Who owns the AI agent you're building on? A 2025–2026 field guide

Six of the tools in our own index changed ownership or raised a historic round between October 2025 and June 2026 — a Salesforce roll-up, the largest venture-backed acquisition on record, a $65B mega-round, and a live case of a rival model vendor absorbing a coding tool’s own usage data to train its own model. None of this is speculative. Every event below is dated, sourced, and — in two cases — has already produced a visible consequence for the product. If you are choosing what to build an integration on, ownership is now a live decision input, not a footnote.

The six events, in order

Date	Tool (our listing)	Event	Terms	Status today
2025-10-20	LangChain	Series B, unicorn valuation	$125M at $1.25B	Independent, VC-backed
2025-12-03	Bun (Claude Code’s runtime)	Acquired by Anthropic	Undisclosed	Anthropic-owned, stays open-source/MIT
2026-04-01	Qualified (Piper)	Acquisition closed	~$1.2B	Salesforce-owned
2026-05-28	Anthropic (parent of Claude Code)	Series H	$65B at $965B valuation	Independent, mega-funded
2026-06-15	Intercom Fin	Definitive agreement signed	~$3.6B	Pending Salesforce (expected close ~fiscal Q4 2027)
2026-06-16	Cursor (Anysphere)	Definitive agreement signed	$60B all-stock	Pending SpaceX — largest venture-backed acquisition on record

Two of these are still pending regulatory close. The rest are done deals with real, observable aftermath — which is the part most “here’s what changed” roundups skip.

Salesforce is rolling up agentic CX, and we can already see how the first one played out

Qualified’s Piper (inbound AI SDR) and Intercom Fin (AI support agent) are not a coincidence — Salesforce has now moved to own both ends of the agentic customer-experience stack it competes in with Agentforce. Qualified is the finished case: Salesforce signed the agreement in December 2025 and closed it April 1, 2026 for an acquisition-date fair value of roughly $1.2 billion. Our own listing on Qualified already documents the result — “now backed by Salesforce’s own balance sheet and roadmap,” but also “buyers should expect the roadmap and pricing to move toward Agentforce bundling rather than standalone evolution.”

Intercom Fin is nine weeks behind Qualified on the same path: Salesforce’s definitive agreement was signed 2026-06-15 for approximately $3.6 billion, expected to close around Salesforce’s fiscal Q4 2027. Intercom’s leadership (CEO Eoghan McCabe, Chief Strategy Officer Des Traynor) is staying on, and Salesforce says Fin will complement Agentforce — but neither company has published what happens to Fin’s pricing, its API, or its support for customers who run it on a non-Salesforce helpdesk once the deal closes. If Qualified is any guide, “complement” is the pre-close language and “bundle” is the post-close reality. A buyer evaluating Fin today is evaluating the vendor Salesforce will very likely make it into, not only the standalone product on the page right now.

The Windsurf precedent: a rival model vendor can cut your model access mid-acquisition

This already happened once, and it is the clearest reason to take “will my model access survive this deal” seriously rather than dismiss it as FUD. In June 2025, as OpenAI moved to acquire Windsurf (then an independent AI coding editor, since folded into Cognition and rebranded Devin Desktop) for roughly $3 billion, Anthropic cut Windsurf’s direct access to Claude 3.x models with under five days’ notice — despite Windsurf’s willingness to pay for full capacity. Anthropic co-founder Jared Kaplan’s public reasoning: “I think it would be odd for us to be selling Claude to OpenAI.” The OpenAI deal then collapsed, Google struck a talent-licensing deal with Windsurf’s leadership, and Cognition acquired the remaining company. Anthropic restored Windsurf’s first-party Claude access only after that ownership picture had changed. Our own Windsurf listing documents the next chapter this saga produced: the editor was rebranded to Devin Desktop in June 2026, and its original Cascade agent — the thing daily users actually depended on — was retired outright on 1 July 2026.

That precedent is directly live again, on a bigger scale, in the Cursor deal. SpaceX’s rival model, Grok, is trained by its own AI division, and Cursor and SpaceXAI have already jointly shipped a model together: Grok 4.5, released 2026-07-08, which Cursor’s own blog says was trained on “trillions of tokens of Cursor data which capture a wide-range of user interactions with codebases and software tools.” That training partnership predates the formal $60B acquisition announcement by nearly two months (the compute deal was announced 2026-04-21) — so by the time SpaceX’s acquisition of Cursor’s parent, Anysphere, was announced 2026-06-16, developer usage data was already flowing into a competing model vendor’s frontier model. Cursor markets itself as model-agnostic — Claude, GPT and Gemini, selectable per request — which is exactly the position Windsurf held before its access got cut. Nothing has been cut yet. The Windsurf case is the reason “yet” is doing real work in that sentence.

Anthropic’s two moves are a different kind of signal — mostly reassuring, with one asterisk

Not every event on this list is a risk. Anthropic’s $65B Series H, announced 2026-05-28 at a $965B post-money valuation (led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, with six more firms co-leading), is a stability signal for anyone building on Claude Code: the company’s own announcement names Claude Code by title as a product it intends to keep investing in, and states run-rate revenue had crossed $47 billion in the weeks before the round. Six months earlier, on 2025-12-03, Anthropic acquired Bun — the JavaScript runtime Claude Code, the Claude Agent SDK and other Anthropic coding tools increasingly run on — explicitly to speed up and stabilize that infrastructure. Anthropic committed publicly that Bun stays open-source and MIT-licensed.

