Originally published on the BuildWithHermes blog.
On May 27, 2026, Lovo AI filed for Chapter 7 bankruptcy, less than three weeks before scheduled motion-to-dismiss arguments in a class-action lawsuit brought by voice actors who alleged their voices were cloned without consent. Chapter 7 means liquidation. It is over.
In the same 11-day stretch, Vapi logged three separate outages: a database outage on May 20 that lasted 4 hours and 37 minutes, a degraded service event on May 26 (38 minutes), and increased call failure rates on May 28 (54 minutes). Vapi's headline uptime is 99.785 percent for the period. That sounds fine until you do the math: 0.2 percent of a month is about 86 minutes of downtime. Vapi used more than that in a single incident.
If you run an AI voice agency, this is the playbook, whether or not either event hit you directly. The structural risk only requires that you are building a client-facing business on infrastructure you do not control.
Why this matters for AI voice agencies
Lovo's bankruptcy is not primarily an IP story. It is a story about what happens when a key upstream vendor in your stack disappears without warning. Agencies that white-labeled Lovo voices now have client deliverables tied to a defunct company. No support. No API continuity guarantee. No contractual obligation to keep anything running.
That scenario does not require a lawsuit to hit the next vendor. It requires a funding round that falls through, a pivot that kills a product line, or an acquisition where the acquirer shuts down the legacy API. These are not rare events in the AI tooling market. They are the baseline.
The Vapi situation is a different category of risk, but it compounds the same way. Three outages in 11 days means missed calls that do not reschedule themselves, leads that went cold before a callback, and clients asking why their appointment confirmation campaigns had gaps. The issue is not Vapi's competence. The issue is that single-vendor dependency means you inherit 100 percent of that vendor's bad days.
The revenue math on a single outage
An agency running 20 clients at an average of 500 minutes per client per month processes about 10,000 minutes monthly. Vapi's May 20 outage lasted 4 hours 37 minutes, roughly 19 percent of a business day. For an agency charging $2,000 per client per month (about $40,000 MRR at 20 clients), that one incident put roughly $2,400 of revenue at risk. Not catastrophic on its own. But it compounds across three outages in 11 days, and outbound lead response is time-sensitive: those conversions are unrecoverable.
Action steps for agencies this week
Audit your upstream dependencies before your clients do. Write down every vendor between your agency and your client: voice AI provider, telephony, CRM, automation, billing. For each, ask what breaks immediately for a client if it went offline for 4 hours or shut down entirely. If the answer is "the whole thing," you have a single point of failure.
Contact any clients who noticed gaps during the May 20 Vapi window. Proactive acknowledgment reads as operational maturity. Silence after an outage reads as not noticing.
Do not white-label voices from a single TTS vendor without a fallback. Lovo is an extreme case, but the dynamic is not. Build client deliverables on a layer that abstracts the TTS provider, not on a direct API integration you will rebuild when terms change.
Use this week's news as a qualification signal in sales conversations. "You may have seen that Lovo filed for bankruptcy this week and Vapi had three outages in May. Here is how we think about vendor risk." Serious prospects ask follow-ups.
Move client delivery to a platform layer if you are still running raw API integrations. Raw Vapi or Retell gives you low per-minute prices, and it also means you absorb the full blast radius of every upstream reliability event. At 3 to 10 clients the math can work. At 10 to 20, maintaining raw integrations starts competing with actually growing the agency.
What we built at BuildWithHermes
Hermes is an operating platform for AI voice agencies, not an API wrapper on a single provider. The architecture assumes any single upstream provider will have bad days: the platform handles provider-layer routing, so an agency's uptime is not tied to any one vendor's status page. Clients see the agency's brand and white-labeled workspace; if a provider needs swapping, it happens at the platform level. Pricing: Starter $149/mo (300 included minutes), Business $399/mo (1,000), Agency $699/mo (2,000), overage $0.24/min.
The bottom line
Two infrastructure events in one week are not a warning shot. They are the normal operating environment for AI voice agencies in 2026. Vendors raise money, pivot, get acquired, go dark, and have bad reliability months. The agencies that scale past 10 clients treat that as a design constraint, not an edge case.
You cannot control whether Vapi has a database outage on a Tuesday. You can control whether your agency's revenue is the same before and after that outage.
Full version with FAQ: buildwithhermes.com/blog
Top comments (0)