AI voice agents: what the verified record actually shows (2026)
2026-08-15
The loudest AI voice agent numbers come from vendor case studies. The one voice deployment a regulator examined was found overstated. Here is what survives verification.
Built on verified case files. The argument below leans on evidence The Internet Ninja validated against the public record and published in full, method included.
You are being sold AI voice agents on a single number: a containment rate, a deflection multiple, a headcount equivalent. The number almost always comes from the company selling the agent. Before you wire one into your call flow, it is worth asking which of those numbers anyone outside the vendor has ever checked.
The answer, for the loudest claims on this beat, is nobody. The most-cited AI voice agent outcomes are vendor case studies with one source: the vendor. That is not a reason to dismiss them. It is a reason to know exactly what tier of evidence you are buying on.
Here is the sharpest data point we have. The one voice-AI deployment an independent regulator put on the record, Presto Automation’s drive-thru order taker, was found to need a human agent on essentially every order while the company told investors it “eliminat[es] human order taking” (source).
What AI voice agents are
An AI voice agent is a software system that holds a spoken conversation with a caller, interprets intent with a language model, and either completes the request or routes it to a person. The measure that matters for a buyer is the share of calls it finishes without a human, and how honestly that share is reported.
How do AI voice agents work
A voice agent transcribes the caller, passes the text to a language model to decide the next action, then speaks a response or triggers a backend task. Some run fully automated. Many keep a human in the loop, either reviewing before an action or taking over when confidence drops.
That human-in-the-loop distinction is where the marketing and the reality separate. A system that drafts and a system that acts are sold with the same phrase, “AI voice agent,” and priced very differently.
How good are AI voice agents
Honest answer: it depends on who is holding the stopwatch, and most of the published stopwatches belong to vendors.
Sierra, Parloa, Cresta and others publish striking voice and chat containment figures tied to named enterprise clients. TIN reviewed several of these for its case-file registry and could not independently corroborate the numbers against a second source, so they were not granted a verified badge. They may be accurate. They are single-origin, and a buyer should treat them as vendor claims, not measured outcomes.
Against that, one voice deployment has a signed regulatory finding behind it, and it points the other way.
AI voice agents examples: what survives verification
Three deployments in TIN’s verified registry bear directly on voice agents. One is voice. Two are adjacent customer-service agents that establish the human-in-the-loop pattern buyers actually get.
Presto Automation (voice, drive-thru). On 14 January 2025 the SEC entered a settled cease-and-desist order finding Presto’s AI claims materially false (source). The original version of its own voice AI “required human agent intervention, including entering the order, in all instances,” using off-site agents in the Philippines and India, while filings reported automation rates of 95% to 99% (source). Even the advanced 2023 pilot needed a human about 70% of the time. The full case file is the Presto SEC AI-washing order.
Octopus Energy (text, adjacent). Kraken’s “Magic Ink” drafted support email replies with a person reviewing and sending each one. By end-April 2023 CEO Greg Jackson said it handled 34% of customer queries, “the work of 250 people,” and the company said there would be no job cuts, announcing 4,000 new roles in 2024 (source). See the Octopus Energy Kraken case file.
Notion and Decagon (text, adjacent). Per Decagon’s case study, Notion saw a 34% improvement in ticket resolution time, a 2x increase in deflection, and a 3.4% ask-for-human rate (source). The figures are vendor-published, but Notion’s Global Head of Customer Experience is named and quoted, which is why it cleared the bar as a single impeccable source. See the Notion Decagon case file.
What TIN verified
The proof section is the part of this page no other blog on the beat can write, because it rests on files TIN checked against the public record, not on opinion.
The pattern across all three is the same, and it is the opposite of the pitch. The system that had an independent examiner, Presto, was found to depend on humans by design. The systems that worked kept humans in the loop on purpose and did not cut staff. The gap between “eliminates human order taking” and “required human agent intervention in all instances” is the whole story of voice-agent marketing in one order.