The asterisk: none of that changes Claude Code’s existing single-vendor constraint. It still runs on Claude exclusively, with no way to point it at a competing model the way Cursor, Aider or OpenCode can. A well-funded, stable vendor is a real answer to “will this still exist in a year,” but it is not an answer to “what if I want a different model later” — that trade-off was already true before the Series H, and a bigger balance sheet doesn’t remove it.

LangChain’s unicorn round is the one case where funding reduces ownership risk

LangChain’s $125M Series B, announced 2025-10-20 at a $1.25B post-money valuation (a unicorn, up more than 6x from its 2024 Series A), was led by IVP with existing investors Sequoia and Benchmark plus new backers including CapitalG, Sapphire Ventures, ServiceNow Ventures, Workday Ventures, Cisco Investments, Datadog and Databricks. It reads like the other events on this list — a nine-figure round, a new valuation milestone — but the ownership-risk math is inverted. LangChain and LangGraph are MIT-licensed and self-hostable regardless of who is on the cap table; a change in investors doesn’t change what you can run or fork. The round funds LangSmith, the paid observability layer, and the company’s headcount — it does not create a lever anyone could pull to cut your access to the open-source core the way Anthropic cut Windsurf’s model access. For a framework you self-host, “well-capitalized and independent” is close to the best ownership status available.

A framework for judging the risk before you adopt

Six events, four genuinely different risk profiles. In order of what to check:

1. Is the core interface open-source and self-hostable? If yes (LangChain, Bun, and the coding agents in our own open-vs-locked comparison), an ownership change matters far less — you can fork or keep running what you have even if the company disappears tomorrow.
2. Does the acquirer compete with the model vendor you depend on? This is the live one. SpaceX owns xAI/Grok and Cursor’s own blog already confirms Cursor usage data trained a competing frontier model; the Windsurf/Anthropic/OpenAI sequence shows a rival vendor cutting model access is not hypothetical — it already happened, on a five-day notice.
3. Has the acquirer already completed a similar deal you can observe? Salesforce closing Qualified nine weeks before signing for Intercom Fin gives you a real, dated preview of where Fin’s roadmap and pricing head next — read Qualified’s own listing as the leading indicator, not just Fin’s.
4. Is the new backing about capability or about consolidation? Anthropic’s Series H funds the product you already use (named specifically in its own announcement); Salesforce’s acquisitions fold the product into a different one (Agentforce). Same-sized headline number, opposite implication for the tool’s future shape.

None of these four questions has a universal answer across all six deals — which is exactly the point. “It depends” is not a verdict; “check these four things, in this order, and here’s what each answer implies” is.

FAQ

Is Cursor still safe to build on while the SpaceX deal is pending? Nothing has been cut yet, and Cursor still routes to Claude, GPT and Gemini per request as of this writing. But developer usage data already trained a competing model (Grok 4.5, released 2026-07-08, confirmed on Cursor’s own blog), and the Windsurf precedent shows a rival model vendor cutting first-party access is a real, already-executed move, not a hypothetical — worth watching once the acquisition formally closes, expected Q3 2026.

Will Intercom Fin’s pricing or API change after Salesforce’s acquisition closes? Neither company has published anything about post-close pricing or API changes. The closest available evidence is Qualified — a smaller, earlier Salesforce acquisition in the same space (support/sales AI agents) — whose own listing on this site already notes a shift toward Agentforce bundling after its April 2026 close.

Does Anthropic’s $65B round change anything about using Claude Code? It’s a stability signal, not a product change — Claude Code is named directly in Anthropic’s own funding announcement as a product it’s investing in. It does not add multi-model support; Claude Code still runs on Claude exclusively.

Is LangChain still safe to build a production stack on after its funding round? Yes, and its open-source, MIT-licensed, self-hostable status is exactly why: unlike a proprietary tool whose owner can cut access, LangChain’s core is yours to keep running even if its investor base or ownership changes again.

Methodology & sources

Every fact above is dated and drawn from a primary source — a company’s own blog or press release — or, where we quote developer/analyst reaction, from named journalism outlets, checked live between 2026-07-20 and 2026-07-21. We did not estimate any figure; where a detail (like post-acquisition pricing) has not been published anywhere, we say so rather than guess. This is the same standard we apply to every listing — see how we research AI agents and our methodology.

Primary sources: Anthropic — Series H announcement · Anthropic — Bun acquisition · Cursor — Grok 4.5 announcement · Cursor — SpaceX model-training partnership · LangChain — Series B announcement · Salesforce — Qualified acquisition · Intercom — Salesforce agreement to acquire Fin.

Reporting sources: TechCrunch — SpaceX to acquire Cursor for $60B · TechCrunch — Windsurf says Anthropic is limiting Claude access · Forbes — Anthropic cuts Windsurf’s Claude access before OpenAI acquisition · TechCrunch — LangChain hits $1.25B valuation · SalesforceBen — Salesforce acquires Fin.

To see how each of these tools is scored and compared today, start with the index, best AI agents, or side-by-side comparisons.