippling collapsed over price after roughly a year of collaboration. That's the most expensive possible moment to discover misalignment. A one-line email early, "assuming the pilot succeeds, budget is in the range of X," would save both sides months. Some deals die at that email. Good. They died cheap.

Third, meter your engineering hours like the finite resource they are. Every hour your best people spend educating a prospect is an hour not spent shipping. Track it. When trial support for one account crosses some threshold you set in advance, that's a signal to escalate: close, charge for the pilot, or walk.

Paid pilots deserve a special mention. Money changes behavior. A prospect that won't pay 10 or 20 grand for a serious evaluation is telling you what your product is worth to them.

If the worst happens, your memory won't be evidence. Your records will.

Keep a disclosure log for every enterprise deal: what was shared, with whom, on what date, under which agreement. A simple spreadsheet works. Roadmap deck sent March 3 to these four people. Architecture call April 10, these topics, these attendees. Repo access granted to these accounts, revoked on this date.

Watermark sensitive documents per recipient. Use view-only links with expiry instead of attachments where you can. Grant repo or environment access to named individuals, never to a group alias, and revoke it the day the trial ends. None of this is exotic; it's an afternoon of setup.

Why bother? Two reasons. In court, specificity wins, and Runlayer's complaint is compelling precisely because it can describe what was shared and when. Outside court, the discipline itself changes your behavior. When sharing something requires logging it, you pause before sharing it.

One more thing: notice how Runlayer learned about the alleged clone. An insider texted the CEO. Networks talk. Treat your reputation with prospects' engineers as an asset, because someday one of them may be your early warning system.

Not every big-company clone deserves your fear. Internal builds fail constantly, quietly, and expensively. The graveyard of "we'll just build it ourselves" projects is vast, because maintaining infrastructure is a decade-long commitment that internal sponsors rarely survive.

The copy threat is real when your product is a feature of their platform, when your buyer and their engineers report to the same executive, and when the hard part of your product is visible in a demo. It's mostly noise when your value lives in accumulated data, cross-customer learning, integrations, or operational grind that a spec can't capture.

This is worth mapping before you ever enter a big trial, not after. Sit down and write out what your durable advantage actually is: what compounds for you that a well-funded copier can't shortcut. If the answer is "we shipped first," you have a head start, not a moat, and your strategy should be speed. If the answer is a data loop or a distribution channel, protect that specific thing fiercely and share the rest more freely. A structured planning tool like Foundra can help here; its competitive analysis templates force you to name your moat in writing rather than gesture at it, which is exactly the clarity this decision needs.

Runlayer, notably, isn't standing still. It raised $42 million from investors including Khosla Ventures and Felicis and launched with backing from founders of eight unicorns. Lawsuit or not, its best defense is shipping.

Every large prospect is a potential competitor. Price that into how you sell.

Sign the NDA and the no-copy trial agreement every time, but treat them as your floor, not your protection. Contracts pay you back after damage; disclosure discipline prevents it.

Build a disclosure ladder: demo to pilot to architecture to (almost never) source code. Make commercial commitment the price of each rung.

Time-box trials with a decision date. Surface budget alignment before deep engineering work begins, not a year in. Charge for serious pilots when you can.

Keep a disclosure log, per-recipient watermarks, and named-individual access you revoke promptly. Specific records are what make a claim credible.

Know which kind of company you are: head-start companies must outrun copies, moat companies must guard the one thing that compounds. Write it down, then sell accordingly.

Should I refuse to sell to companies that could copy me? No. Big customers are how infrastructure startups grow. Sell to them with a disclosure ladder, a time-boxed trial, and early price alignment. The goal is discipline, not avoidance.

Is sharing source code ever okay in a sales process? Rarely. If a buyer needs continuity assurance, offer third-party code escrow that releases only on defined triggers like your company shutting down. A supervised, paid audit is the outer limit for most deals.

Do trade secret lawsuits work? Sometimes, but they're slow and expensive, and independent development is a strong defense. Treat litigation as a last resort you prepare evidence for, never as a strategy you rely on.

What if a prospect insists on an open-ended free trial? That's a negotiation, not a requirement. Counter with a scoped pilot, a decision date, and ideally a fee. A prospect that refuses all three is telling you they value the education more than the product.

Does an insider tip like Runlayer's change anything legally? It gave notice and likely shaped the complaint, but the case will turn on evidence of what was accessed and used. Which is why your own disclosure log matters.

1. TechCrunch: MCP startup Runlayer accuses Rippling of stealing its product idea (Jul 28, 2026)
2. TechCrunch: Anthropic proposes a way to connect data to AI chatbots (MCP launch, Nov 2024)
3. TechCrunch: MCP AI agent security startup Runlayer launches with 8 unicorns, $11M (Nov 2025)
4. Yahoo Finance: Runlayer raises $30M Series A