AI voice agents vs the claims: a comparison
| Deployment | Modality | Claimed | On the verified record | Evidence tier |
|---|---|---|---|---|
| Presto Automation | Voice | ”Eliminat[es] human order taking,” 95% to 99% automated | Human intervention in all instances (original); ~70% (advanced pilot) | Tier 1, signed SEC order |
| Octopus Energy / Kraken | Text | 34% of queries handled, “work of 250 people” | 34% handled, a human reviews and sends every reply, no job cuts | Tier 2, CEO relayed by independent press |
| Notion / Decagon | Text | 34% faster resolution, 2x deflection | Same, vendor-sourced, client CX head named and quoted | Tier 3, single impeccable source |
Read the tier column before the numbers. A signed regulatory order and a vendor case study are not the same kind of fact, even when both mention a percentage.
How much do AI voice agents cost
Pricing on this beat is quoted per minute, per resolution, or per seat, and vendors rarely publish it. TIN has no verified pricing figure to cite, so this page states none. Treat any per-call cost saving a vendor quotes with the same scrutiny as its containment rate: ask for the denominator.
The bottom line
AI voice agents are real and some deliver. But in 2026 the loudest numbers are vendor-published and single-source, and the only voice deployment an independent regulator examined was found to overstate what the AI did. Buy on the tier of evidence, not the size of the percentage. Ask who measured it, ask whether a human is still in the loop, and ask for the number the vendor did not lead with.
FAQ
What are AI voice agents? AI voice agents are software that holds a spoken conversation, understands intent with a language model, and completes or routes the request. On the verified record, many still hand off to a human on a meaningful share of calls.
How good are AI voice agents? It depends on who is measuring. Vendor case studies report high containment, but they are single-source. The one voice deployment a regulator examined, Presto’s drive-thru, needed a human on nearly every order.
Do AI voice agents replace human agents? Not in the cases with a public record. Octopus Energy kept a human reviewing every reply and hired thousands more staff. Presto’s system was found to depend on off-site humans by design.
Sources
- U.S. Securities and Exchange Commission, “Order Instituting Cease-and-Desist Proceedings: Presto Automation Inc.,” Securities Act Release No. 11352, 2025-01-14. https://www.sec.gov/files/litigation/admin/2025/33-11352.pdf
- Cooley PubCo, “SEC charges ‘AI-washing’,” 2025-01-30. https://cooleypubco.com/2025/01/30/sec-charges-ai-washing/
- Sifted, “Octopus Energy CEO on AI answering customer emails,” 2023-05-09. https://sifted.eu/articles/octopus-energy-ceo-chatgpt-news
- Decagon, “Notion case study,” checked 2026-07-08. https://decagon.ai/case-studies/notion
Questions
What are AI voice agents?
AI voice agents are software systems that hold a spoken conversation with a caller, understand intent, and either complete a task or route it. The verified record shows many still hand off to a human on a large share of calls.
How good are AI voice agents?
It depends entirely on who is measuring. Vendor case studies report high containment rates, but they are single-source. The one voice deployment a regulator examined, Presto's drive-thru, needed a human on nearly every order despite claims it eliminated human order taking.
Do AI voice agents replace human agents?
Not in the cases with a verifiable public record. Octopus Energy kept a human reviewing and sending every AI-drafted reply and announced 4,000 new jobs. Presto's system was found to depend on off-site human agents by design.
Sources
- U.S. Securities and Exchange Commission, Order Instituting Cease-and-Desist Proceedings: Presto Automation Inc. (Securities Act Release No. 11352) , 2025-01-14
- Cooley PubCo, SEC charges 'AI-washing' , 2025-01-30
- Sifted, Octopus Energy CEO on AI answering customer emails , 2023-05-09
- Decagon, Notion case study , 2026-07-08
This is analysis, not a verified outcome. It carries no verification badge and never will. The proof lives in the case files, where every figure is checked against the public record and the method is printed on the page.